Merge pull request #2513 from secondlife/lua-merge-dev

Merge develop branch into Lua project branch.

20 files changed, 4834 insertions, 216 deletions

diff --git a/indra/newview/app_settings/shaders/class1/deferred/CASF.glsl b/indra/newview/app_settings/shaders/class1/deferred/CASF.glsl
new file mode 100644
index 0000000000..e80c59b39f
--- /dev/null
+++ b/indra/newview/app_settings/shaders/class1/deferred/CASF.glsl
@@ -0,0 +1,2556 @@
+/**
+ * @file CASF.glsl
+ *
+ * $LicenseInfo:firstyear=2024&license=viewerlgpl$
+ * Second Life Viewer Source Code
+ * Copyright (C) 2024, Linden Research, Inc.
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation;
+ * version 2.1 of the License only.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
+ *
+ * Linden Research, Inc., 945 Battery Street, San Francisco, CA  94111  USA
+ * $/LicenseInfo$
+ */
+
+/*[EXTRA_CODE_HERE]*/
+
+#ifndef A_CPU
+#define A_GPU
+#define A_GLSL
+#define CAS_BETTER_DIAGONALS
+#define CAS_SLOW
+
+out vec4 frag_color;
+in vec2 vary_fragcoord;
+
+uniform sampler2D diffuseRect;
+uniform vec2 out_screen_res;
+uniform uvec4 cas_param_0;
+uniform uvec4 cas_param_1;
+
+vec3 srgb_to_linear(vec3 cs);
+vec3 linear_to_srgb(vec3 cl);
+#endif
+
+#ifndef SHADER_PORTABILITY
+//==============================================================================================================================
+//
+//                                               [A] SHADER PORTABILITY 1.20210629
+//
+//==============================================================================================================================
+// FidelityFX Super Resolution Sample
+//
+// Copyright (c) 2021 Advanced Micro Devices, Inc. All rights reserved.
+// Permission is hereby granted, free of charge, to any person obtaining a copy
+// of this software and associated documentation files(the "Software"), to deal
+// in the Software without restriction, including without limitation the rights
+// to use, copy, modify, merge, publish, distribute, sublicense, and / or sell
+// copies of the Software, and to permit persons to whom the Software is
+// furnished to do so, subject to the following conditions :
+// The above copyright notice and this permission notice shall be included in
+// all copies or substantial portions of the Software.
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.IN NO EVENT SHALL THE
+// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+// THE SOFTWARE.
+//------------------------------------------------------------------------------------------------------------------------------
+// MIT LICENSE
+// ===========
+// Copyright (c) 2014 Michal Drobot (for concepts used in "FLOAT APPROXIMATIONS").
+// -----------
+// Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation
+// files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy,
+// modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the
+// Software is furnished to do so, subject to the following conditions:
+// -----------
+// The above copyright notice and this permission notice shall be included in all copies or substantial portions of the
+// Software.
+// -----------
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE
+// WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE AUTHORS OR
+// COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+// ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+//------------------------------------------------------------------------------------------------------------------------------
+#define A_2PI 6.28318530718
+#ifdef A_CPU
+ // Supporting user defined overrides.
+ #ifndef A_RESTRICT
+  #define A_RESTRICT __restrict
+ #endif
+//------------------------------------------------------------------------------------------------------------------------------
+ #ifndef A_STATIC
+  #define A_STATIC static
+ #endif
+//------------------------------------------------------------------------------------------------------------------------------
+ // Same types across CPU and GPU.
+ // Predicate uses 32-bit integer (C friendly bool).
+ typedef uint32_t AP1;
+ typedef float AF1;
+ typedef double AD1;
+ typedef uint8_t AB1;
+ typedef uint16_t AW1;
+ typedef uint32_t AU1;
+ typedef uint64_t AL1;
+ typedef int8_t ASB1;
+ typedef int16_t ASW1;
+ typedef int32_t ASU1;
+ typedef int64_t ASL1;
+//------------------------------------------------------------------------------------------------------------------------------
+ #define AD1_(a) ((AD1)(a))
+ #define AF1_(a) ((AF1)(a))
+ #define AL1_(a) ((AL1)(a))
+ #define AU1_(a) ((AU1)(a))
+//------------------------------------------------------------------------------------------------------------------------------
+ #define ASL1_(a) ((ASL1)(a))
+ #define ASU1_(a) ((ASU1)(a))
+//------------------------------------------------------------------------------------------------------------------------------
+ A_STATIC AU1 AU1_AF1(AF1 a){union{AF1 f;AU1 u;}bits;bits.f=a;return bits.u;}
+//------------------------------------------------------------------------------------------------------------------------------
+ #define A_TRUE 1
+ #define A_FALSE 0
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+//_____________________________________________________________/\_______________________________________________________________
+//==============================================================================================================================
+//
+//                                                       CPU/GPU PORTING
+//
+//------------------------------------------------------------------------------------------------------------------------------
+// Get CPU and GPU to share all setup code, without duplicate code paths.
+// This uses a lower-case prefix for special vector constructs.
+//  - In C restrict pointers are used.
+//  - In the shading language, in/inout/out arguments are used.
+// This depends on the ability to access a vector value in both languages via array syntax (aka color[2]).
+//==============================================================================================================================
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+//_____________________________________________________________/\_______________________________________________________________
+//==============================================================================================================================
+//                                     VECTOR ARGUMENT/RETURN/INITIALIZATION PORTABILITY
+//==============================================================================================================================
+ #define retAD2 AD1 *A_RESTRICT
+ #define retAD3 AD1 *A_RESTRICT
+ #define retAD4 AD1 *A_RESTRICT
+ #define retAF2 AF1 *A_RESTRICT
+ #define retAF3 AF1 *A_RESTRICT
+ #define retAF4 AF1 *A_RESTRICT
+ #define retAL2 AL1 *A_RESTRICT
+ #define retAL3 AL1 *A_RESTRICT
+ #define retAL4 AL1 *A_RESTRICT
+ #define retAU2 AU1 *A_RESTRICT
+ #define retAU3 AU1 *A_RESTRICT
+ #define retAU4 AU1 *A_RESTRICT
+//------------------------------------------------------------------------------------------------------------------------------
+ #define inAD2 AD1 *A_RESTRICT
+ #define inAD3 AD1 *A_RESTRICT
+ #define inAD4 AD1 *A_RESTRICT
+ #define inAF2 AF1 *A_RESTRICT
+ #define inAF3 AF1 *A_RESTRICT
+ #define inAF4 AF1 *A_RESTRICT
+ #define inAL2 AL1 *A_RESTRICT
+ #define inAL3 AL1 *A_RESTRICT
+ #define inAL4 AL1 *A_RESTRICT
+ #define inAU2 AU1 *A_RESTRICT
+ #define inAU3 AU1 *A_RESTRICT
+ #define inAU4 AU1 *A_RESTRICT
+//------------------------------------------------------------------------------------------------------------------------------
+ #define inoutAD2 AD1 *A_RESTRICT
+ #define inoutAD3 AD1 *A_RESTRICT
+ #define inoutAD4 AD1 *A_RESTRICT
+ #define inoutAF2 AF1 *A_RESTRICT
+ #define inoutAF3 AF1 *A_RESTRICT
+ #define inoutAF4 AF1 *A_RESTRICT
+ #define inoutAL2 AL1 *A_RESTRICT
+ #define inoutAL3 AL1 *A_RESTRICT
+ #define inoutAL4 AL1 *A_RESTRICT
+ #define inoutAU2 AU1 *A_RESTRICT
+ #define inoutAU3 AU1 *A_RESTRICT
+ #define inoutAU4 AU1 *A_RESTRICT
+//------------------------------------------------------------------------------------------------------------------------------
+ #define outAD2 AD1 *A_RESTRICT
+ #define outAD3 AD1 *A_RESTRICT
+ #define outAD4 AD1 *A_RESTRICT
+ #define outAF2 AF1 *A_RESTRICT
+ #define outAF3 AF1 *A_RESTRICT
+ #define outAF4 AF1 *A_RESTRICT
+ #define outAL2 AL1 *A_RESTRICT
+ #define outAL3 AL1 *A_RESTRICT
+ #define outAL4 AL1 *A_RESTRICT
+ #define outAU2 AU1 *A_RESTRICT
+ #define outAU3 AU1 *A_RESTRICT
+ #define outAU4 AU1 *A_RESTRICT
+//------------------------------------------------------------------------------------------------------------------------------
+ #define varAD2(x) AD1 x[2]
+ #define varAD3(x) AD1 x[3]
+ #define varAD4(x) AD1 x[4]
+ #define varAF2(x) AF1 x[2]
+ #define varAF3(x) AF1 x[3]
+ #define varAF4(x) AF1 x[4]
+ #define varAL2(x) AL1 x[2]
+ #define varAL3(x) AL1 x[3]
+ #define varAL4(x) AL1 x[4]
+ #define varAU2(x) AU1 x[2]
+ #define varAU3(x) AU1 x[3]
+ #define varAU4(x) AU1 x[4]
+//------------------------------------------------------------------------------------------------------------------------------
+ #define initAD2(x,y) {x,y}
+ #define initAD3(x,y,z) {x,y,z}
+ #define initAD4(x,y,z,w) {x,y,z,w}
+ #define initAF2(x,y) {x,y}
+ #define initAF3(x,y,z) {x,y,z}
+ #define initAF4(x,y,z,w) {x,y,z,w}
+ #define initAL2(x,y) {x,y}
+ #define initAL3(x,y,z) {x,y,z}
+ #define initAL4(x,y,z,w) {x,y,z,w}
+ #define initAU2(x,y) {x,y}
+ #define initAU3(x,y,z) {x,y,z}
+ #define initAU4(x,y,z,w) {x,y,z,w}
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+//_____________________________________________________________/\_______________________________________________________________
+//==============================================================================================================================
+//                                                     SCALAR RETURN OPS
+//------------------------------------------------------------------------------------------------------------------------------
+// TODO
+// ====
+//  - Replace transcendentals with manual versions.
+//==============================================================================================================================
+ #ifdef A_GCC
+  A_STATIC AD1 AAbsD1(AD1 a){return __builtin_fabs(a);}
+  A_STATIC AF1 AAbsF1(AF1 a){return __builtin_fabsf(a);}
+  A_STATIC AU1 AAbsSU1(AU1 a){return AU1_(__builtin_abs(ASU1_(a)));}
+  A_STATIC AL1 AAbsSL1(AL1 a){return AL1_(__builtin_llabs(ASL1_(a)));}
+ #else
+  A_STATIC AD1 AAbsD1(AD1 a){return fabs(a);}
+  A_STATIC AF1 AAbsF1(AF1 a){return fabsf(a);}
+  A_STATIC AU1 AAbsSU1(AU1 a){return AU1_(abs(ASU1_(a)));}
+  A_STATIC AL1 AAbsSL1(AL1 a){return AL1_(labs((long)ASL1_(a)));}
+ #endif
+//------------------------------------------------------------------------------------------------------------------------------
+ #ifdef A_GCC
+  A_STATIC AD1 ACosD1(AD1 a){return __builtin_cos(a);}
+  A_STATIC AF1 ACosF1(AF1 a){return __builtin_cosf(a);}
+ #else
+  A_STATIC AD1 ACosD1(AD1 a){return cos(a);}
+  A_STATIC AF1 ACosF1(AF1 a){return cosf(a);}
+ #endif
+//------------------------------------------------------------------------------------------------------------------------------
+ A_STATIC AD1 ADotD2(inAD2 a,inAD2 b){return a[0]*b[0]+a[1]*b[1];}
+ A_STATIC AD1 ADotD3(inAD3 a,inAD3 b){return a[0]*b[0]+a[1]*b[1]+a[2]*b[2];}
+ A_STATIC AD1 ADotD4(inAD4 a,inAD4 b){return a[0]*b[0]+a[1]*b[1]+a[2]*b[2]+a[3]*b[3];}
+ A_STATIC AF1 ADotF2(inAF2 a,inAF2 b){return a[0]*b[0]+a[1]*b[1];}
+ A_STATIC AF1 ADotF3(inAF3 a,inAF3 b){return a[0]*b[0]+a[1]*b[1]+a[2]*b[2];}
+ A_STATIC AF1 ADotF4(inAF4 a,inAF4 b){return a[0]*b[0]+a[1]*b[1]+a[2]*b[2]+a[3]*b[3];}
+//------------------------------------------------------------------------------------------------------------------------------
+ #ifdef A_GCC
+  A_STATIC AD1 AExp2D1(AD1 a){return __builtin_exp2(a);}
+  A_STATIC AF1 AExp2F1(AF1 a){return __builtin_exp2f(a);}
+ #else
+  A_STATIC AD1 AExp2D1(AD1 a){return exp2(a);}
+  A_STATIC AF1 AExp2F1(AF1 a){return exp2f(a);}
+ #endif
+//------------------------------------------------------------------------------------------------------------------------------
+ #ifdef A_GCC
+  A_STATIC AD1 AFloorD1(AD1 a){return __builtin_floor(a);}
+  A_STATIC AF1 AFloorF1(AF1 a){return __builtin_floorf(a);}
+ #else
+  A_STATIC AD1 AFloorD1(AD1 a){return floor(a);}
+  A_STATIC AF1 AFloorF1(AF1 a){return floorf(a);}
+ #endif
+//------------------------------------------------------------------------------------------------------------------------------
+ A_STATIC AD1 ALerpD1(AD1 a,AD1 b,AD1 c){return b*c+(-a*c+a);}
+ A_STATIC AF1 ALerpF1(AF1 a,AF1 b,AF1 c){return b*c+(-a*c+a);}
+//------------------------------------------------------------------------------------------------------------------------------
+ #ifdef A_GCC
+  A_STATIC AD1 ALog2D1(AD1 a){return __builtin_log2(a);}
+  A_STATIC AF1 ALog2F1(AF1 a){return __builtin_log2f(a);}
+ #else
+  A_STATIC AD1 ALog2D1(AD1 a){return log2(a);}
+  A_STATIC AF1 ALog2F1(AF1 a){return log2f(a);}
+ #endif
+//------------------------------------------------------------------------------------------------------------------------------
+ A_STATIC AD1 AMaxD1(AD1 a,AD1 b){return a>b?a:b;}
+ A_STATIC AF1 AMaxF1(AF1 a,AF1 b){return a>b?a:b;}
+ A_STATIC AL1 AMaxL1(AL1 a,AL1 b){return a>b?a:b;}
+ A_STATIC AU1 AMaxU1(AU1 a,AU1 b){return a>b?a:b;}
+//------------------------------------------------------------------------------------------------------------------------------
+ // These follow the convention that A integer types don't have signage, until they are operated on.
+ A_STATIC AL1 AMaxSL1(AL1 a,AL1 b){return (ASL1_(a)>ASL1_(b))?a:b;}
+ A_STATIC AU1 AMaxSU1(AU1 a,AU1 b){return (ASU1_(a)>ASU1_(b))?a:b;}
+//------------------------------------------------------------------------------------------------------------------------------
+ A_STATIC AD1 AMinD1(AD1 a,AD1 b){return a<b?a:b;}
+ A_STATIC AF1 AMinF1(AF1 a,AF1 b){return a<b?a:b;}
+ A_STATIC AL1 AMinL1(AL1 a,AL1 b){return a<b?a:b;}
+ A_STATIC AU1 AMinU1(AU1 a,AU1 b){return a<b?a:b;}
+//------------------------------------------------------------------------------------------------------------------------------
+ A_STATIC AL1 AMinSL1(AL1 a,AL1 b){return (ASL1_(a)<ASL1_(b))?a:b;}
+ A_STATIC AU1 AMinSU1(AU1 a,AU1 b){return (ASU1_(a)<ASU1_(b))?a:b;}
+//------------------------------------------------------------------------------------------------------------------------------
+ A_STATIC AD1 ARcpD1(AD1 a){return 1.0/a;}
+ A_STATIC AF1 ARcpF1(AF1 a){return 1.0f/a;}
+//------------------------------------------------------------------------------------------------------------------------------
+ A_STATIC AL1 AShrSL1(AL1 a,AL1 b){return AL1_(ASL1_(a)>>ASL1_(b));}
+ A_STATIC AU1 AShrSU1(AU1 a,AU1 b){return AU1_(ASU1_(a)>>ASU1_(b));}
+//------------------------------------------------------------------------------------------------------------------------------
+ #ifdef A_GCC
+  A_STATIC AD1 ASinD1(AD1 a){return __builtin_sin(a);}
+  A_STATIC AF1 ASinF1(AF1 a){return __builtin_sinf(a);}
+ #else
+  A_STATIC AD1 ASinD1(AD1 a){return sin(a);}
+  A_STATIC AF1 ASinF1(AF1 a){return sinf(a);}
+ #endif
+//------------------------------------------------------------------------------------------------------------------------------
+ #ifdef A_GCC
+  A_STATIC AD1 ASqrtD1(AD1 a){return __builtin_sqrt(a);}
+  A_STATIC AF1 ASqrtF1(AF1 a){return __builtin_sqrtf(a);}
+ #else
+  A_STATIC AD1 ASqrtD1(AD1 a){return sqrt(a);}
+  A_STATIC AF1 ASqrtF1(AF1 a){return sqrtf(a);}
+ #endif
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+//_____________________________________________________________/\_______________________________________________________________
+//==============================================================================================================================
+//                                               SCALAR RETURN OPS - DEPENDENT
+//==============================================================================================================================
+ A_STATIC AD1 AClampD1(AD1 x,AD1 n,AD1 m){return AMaxD1(n,AMinD1(x,m));}
+ A_STATIC AF1 AClampF1(AF1 x,AF1 n,AF1 m){return AMaxF1(n,AMinF1(x,m));}
+//------------------------------------------------------------------------------------------------------------------------------
+ A_STATIC AD1 AFractD1(AD1 a){return a-AFloorD1(a);}
+ A_STATIC AF1 AFractF1(AF1 a){return a-AFloorF1(a);}
+//------------------------------------------------------------------------------------------------------------------------------
+ A_STATIC AD1 APowD1(AD1 a,AD1 b){return AExp2D1(b*ALog2D1(a));}
+ A_STATIC AF1 APowF1(AF1 a,AF1 b){return AExp2F1(b*ALog2F1(a));}
+//------------------------------------------------------------------------------------------------------------------------------
+ A_STATIC AD1 ARsqD1(AD1 a){return ARcpD1(ASqrtD1(a));}
+ A_STATIC AF1 ARsqF1(AF1 a){return ARcpF1(ASqrtF1(a));}
+//------------------------------------------------------------------------------------------------------------------------------
+ A_STATIC AD1 ASatD1(AD1 a){return AMinD1(1.0,AMaxD1(0.0,a));}
+ A_STATIC AF1 ASatF1(AF1 a){return AMinF1(1.0f,AMaxF1(0.0f,a));}
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+//_____________________________________________________________/\_______________________________________________________________
+//==============================================================================================================================
+//                                                         VECTOR OPS
+//------------------------------------------------------------------------------------------------------------------------------
+// These are added as needed for production or prototyping, so not necessarily a complete set.
+// They follow a convention of taking in a destination and also returning the destination value to increase utility.
+//==============================================================================================================================
+ A_STATIC retAD2 opAAbsD2(outAD2 d,inAD2 a){d[0]=AAbsD1(a[0]);d[1]=AAbsD1(a[1]);return d;}
+ A_STATIC retAD3 opAAbsD3(outAD3 d,inAD3 a){d[0]=AAbsD1(a[0]);d[1]=AAbsD1(a[1]);d[2]=AAbsD1(a[2]);return d;}
+ A_STATIC retAD4 opAAbsD4(outAD4 d,inAD4 a){d[0]=AAbsD1(a[0]);d[1]=AAbsD1(a[1]);d[2]=AAbsD1(a[2]);d[3]=AAbsD1(a[3]);return d;}
+//------------------------------------------------------------------------------------------------------------------------------
+ A_STATIC retAF2 opAAbsF2(outAF2 d,inAF2 a){d[0]=AAbsF1(a[0]);d[1]=AAbsF1(a[1]);return d;}
+ A_STATIC retAF3 opAAbsF3(outAF3 d,inAF3 a){d[0]=AAbsF1(a[0]);d[1]=AAbsF1(a[1]);d[2]=AAbsF1(a[2]);return d;}
+ A_STATIC retAF4 opAAbsF4(outAF4 d,inAF4 a){d[0]=AAbsF1(a[0]);d[1]=AAbsF1(a[1]);d[2]=AAbsF1(a[2]);d[3]=AAbsF1(a[3]);return d;}
+//==============================================================================================================================
+ A_STATIC retAD2 opAAddD2(outAD2 d,inAD2 a,inAD2 b){d[0]=a[0]+b[0];d[1]=a[1]+b[1];return d;}
+ A_STATIC retAD3 opAAddD3(outAD3 d,inAD3 a,inAD3 b){d[0]=a[0]+b[0];d[1]=a[1]+b[1];d[2]=a[2]+b[2];return d;}
+ A_STATIC retAD4 opAAddD4(outAD4 d,inAD4 a,inAD4 b){d[0]=a[0]+b[0];d[1]=a[1]+b[1];d[2]=a[2]+b[2];d[3]=a[3]+b[3];return d;}
+//------------------------------------------------------------------------------------------------------------------------------
+ A_STATIC retAF2 opAAddF2(outAF2 d,inAF2 a,inAF2 b){d[0]=a[0]+b[0];d[1]=a[1]+b[1];return d;}
+ A_STATIC retAF3 opAAddF3(outAF3 d,inAF3 a,inAF3 b){d[0]=a[0]+b[0];d[1]=a[1]+b[1];d[2]=a[2]+b[2];return d;}
+ A_STATIC retAF4 opAAddF4(outAF4 d,inAF4 a,inAF4 b){d[0]=a[0]+b[0];d[1]=a[1]+b[1];d[2]=a[2]+b[2];d[3]=a[3]+b[3];return d;}
+//==============================================================================================================================
+ A_STATIC retAD2 opAAddOneD2(outAD2 d,inAD2 a,AD1 b){d[0]=a[0]+b;d[1]=a[1]+b;return d;}
+ A_STATIC retAD3 opAAddOneD3(outAD3 d,inAD3 a,AD1 b){d[0]=a[0]+b;d[1]=a[1]+b;d[2]=a[2]+b;return d;}
+ A_STATIC retAD4 opAAddOneD4(outAD4 d,inAD4 a,AD1 b){d[0]=a[0]+b;d[1]=a[1]+b;d[2]=a[2]+b;d[3]=a[3]+b;return d;}
+//------------------------------------------------------------------------------------------------------------------------------
+ A_STATIC retAF2 opAAddOneF2(outAF2 d,inAF2 a,AF1 b){d[0]=a[0]+b;d[1]=a[1]+b;return d;}
+ A_STATIC retAF3 opAAddOneF3(outAF3 d,inAF3 a,AF1 b){d[0]=a[0]+b;d[1]=a[1]+b;d[2]=a[2]+b;return d;}
+ A_STATIC retAF4 opAAddOneF4(outAF4 d,inAF4 a,AF1 b){d[0]=a[0]+b;d[1]=a[1]+b;d[2]=a[2]+b;d[3]=a[3]+b;return d;}
+//==============================================================================================================================
+ A_STATIC retAD2 opACpyD2(outAD2 d,inAD2 a){d[0]=a[0];d[1]=a[1];return d;}
+ A_STATIC retAD3 opACpyD3(outAD3 d,inAD3 a){d[0]=a[0];d[1]=a[1];d[2]=a[2];return d;}
+ A_STATIC retAD4 opACpyD4(outAD4 d,inAD4 a){d[0]=a[0];d[1]=a[1];d[2]=a[2];d[3]=a[3];return d;}
+//------------------------------------------------------------------------------------------------------------------------------
+ A_STATIC retAF2 opACpyF2(outAF2 d,inAF2 a){d[0]=a[0];d[1]=a[1];return d;}
+ A_STATIC retAF3 opACpyF3(outAF3 d,inAF3 a){d[0]=a[0];d[1]=a[1];d[2]=a[2];return d;}
+ A_STATIC retAF4 opACpyF4(outAF4 d,inAF4 a){d[0]=a[0];d[1]=a[1];d[2]=a[2];d[3]=a[3];return d;}
+//==============================================================================================================================
+ A_STATIC retAD2 opALerpD2(outAD2 d,inAD2 a,inAD2 b,inAD2 c){d[0]=ALerpD1(a[0],b[0],c[0]);d[1]=ALerpD1(a[1],b[1],c[1]);return d;}
+ A_STATIC retAD3 opALerpD3(outAD3 d,inAD3 a,inAD3 b,inAD3 c){d[0]=ALerpD1(a[0],b[0],c[0]);d[1]=ALerpD1(a[1],b[1],c[1]);d[2]=ALerpD1(a[2],b[2],c[2]);return d;}
+ A_STATIC retAD4 opALerpD4(outAD4 d,inAD4 a,inAD4 b,inAD4 c){d[0]=ALerpD1(a[0],b[0],c[0]);d[1]=ALerpD1(a[1],b[1],c[1]);d[2]=ALerpD1(a[2],b[2],c[2]);d[3]=ALerpD1(a[3],b[3],c[3]);return d;}
+//------------------------------------------------------------------------------------------------------------------------------
+ A_STATIC retAF2 opALerpF2(outAF2 d,inAF2 a,inAF2 b,inAF2 c){d[0]=ALerpF1(a[0],b[0],c[0]);d[1]=ALerpF1(a[1],b[1],c[1]);return d;}
+ A_STATIC retAF3 opALerpF3(outAF3 d,inAF3 a,inAF3 b,inAF3 c){d[0]=ALerpF1(a[0],b[0],c[0]);d[1]=ALerpF1(a[1],b[1],c[1]);d[2]=ALerpF1(a[2],b[2],c[2]);return d;}
+ A_STATIC retAF4 opALerpF4(outAF4 d,inAF4 a,inAF4 b,inAF4 c){d[0]=ALerpF1(a[0],b[0],c[0]);d[1]=ALerpF1(a[1],b[1],c[1]);d[2]=ALerpF1(a[2],b[2],c[2]);d[3]=ALerpF1(a[3],b[3],c[3]);return d;}
+//==============================================================================================================================
+ A_STATIC retAD2 opALerpOneD2(outAD2 d,inAD2 a,inAD2 b,AD1 c){d[0]=ALerpD1(a[0],b[0],c);d[1]=ALerpD1(a[1],b[1],c);return d;}
+ A_STATIC retAD3 opALerpOneD3(outAD3 d,inAD3 a,inAD3 b,AD1 c){d[0]=ALerpD1(a[0],b[0],c);d[1]=ALerpD1(a[1],b[1],c);d[2]=ALerpD1(a[2],b[2],c);return d;}
+ A_STATIC retAD4 opALerpOneD4(outAD4 d,inAD4 a,inAD4 b,AD1 c){d[0]=ALerpD1(a[0],b[0],c);d[1]=ALerpD1(a[1],b[1],c);d[2]=ALerpD1(a[2],b[2],c);d[3]=ALerpD1(a[3],b[3],c);return d;}
+//------------------------------------------------------------------------------------------------------------------------------
+ A_STATIC retAF2 opALerpOneF2(outAF2 d,inAF2 a,inAF2 b,AF1 c){d[0]=ALerpF1(a[0],b[0],c);d[1]=ALerpF1(a[1],b[1],c);return d;}
+ A_STATIC retAF3 opALerpOneF3(outAF3 d,inAF3 a,inAF3 b,AF1 c){d[0]=ALerpF1(a[0],b[0],c);d[1]=ALerpF1(a[1],b[1],c);d[2]=ALerpF1(a[2],b[2],c);return d;}
+ A_STATIC retAF4 opALerpOneF4(outAF4 d,inAF4 a,inAF4 b,AF1 c){d[0]=ALerpF1(a[0],b[0],c);d[1]=ALerpF1(a[1],b[1],c);d[2]=ALerpF1(a[2],b[2],c);d[3]=ALerpF1(a[3],b[3],c);return d;}
+//==============================================================================================================================
+ A_STATIC retAD2 opAMaxD2(outAD2 d,inAD2 a,inAD2 b){d[0]=AMaxD1(a[0],b[0]);d[1]=AMaxD1(a[1],b[1]);return d;}
+ A_STATIC retAD3 opAMaxD3(outAD3 d,inAD3 a,inAD3 b){d[0]=AMaxD1(a[0],b[0]);d[1]=AMaxD1(a[1],b[1]);d[2]=AMaxD1(a[2],b[2]);return d;}
+ A_STATIC retAD4 opAMaxD4(outAD4 d,inAD4 a,inAD4 b){d[0]=AMaxD1(a[0],b[0]);d[1]=AMaxD1(a[1],b[1]);d[2]=AMaxD1(a[2],b[2]);d[3]=AMaxD1(a[3],b[3]);return d;}
+//------------------------------------------------------------------------------------------------------------------------------
+ A_STATIC retAF2 opAMaxF2(outAF2 d,inAF2 a,inAF2 b){d[0]=AMaxF1(a[0],b[0]);d[1]=AMaxF1(a[1],b[1]);return d;}
+ A_STATIC retAF3 opAMaxF3(outAF3 d,inAF3 a,inAF3 b){d[0]=AMaxF1(a[0],b[0]);d[1]=AMaxF1(a[1],b[1]);d[2]=AMaxF1(a[2],b[2]);return d;}
+ A_STATIC retAF4 opAMaxF4(outAF4 d,inAF4 a,inAF4 b){d[0]=AMaxF1(a[0],b[0]);d[1]=AMaxF1(a[1],b[1]);d[2]=AMaxF1(a[2],b[2]);d[3]=AMaxF1(a[3],b[3]);return d;}
+//==============================================================================================================================
+ A_STATIC retAD2 opAMinD2(outAD2 d,inAD2 a,inAD2 b){d[0]=AMinD1(a[0],b[0]);d[1]=AMinD1(a[1],b[1]);return d;}
+ A_STATIC retAD3 opAMinD3(outAD3 d,inAD3 a,inAD3 b){d[0]=AMinD1(a[0],b[0]);d[1]=AMinD1(a[1],b[1]);d[2]=AMinD1(a[2],b[2]);return d;}
+ A_STATIC retAD4 opAMinD4(outAD4 d,inAD4 a,inAD4 b){d[0]=AMinD1(a[0],b[0]);d[1]=AMinD1(a[1],b[1]);d[2]=AMinD1(a[2],b[2]);d[3]=AMinD1(a[3],b[3]);return d;}
+//------------------------------------------------------------------------------------------------------------------------------
+ A_STATIC retAF2 opAMinF2(outAF2 d,inAF2 a,inAF2 b){d[0]=AMinF1(a[0],b[0]);d[1]=AMinF1(a[1],b[1]);return d;}
+ A_STATIC retAF3 opAMinF3(outAF3 d,inAF3 a,inAF3 b){d[0]=AMinF1(a[0],b[0]);d[1]=AMinF1(a[1],b[1]);d[2]=AMinF1(a[2],b[2]);return d;}
+ A_STATIC retAF4 opAMinF4(outAF4 d,inAF4 a,inAF4 b){d[0]=AMinF1(a[0],b[0]);d[1]=AMinF1(a[1],b[1]);d[2]=AMinF1(a[2],b[2]);d[3]=AMinF1(a[3],b[3]);return d;}
+//==============================================================================================================================
+ A_STATIC retAD2 opAMulD2(outAD2 d,inAD2 a,inAD2 b){d[0]=a[0]*b[0];d[1]=a[1]*b[1];return d;}
+ A_STATIC retAD3 opAMulD3(outAD3 d,inAD3 a,inAD3 b){d[0]=a[0]*b[0];d[1]=a[1]*b[1];d[2]=a[2]*b[2];return d;}
+ A_STATIC retAD4 opAMulD4(outAD4 d,inAD4 a,inAD4 b){d[0]=a[0]*b[0];d[1]=a[1]*b[1];d[2]=a[2]*b[2];d[3]=a[3]*b[3];return d;}
+//------------------------------------------------------------------------------------------------------------------------------
+ A_STATIC retAF2 opAMulF2(outAF2 d,inAF2 a,inAF2 b){d[0]=a[0]*b[0];d[1]=a[1]*b[1];return d;}
+ A_STATIC retAF3 opAMulF3(outAF3 d,inAF3 a,inAF3 b){d[0]=a[0]*b[0];d[1]=a[1]*b[1];d[2]=a[2]*b[2];return d;}
+ A_STATIC retAF4 opAMulF4(outAF4 d,inAF4 a,inAF4 b){d[0]=a[0]*b[0];d[1]=a[1]*b[1];d[2]=a[2]*b[2];d[3]=a[3]*b[3];return d;}
+//==============================================================================================================================
+ A_STATIC retAD2 opAMulOneD2(outAD2 d,inAD2 a,AD1 b){d[0]=a[0]*b;d[1]=a[1]*b;return d;}
+ A_STATIC retAD3 opAMulOneD3(outAD3 d,inAD3 a,AD1 b){d[0]=a[0]*b;d[1]=a[1]*b;d[2]=a[2]*b;return d;}
+ A_STATIC retAD4 opAMulOneD4(outAD4 d,inAD4 a,AD1 b){d[0]=a[0]*b;d[1]=a[1]*b;d[2]=a[2]*b;d[3]=a[3]*b;return d;}
+//------------------------------------------------------------------------------------------------------------------------------
+ A_STATIC retAF2 opAMulOneF2(outAF2 d,inAF2 a,AF1 b){d[0]=a[0]*b;d[1]=a[1]*b;return d;}
+ A_STATIC retAF3 opAMulOneF3(outAF3 d,inAF3 a,AF1 b){d[0]=a[0]*b;d[1]=a[1]*b;d[2]=a[2]*b;return d;}
+ A_STATIC retAF4 opAMulOneF4(outAF4 d,inAF4 a,AF1 b){d[0]=a[0]*b;d[1]=a[1]*b;d[2]=a[2]*b;d[3]=a[3]*b;return d;}
+//==============================================================================================================================
+ A_STATIC retAD2 opANegD2(outAD2 d,inAD2 a){d[0]=-a[0];d[1]=-a[1];return d;}
+ A_STATIC retAD3 opANegD3(outAD3 d,inAD3 a){d[0]=-a[0];d[1]=-a[1];d[2]=-a[2];return d;}
+ A_STATIC retAD4 opANegD4(outAD4 d,inAD4 a){d[0]=-a[0];d[1]=-a[1];d[2]=-a[2];d[3]=-a[3];return d;}
+//------------------------------------------------------------------------------------------------------------------------------
+ A_STATIC retAF2 opANegF2(outAF2 d,inAF2 a){d[0]=-a[0];d[1]=-a[1];return d;}
+ A_STATIC retAF3 opANegF3(outAF3 d,inAF3 a){d[0]=-a[0];d[1]=-a[1];d[2]=-a[2];return d;}
+ A_STATIC retAF4 opANegF4(outAF4 d,inAF4 a){d[0]=-a[0];d[1]=-a[1];d[2]=-a[2];d[3]=-a[3];return d;}
+//==============================================================================================================================
+ A_STATIC retAD2 opARcpD2(outAD2 d,inAD2 a){d[0]=ARcpD1(a[0]);d[1]=ARcpD1(a[1]);return d;}
+ A_STATIC retAD3 opARcpD3(outAD3 d,inAD3 a){d[0]=ARcpD1(a[0]);d[1]=ARcpD1(a[1]);d[2]=ARcpD1(a[2]);return d;}
+ A_STATIC retAD4 opARcpD4(outAD4 d,inAD4 a){d[0]=ARcpD1(a[0]);d[1]=ARcpD1(a[1]);d[2]=ARcpD1(a[2]);d[3]=ARcpD1(a[3]);return d;}
+//------------------------------------------------------------------------------------------------------------------------------
+ A_STATIC retAF2 opARcpF2(outAF2 d,inAF2 a){d[0]=ARcpF1(a[0]);d[1]=ARcpF1(a[1]);return d;}
+ A_STATIC retAF3 opARcpF3(outAF3 d,inAF3 a){d[0]=ARcpF1(a[0]);d[1]=ARcpF1(a[1]);d[2]=ARcpF1(a[2]);return d;}
+ A_STATIC retAF4 opARcpF4(outAF4 d,inAF4 a){d[0]=ARcpF1(a[0]);d[1]=ARcpF1(a[1]);d[2]=ARcpF1(a[2]);d[3]=ARcpF1(a[3]);return d;}
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+//_____________________________________________________________/\_______________________________________________________________
+//==============================================================================================================================
+//                                                     HALF FLOAT PACKING
+//==============================================================================================================================
+ // Convert float to half (in lower 16-bits of output).
+ // Same fast technique as documented here: ftp://ftp.fox-toolkit.org/pub/fasthalffloatconversion.pdf
+ // Supports denormals.
+ // Conversion rules are to make computations possibly "safer" on the GPU,
+ //  -INF & -NaN -> -65504
+ //  +INF & +NaN -> +65504
+ A_STATIC AU1 AU1_AH1_AF1(AF1 f){
+  static AW1 base[512]={
+   0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,
+   0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,
+   0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,
+   0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,
+   0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,
+   0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,
+   0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0000,0x0001,0x0002,0x0004,0x0008,0x0010,0x0020,0x0040,0x0080,0x0100,
+   0x0200,0x0400,0x0800,0x0c00,0x1000,0x1400,0x1800,0x1c00,0x2000,0x2400,0x2800,0x2c00,0x3000,0x3400,0x3800,0x3c00,
+   0x4000,0x4400,0x4800,0x4c00,0x5000,0x5400,0x5800,0x5c00,0x6000,0x6400,0x6800,0x6c00,0x7000,0x7400,0x7800,0x7bff,
+   0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,
+   0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,
+   0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,
+   0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,
+   0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,
+   0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,
+   0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,0x7bff,
+   0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,
+   0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,
+   0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,
+   0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,
+   0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,
+   0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,
+   0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8000,0x8001,0x8002,0x8004,0x8008,0x8010,0x8020,0x8040,0x8080,0x8100,
+   0x8200,0x8400,0x8800,0x8c00,0x9000,0x9400,0x9800,0x9c00,0xa000,0xa400,0xa800,0xac00,0xb000,0xb400,0xb800,0xbc00,
+   0xc000,0xc400,0xc800,0xcc00,0xd000,0xd400,0xd800,0xdc00,0xe000,0xe400,0xe800,0xec00,0xf000,0xf400,0xf800,0xfbff,
+   0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,
+   0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,
+   0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,
+   0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,
+   0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,
+   0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,
+   0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff,0xfbff};
+  static AB1 shift[512]={
+   0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,
+   0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,
+   0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,
+   0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,
+   0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,
+   0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,
+   0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x17,0x16,0x15,0x14,0x13,0x12,0x11,0x10,0x0f,
+   0x0e,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,
+   0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x18,
+   0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,
+   0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,
+   0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,
+   0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,
+   0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,
+   0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,
+   0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,
+   0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,
+   0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,
+   0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,
+   0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,
+   0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,
+   0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,
+   0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x17,0x16,0x15,0x14,0x13,0x12,0x11,0x10,0x0f,
+   0x0e,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,
+   0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x0d,0x18,
+   0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,
+   0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,
+   0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,
+   0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,
+   0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,
+   0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,
+   0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18,0x18};
+  union{AF1 f;AU1 u;}bits;bits.f=f;AU1 u=bits.u;AU1 i=u>>23;return (AU1)(base[i])+((u&0x7fffff)>>shift[i]);}
+//------------------------------------------------------------------------------------------------------------------------------
+ // Used to output packed constant.
+ A_STATIC AU1 AU1_AH2_AF2(inAF2 a){return AU1_AH1_AF1(a[0])+(AU1_AH1_AF1(a[1])<<16);}
+#endif
+//_____________________________________________________________/\_______________________________________________________________
+//==============================================================================================================================
+//                                                            GLSL
+//==============================================================================================================================
+#if defined(A_GLSL) && defined(A_GPU)
+ #ifndef A_SKIP_EXT
+  #ifdef A_LONG
+   #extension GL_ARB_gpu_shader_int64:require
+   #extension GL_NV_shader_atomic_int64:require
+  #endif
+//------------------------------------------------------------------------------------------------------------------------------
+  #ifdef A_WAVE
+   #extension GL_KHR_shader_subgroup_arithmetic:require
+   #extension GL_KHR_shader_subgroup_ballot:require
+   #extension GL_KHR_shader_subgroup_quad:require
+   #extension GL_KHR_shader_subgroup_shuffle:require
+  #endif
+ #endif
+//==============================================================================================================================
+ #define AP1 bool
+ #define AP2 bvec2
+ #define AP3 bvec3
+ #define AP4 bvec4
+//------------------------------------------------------------------------------------------------------------------------------
+ #define AF1 float
+ #define AF2 vec2
+ #define AF3 vec3
+ #define AF4 vec4
+//------------------------------------------------------------------------------------------------------------------------------
+ #define AU1 uint
+ #define AU2 uvec2
+ #define AU3 uvec3
+ #define AU4 uvec4
+//------------------------------------------------------------------------------------------------------------------------------
+ #define ASU1 int
+ #define ASU2 ivec2
+ #define ASU3 ivec3
+ #define ASU4 ivec4
+//==============================================================================================================================
+ #define AF1_AU1(x) uintBitsToFloat(AU1(x))
+ #define AF2_AU2(x) uintBitsToFloat(AU2(x))
+ #define AF3_AU3(x) uintBitsToFloat(AU3(x))
+ #define AF4_AU4(x) uintBitsToFloat(AU4(x))
+//------------------------------------------------------------------------------------------------------------------------------
+ #define AU1_AF1(x) floatBitsToUint(AF1(x))
+ #define AU2_AF2(x) floatBitsToUint(AF2(x))
+ #define AU3_AF3(x) floatBitsToUint(AF3(x))
+ #define AU4_AF4(x) floatBitsToUint(AF4(x))
+//==============================================================================================================================
+ AF1 AF1_x(AF1 a){return AF1(a);}
+ AF2 AF2_x(AF1 a){return AF2(a,a);}
+ AF3 AF3_x(AF1 a){return AF3(a,a,a);}
+ AF4 AF4_x(AF1 a){return AF4(a,a,a,a);}
+ #define AF1_(a) AF1_x(AF1(a))
+ #define AF2_(a) AF2_x(AF1(a))
+ #define AF3_(a) AF3_x(AF1(a))
+ #define AF4_(a) AF4_x(AF1(a))
+//------------------------------------------------------------------------------------------------------------------------------
+ AU1 AU1_x(AU1 a){return AU1(a);}
+ AU2 AU2_x(AU1 a){return AU2(a,a);}
+ AU3 AU3_x(AU1 a){return AU3(a,a,a);}
+ AU4 AU4_x(AU1 a){return AU4(a,a,a,a);}
+ #define AU1_(a) AU1_x(AU1(a))
+ #define AU2_(a) AU2_x(AU1(a))
+ #define AU3_(a) AU3_x(AU1(a))
+ #define AU4_(a) AU4_x(AU1(a))
+//==============================================================================================================================
+ AU1 AAbsSU1(AU1 a){return AU1(abs(ASU1(a)));}
+ AU2 AAbsSU2(AU2 a){return AU2(abs(ASU2(a)));}
+ AU3 AAbsSU3(AU3 a){return AU3(abs(ASU3(a)));}
+ AU4 AAbsSU4(AU4 a){return AU4(abs(ASU4(a)));}
+//------------------------------------------------------------------------------------------------------------------------------
+ AU1 ABfe(AU1 src,AU1 off,AU1 bits){return bitfieldExtract(src,ASU1(off),ASU1(bits));}
+ AU1 ABfi(AU1 src,AU1 ins,AU1 mask){return (ins&mask)|(src&(~mask));}
+ // Proxy for V_BFI_B32 where the 'mask' is set as 'bits', 'mask=(1<<bits)-1', and 'bits' needs to be an immediate.
+ AU1 ABfiM(AU1 src,AU1 ins,AU1 bits){return bitfieldInsert(src,ins,0,ASU1(bits));}
+//------------------------------------------------------------------------------------------------------------------------------
+ // V_MED3_F32.
+ AF1 AClampF1(AF1 x,AF1 n,AF1 m){return clamp(x,n,m);}
+ AF2 AClampF2(AF2 x,AF2 n,AF2 m){return clamp(x,n,m);}
+ AF3 AClampF3(AF3 x,AF3 n,AF3 m){return clamp(x,n,m);}
+ AF4 AClampF4(AF4 x,AF4 n,AF4 m){return clamp(x,n,m);}
+//------------------------------------------------------------------------------------------------------------------------------
+ // V_FRACT_F32 (note DX frac() is different).
+ AF1 AFractF1(AF1 x){return fract(x);}
+ AF2 AFractF2(AF2 x){return fract(x);}
+ AF3 AFractF3(AF3 x){return fract(x);}
+ AF4 AFractF4(AF4 x){return fract(x);}
+//------------------------------------------------------------------------------------------------------------------------------
+ AF1 ALerpF1(AF1 x,AF1 y,AF1 a){return mix(x,y,a);}
+ AF2 ALerpF2(AF2 x,AF2 y,AF2 a){return mix(x,y,a);}
+ AF3 ALerpF3(AF3 x,AF3 y,AF3 a){return mix(x,y,a);}
+ AF4 ALerpF4(AF4 x,AF4 y,AF4 a){return mix(x,y,a);}
+//------------------------------------------------------------------------------------------------------------------------------
+ // V_MAX3_F32.
+ AF1 AMax3F1(AF1 x,AF1 y,AF1 z){return max(x,max(y,z));}
+ AF2 AMax3F2(AF2 x,AF2 y,AF2 z){return max(x,max(y,z));}
+ AF3 AMax3F3(AF3 x,AF3 y,AF3 z){return max(x,max(y,z));}
+ AF4 AMax3F4(AF4 x,AF4 y,AF4 z){return max(x,max(y,z));}
+//------------------------------------------------------------------------------------------------------------------------------
+ AU1 AMax3SU1(AU1 x,AU1 y,AU1 z){return AU1(max(ASU1(x),max(ASU1(y),ASU1(z))));}
+ AU2 AMax3SU2(AU2 x,AU2 y,AU2 z){return AU2(max(ASU2(x),max(ASU2(y),ASU2(z))));}
+ AU3 AMax3SU3(AU3 x,AU3 y,AU3 z){return AU3(max(ASU3(x),max(ASU3(y),ASU3(z))));}
+ AU4 AMax3SU4(AU4 x,AU4 y,AU4 z){return AU4(max(ASU4(x),max(ASU4(y),ASU4(z))));}
+//------------------------------------------------------------------------------------------------------------------------------
+ AU1 AMax3U1(AU1 x,AU1 y,AU1 z){return max(x,max(y,z));}
+ AU2 AMax3U2(AU2 x,AU2 y,AU2 z){return max(x,max(y,z));}
+ AU3 AMax3U3(AU3 x,AU3 y,AU3 z){return max(x,max(y,z));}
+ AU4 AMax3U4(AU4 x,AU4 y,AU4 z){return max(x,max(y,z));}
+//------------------------------------------------------------------------------------------------------------------------------
+ AU1 AMaxSU1(AU1 a,AU1 b){return AU1(max(ASU1(a),ASU1(b)));}
+ AU2 AMaxSU2(AU2 a,AU2 b){return AU2(max(ASU2(a),ASU2(b)));}
+ AU3 AMaxSU3(AU3 a,AU3 b){return AU3(max(ASU3(a),ASU3(b)));}
+ AU4 AMaxSU4(AU4 a,AU4 b){return AU4(max(ASU4(a),ASU4(b)));}
+//------------------------------------------------------------------------------------------------------------------------------
+ // Clamp has an easier pattern match for med3 when some ordering is known.
+ // V_MED3_F32.
+ AF1 AMed3F1(AF1 x,AF1 y,AF1 z){return max(min(x,y),min(max(x,y),z));}
+ AF2 AMed3F2(AF2 x,AF2 y,AF2 z){return max(min(x,y),min(max(x,y),z));}
+ AF3 AMed3F3(AF3 x,AF3 y,AF3 z){return max(min(x,y),min(max(x,y),z));}
+ AF4 AMed3F4(AF4 x,AF4 y,AF4 z){return max(min(x,y),min(max(x,y),z));}
+//------------------------------------------------------------------------------------------------------------------------------
+ // V_MIN3_F32.
+ AF1 AMin3F1(AF1 x,AF1 y,AF1 z){return min(x,min(y,z));}
+ AF2 AMin3F2(AF2 x,AF2 y,AF2 z){return min(x,min(y,z));}
+ AF3 AMin3F3(AF3 x,AF3 y,AF3 z){return min(x,min(y,z));}
+ AF4 AMin3F4(AF4 x,AF4 y,AF4 z){return min(x,min(y,z));}
+//------------------------------------------------------------------------------------------------------------------------------
+ AU1 AMin3SU1(AU1 x,AU1 y,AU1 z){return AU1(min(ASU1(x),min(ASU1(y),ASU1(z))));}
+ AU2 AMin3SU2(AU2 x,AU2 y,AU2 z){return AU2(min(ASU2(x),min(ASU2(y),ASU2(z))));}
+ AU3 AMin3SU3(AU3 x,AU3 y,AU3 z){return AU3(min(ASU3(x),min(ASU3(y),ASU3(z))));}
+ AU4 AMin3SU4(AU4 x,AU4 y,AU4 z){return AU4(min(ASU4(x),min(ASU4(y),ASU4(z))));}
+//------------------------------------------------------------------------------------------------------------------------------
+ AU1 AMin3U1(AU1 x,AU1 y,AU1 z){return min(x,min(y,z));}
+ AU2 AMin3U2(AU2 x,AU2 y,AU2 z){return min(x,min(y,z));}
+ AU3 AMin3U3(AU3 x,AU3 y,AU3 z){return min(x,min(y,z));}
+ AU4 AMin3U4(AU4 x,AU4 y,AU4 z){return min(x,min(y,z));}
+//------------------------------------------------------------------------------------------------------------------------------
+ AU1 AMinSU1(AU1 a,AU1 b){return AU1(min(ASU1(a),ASU1(b)));}
+ AU2 AMinSU2(AU2 a,AU2 b){return AU2(min(ASU2(a),ASU2(b)));}
+ AU3 AMinSU3(AU3 a,AU3 b){return AU3(min(ASU3(a),ASU3(b)));}
+ AU4 AMinSU4(AU4 a,AU4 b){return AU4(min(ASU4(a),ASU4(b)));}
+//------------------------------------------------------------------------------------------------------------------------------
+ // Normalized trig. Valid input domain is {-256 to +256}. No GLSL compiler intrinsic exists to map to this currently.
+ // V_COS_F32.
+ AF1 ANCosF1(AF1 x){return cos(x*AF1_(A_2PI));}
+ AF2 ANCosF2(AF2 x){return cos(x*AF2_(A_2PI));}
+ AF3 ANCosF3(AF3 x){return cos(x*AF3_(A_2PI));}
+ AF4 ANCosF4(AF4 x){return cos(x*AF4_(A_2PI));}
+//------------------------------------------------------------------------------------------------------------------------------
+ // Normalized trig. Valid input domain is {-256 to +256}. No GLSL compiler intrinsic exists to map to this currently.
+ // V_SIN_F32.
+ AF1 ANSinF1(AF1 x){return sin(x*AF1_(A_2PI));}
+ AF2 ANSinF2(AF2 x){return sin(x*AF2_(A_2PI));}
+ AF3 ANSinF3(AF3 x){return sin(x*AF3_(A_2PI));}
+ AF4 ANSinF4(AF4 x){return sin(x*AF4_(A_2PI));}
+//------------------------------------------------------------------------------------------------------------------------------
+ AF1 ARcpF1(AF1 x){return AF1_(1.0)/x;}
+ AF2 ARcpF2(AF2 x){return AF2_(1.0)/x;}
+ AF3 ARcpF3(AF3 x){return AF3_(1.0)/x;}
+ AF4 ARcpF4(AF4 x){return AF4_(1.0)/x;}
+//------------------------------------------------------------------------------------------------------------------------------
+ AF1 ARsqF1(AF1 x){return AF1_(1.0)/sqrt(x);}
+ AF2 ARsqF2(AF2 x){return AF2_(1.0)/sqrt(x);}
+ AF3 ARsqF3(AF3 x){return AF3_(1.0)/sqrt(x);}
+ AF4 ARsqF4(AF4 x){return AF4_(1.0)/sqrt(x);}
+//------------------------------------------------------------------------------------------------------------------------------
+ AF1 ASatF1(AF1 x){return clamp(x,AF1_(0.0),AF1_(1.0));}
+ AF2 ASatF2(AF2 x){return clamp(x,AF2_(0.0),AF2_(1.0));}
+ AF3 ASatF3(AF3 x){return clamp(x,AF3_(0.0),AF3_(1.0));}
+ AF4 ASatF4(AF4 x){return clamp(x,AF4_(0.0),AF4_(1.0));}
+//------------------------------------------------------------------------------------------------------------------------------
+ AU1 AShrSU1(AU1 a,AU1 b){return AU1(ASU1(a)>>ASU1(b));}
+ AU2 AShrSU2(AU2 a,AU2 b){return AU2(ASU2(a)>>ASU2(b));}
+ AU3 AShrSU3(AU3 a,AU3 b){return AU3(ASU3(a)>>ASU3(b));}
+ AU4 AShrSU4(AU4 a,AU4 b){return AU4(ASU4(a)>>ASU4(b));}
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+//_____________________________________________________________/\_______________________________________________________________
+//==============================================================================================================================
+//                                                          GLSL BYTE
+//==============================================================================================================================
+ #ifdef A_BYTE
+  #define AB1 uint8_t
+  #define AB2 u8vec2
+  #define AB3 u8vec3
+  #define AB4 u8vec4
+//------------------------------------------------------------------------------------------------------------------------------
+  #define ASB1 int8_t
+  #define ASB2 i8vec2
+  #define ASB3 i8vec3
+  #define ASB4 i8vec4
+//------------------------------------------------------------------------------------------------------------------------------
+  AB1 AB1_x(AB1 a){return AB1(a);}
+  AB2 AB2_x(AB1 a){return AB2(a,a);}
+  AB3 AB3_x(AB1 a){return AB3(a,a,a);}
+  AB4 AB4_x(AB1 a){return AB4(a,a,a,a);}
+  #define AB1_(a) AB1_x(AB1(a))
+  #define AB2_(a) AB2_x(AB1(a))
+  #define AB3_(a) AB3_x(AB1(a))
+  #define AB4_(a) AB4_x(AB1(a))
+ #endif
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+//_____________________________________________________________/\_______________________________________________________________
+//==============================================================================================================================
+//                                                         GLSL DOUBLE
+//==============================================================================================================================
+ #ifdef A_DUBL
+  #define AD1 double
+  #define AD2 dvec2
+  #define AD3 dvec3
+  #define AD4 dvec4
+//------------------------------------------------------------------------------------------------------------------------------
+  AD1 AD1_x(AD1 a){return AD1(a);}
+  AD2 AD2_x(AD1 a){return AD2(a,a);}
+  AD3 AD3_x(AD1 a){return AD3(a,a,a);}
+  AD4 AD4_x(AD1 a){return AD4(a,a,a,a);}
+  #define AD1_(a) AD1_x(AD1(a))
+  #define AD2_(a) AD2_x(AD1(a))
+  #define AD3_(a) AD3_x(AD1(a))
+  #define AD4_(a) AD4_x(AD1(a))
+//==============================================================================================================================
+  AD1 AFractD1(AD1 x){return fract(x);}
+  AD2 AFractD2(AD2 x){return fract(x);}
+  AD3 AFractD3(AD3 x){return fract(x);}
+  AD4 AFractD4(AD4 x){return fract(x);}
+//------------------------------------------------------------------------------------------------------------------------------
+  AD1 ALerpD1(AD1 x,AD1 y,AD1 a){return mix(x,y,a);}
+  AD2 ALerpD2(AD2 x,AD2 y,AD2 a){return mix(x,y,a);}
+  AD3 ALerpD3(AD3 x,AD3 y,AD3 a){return mix(x,y,a);}
+  AD4 ALerpD4(AD4 x,AD4 y,AD4 a){return mix(x,y,a);}
+//------------------------------------------------------------------------------------------------------------------------------
+  AD1 ARcpD1(AD1 x){return AD1_(1.0)/x;}
+  AD2 ARcpD2(AD2 x){return AD2_(1.0)/x;}
+  AD3 ARcpD3(AD3 x){return AD3_(1.0)/x;}
+  AD4 ARcpD4(AD4 x){return AD4_(1.0)/x;}
+//------------------------------------------------------------------------------------------------------------------------------
+  AD1 ARsqD1(AD1 x){return AD1_(1.0)/sqrt(x);}
+  AD2 ARsqD2(AD2 x){return AD2_(1.0)/sqrt(x);}
+  AD3 ARsqD3(AD3 x){return AD3_(1.0)/sqrt(x);}
+  AD4 ARsqD4(AD4 x){return AD4_(1.0)/sqrt(x);}
+//------------------------------------------------------------------------------------------------------------------------------
+  AD1 ASatD1(AD1 x){return clamp(x,AD1_(0.0),AD1_(1.0));}
+  AD2 ASatD2(AD2 x){return clamp(x,AD2_(0.0),AD2_(1.0));}
+  AD3 ASatD3(AD3 x){return clamp(x,AD3_(0.0),AD3_(1.0));}
+  AD4 ASatD4(AD4 x){return clamp(x,AD4_(0.0),AD4_(1.0));}
+ #endif
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+//_____________________________________________________________/\_______________________________________________________________
+//==============================================================================================================================
+//                                                         GLSL LONG
+//==============================================================================================================================
+ #ifdef A_LONG
+  #define AL1 uint64_t
+  #define AL2 u64vec2
+  #define AL3 u64vec3
+  #define AL4 u64vec4
+//------------------------------------------------------------------------------------------------------------------------------
+  #define ASL1 int64_t
+  #define ASL2 i64vec2
+  #define ASL3 i64vec3
+  #define ASL4 i64vec4
+//------------------------------------------------------------------------------------------------------------------------------
+  #define AL1_AU2(x) packUint2x32(AU2(x))
+  #define AU2_AL1(x) unpackUint2x32(AL1(x))
+//------------------------------------------------------------------------------------------------------------------------------
+  AL1 AL1_x(AL1 a){return AL1(a);}
+  AL2 AL2_x(AL1 a){return AL2(a,a);}
+  AL3 AL3_x(AL1 a){return AL3(a,a,a);}
+  AL4 AL4_x(AL1 a){return AL4(a,a,a,a);}
+  #define AL1_(a) AL1_x(AL1(a))
+  #define AL2_(a) AL2_x(AL1(a))
+  #define AL3_(a) AL3_x(AL1(a))
+  #define AL4_(a) AL4_x(AL1(a))
+//==============================================================================================================================
+  AL1 AAbsSL1(AL1 a){return AL1(abs(ASL1(a)));}
+  AL2 AAbsSL2(AL2 a){return AL2(abs(ASL2(a)));}
+  AL3 AAbsSL3(AL3 a){return AL3(abs(ASL3(a)));}
+  AL4 AAbsSL4(AL4 a){return AL4(abs(ASL4(a)));}
+//------------------------------------------------------------------------------------------------------------------------------
+  AL1 AMaxSL1(AL1 a,AL1 b){return AL1(max(ASU1(a),ASU1(b)));}
+  AL2 AMaxSL2(AL2 a,AL2 b){return AL2(max(ASU2(a),ASU2(b)));}
+  AL3 AMaxSL3(AL3 a,AL3 b){return AL3(max(ASU3(a),ASU3(b)));}
+  AL4 AMaxSL4(AL4 a,AL4 b){return AL4(max(ASU4(a),ASU4(b)));}
+//------------------------------------------------------------------------------------------------------------------------------
+  AL1 AMinSL1(AL1 a,AL1 b){return AL1(min(ASU1(a),ASU1(b)));}
+  AL2 AMinSL2(AL2 a,AL2 b){return AL2(min(ASU2(a),ASU2(b)));}
+  AL3 AMinSL3(AL3 a,AL3 b){return AL3(min(ASU3(a),ASU3(b)));}
+  AL4 AMinSL4(AL4 a,AL4 b){return AL4(min(ASU4(a),ASU4(b)));}
+ #endif
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+//_____________________________________________________________/\_______________________________________________________________
+//==============================================================================================================================
+//                                                      WAVE OPERATIONS
+//==============================================================================================================================
+ #ifdef A_WAVE
+  // Where 'x' must be a compile time literal.
+  AF1 AWaveXorF1(AF1 v,AU1 x){return subgroupShuffleXor(v,x);}
+  AF2 AWaveXorF2(AF2 v,AU1 x){return subgroupShuffleXor(v,x);}
+  AF3 AWaveXorF3(AF3 v,AU1 x){return subgroupShuffleXor(v,x);}
+  AF4 AWaveXorF4(AF4 v,AU1 x){return subgroupShuffleXor(v,x);}
+  AU1 AWaveXorU1(AU1 v,AU1 x){return subgroupShuffleXor(v,x);}
+  AU2 AWaveXorU2(AU2 v,AU1 x){return subgroupShuffleXor(v,x);}
+  AU3 AWaveXorU3(AU3 v,AU1 x){return subgroupShuffleXor(v,x);}
+  AU4 AWaveXorU4(AU4 v,AU1 x){return subgroupShuffleXor(v,x);}
+ #endif
+//==============================================================================================================================
+#endif
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+//_____________________________________________________________/\_______________________________________________________________
+//==============================================================================================================================
+//
+//
+//                                                            HLSL
+//
+//
+//==============================================================================================================================
+#if defined(A_HLSL) && defined(A_GPU)
+ #ifdef A_HLSL_6_2
+  #define AP1 bool
+  #define AP2 bool2
+  #define AP3 bool3
+  #define AP4 bool4
+//------------------------------------------------------------------------------------------------------------------------------
+  #define AF1 float32_t
+  #define AF2 float32_t2
+  #define AF3 float32_t3
+  #define AF4 float32_t4
+//------------------------------------------------------------------------------------------------------------------------------
+  #define AU1 uint32_t
+  #define AU2 uint32_t2
+  #define AU3 uint32_t3
+  #define AU4 uint32_t4
+//------------------------------------------------------------------------------------------------------------------------------
+  #define ASU1 int32_t
+  #define ASU2 int32_t2
+  #define ASU3 int32_t3
+  #define ASU4 int32_t4
+ #else
+  #define AP1 bool
+  #define AP2 bool2
+  #define AP3 bool3
+  #define AP4 bool4
+//------------------------------------------------------------------------------------------------------------------------------
+  #define AF1 float
+  #define AF2 float2
+  #define AF3 float3
+  #define AF4 float4
+//------------------------------------------------------------------------------------------------------------------------------
+  #define AU1 uint
+  #define AU2 uint2
+  #define AU3 uint3
+  #define AU4 uint4
+//------------------------------------------------------------------------------------------------------------------------------
+  #define ASU1 int
+  #define ASU2 int2
+  #define ASU3 int3
+  #define ASU4 int4
+ #endif
+//==============================================================================================================================
+ #define AF1_AU1(x) asfloat(AU1(x))
+ #define AF2_AU2(x) asfloat(AU2(x))
+ #define AF3_AU3(x) asfloat(AU3(x))
+ #define AF4_AU4(x) asfloat(AU4(x))
+//------------------------------------------------------------------------------------------------------------------------------
+ #define AU1_AF1(x) asuint(AF1(x))
+ #define AU2_AF2(x) asuint(AF2(x))
+ #define AU3_AF3(x) asuint(AF3(x))
+ #define AU4_AF4(x) asuint(AF4(x))
+//------------------------------------------------------------------------------------------------------------------------------
+ AU1 AU1_AH1_AF1_x(AF1 a){return f32tof16(a);}
+ #define AU1_AH1_AF1(a) AU1_AH1_AF1_x(AF1(a))
+//------------------------------------------------------------------------------------------------------------------------------
+ AU1 AU1_AH2_AF2_x(AF2 a){return f32tof16(a.x)|(f32tof16(a.y)<<16);}
+ #define AU1_AH2_AF2(a) AU1_AH2_AF2_x(AF2(a))
+ #define AU1_AB4Unorm_AF4(x) D3DCOLORtoUBYTE4(AF4(x))
+//------------------------------------------------------------------------------------------------------------------------------
+ AF2 AF2_AH2_AU1_x(AU1 x){return AF2(f16tof32(x&0xFFFF),f16tof32(x>>16));}
+ #define AF2_AH2_AU1(x) AF2_AH2_AU1_x(AU1(x))
+//==============================================================================================================================
+ AF1 AF1_x(AF1 a){return AF1(a);}
+ AF2 AF2_x(AF1 a){return AF2(a,a);}
+ AF3 AF3_x(AF1 a){return AF3(a,a,a);}
+ AF4 AF4_x(AF1 a){return AF4(a,a,a,a);}
+ #define AF1_(a) AF1_x(AF1(a))
+ #define AF2_(a) AF2_x(AF1(a))
+ #define AF3_(a) AF3_x(AF1(a))
+ #define AF4_(a) AF4_x(AF1(a))
+//------------------------------------------------------------------------------------------------------------------------------
+ AU1 AU1_x(AU1 a){return AU1(a);}
+ AU2 AU2_x(AU1 a){return AU2(a,a);}
+ AU3 AU3_x(AU1 a){return AU3(a,a,a);}
+ AU4 AU4_x(AU1 a){return AU4(a,a,a,a);}
+ #define AU1_(a) AU1_x(AU1(a))
+ #define AU2_(a) AU2_x(AU1(a))
+ #define AU3_(a) AU3_x(AU1(a))
+ #define AU4_(a) AU4_x(AU1(a))
+//==============================================================================================================================
+ AU1 AAbsSU1(AU1 a){return AU1(abs(ASU1(a)));}
+ AU2 AAbsSU2(AU2 a){return AU2(abs(ASU2(a)));}
+ AU3 AAbsSU3(AU3 a){return AU3(abs(ASU3(a)));}
+ AU4 AAbsSU4(AU4 a){return AU4(abs(ASU4(a)));}
+//------------------------------------------------------------------------------------------------------------------------------
+ AU1 ABfe(AU1 src,AU1 off,AU1 bits){AU1 mask=(1u<<bits)-1;return (src>>off)&mask;}
+ AU1 ABfi(AU1 src,AU1 ins,AU1 mask){return (ins&mask)|(src&(~mask));}
+ AU1 ABfiM(AU1 src,AU1 ins,AU1 bits){AU1 mask=(1u<<bits)-1;return (ins&mask)|(src&(~mask));}
+//------------------------------------------------------------------------------------------------------------------------------
+ AF1 AClampF1(AF1 x,AF1 n,AF1 m){return max(n,min(x,m));}
+ AF2 AClampF2(AF2 x,AF2 n,AF2 m){return max(n,min(x,m));}
+ AF3 AClampF3(AF3 x,AF3 n,AF3 m){return max(n,min(x,m));}
+ AF4 AClampF4(AF4 x,AF4 n,AF4 m){return max(n,min(x,m));}
+//------------------------------------------------------------------------------------------------------------------------------
+ AF1 AFractF1(AF1 x){return x-floor(x);}
+ AF2 AFractF2(AF2 x){return x-floor(x);}
+ AF3 AFractF3(AF3 x){return x-floor(x);}
+ AF4 AFractF4(AF4 x){return x-floor(x);}
+//------------------------------------------------------------------------------------------------------------------------------
+ AF1 ALerpF1(AF1 x,AF1 y,AF1 a){return lerp(x,y,a);}
+ AF2 ALerpF2(AF2 x,AF2 y,AF2 a){return lerp(x,y,a);}
+ AF3 ALerpF3(AF3 x,AF3 y,AF3 a){return lerp(x,y,a);}
+ AF4 ALerpF4(AF4 x,AF4 y,AF4 a){return lerp(x,y,a);}
+//------------------------------------------------------------------------------------------------------------------------------
+ AF1 AMax3F1(AF1 x,AF1 y,AF1 z){return max(x,max(y,z));}
+ AF2 AMax3F2(AF2 x,AF2 y,AF2 z){return max(x,max(y,z));}
+ AF3 AMax3F3(AF3 x,AF3 y,AF3 z){return max(x,max(y,z));}
+ AF4 AMax3F4(AF4 x,AF4 y,AF4 z){return max(x,max(y,z));}
+//------------------------------------------------------------------------------------------------------------------------------
+ AU1 AMax3SU1(AU1 x,AU1 y,AU1 z){return AU1(max(ASU1(x),max(ASU1(y),ASU1(z))));}
+ AU2 AMax3SU2(AU2 x,AU2 y,AU2 z){return AU2(max(ASU2(x),max(ASU2(y),ASU2(z))));}
+ AU3 AMax3SU3(AU3 x,AU3 y,AU3 z){return AU3(max(ASU3(x),max(ASU3(y),ASU3(z))));}
+ AU4 AMax3SU4(AU4 x,AU4 y,AU4 z){return AU4(max(ASU4(x),max(ASU4(y),ASU4(z))));}
+//------------------------------------------------------------------------------------------------------------------------------
+ AU1 AMax3U1(AU1 x,AU1 y,AU1 z){return max(x,max(y,z));}
+ AU2 AMax3U2(AU2 x,AU2 y,AU2 z){return max(x,max(y,z));}
+ AU3 AMax3U3(AU3 x,AU3 y,AU3 z){return max(x,max(y,z));}
+ AU4 AMax3U4(AU4 x,AU4 y,AU4 z){return max(x,max(y,z));}
+//------------------------------------------------------------------------------------------------------------------------------
+ AU1 AMaxSU1(AU1 a,AU1 b){return AU1(max(ASU1(a),ASU1(b)));}
+ AU2 AMaxSU2(AU2 a,AU2 b){return AU2(max(ASU2(a),ASU2(b)));}
+ AU3 AMaxSU3(AU3 a,AU3 b){return AU3(max(ASU3(a),ASU3(b)));}
+ AU4 AMaxSU4(AU4 a,AU4 b){return AU4(max(ASU4(a),ASU4(b)));}
+//------------------------------------------------------------------------------------------------------------------------------
+ AF1 AMed3F1(AF1 x,AF1 y,AF1 z){return max(min(x,y),min(max(x,y),z));}
+ AF2 AMed3F2(AF2 x,AF2 y,AF2 z){return max(min(x,y),min(max(x,y),z));}
+ AF3 AMed3F3(AF3 x,AF3 y,AF3 z){return max(min(x,y),min(max(x,y),z));}
+ AF4 AMed3F4(AF4 x,AF4 y,AF4 z){return max(min(x,y),min(max(x,y),z));}
+//------------------------------------------------------------------------------------------------------------------------------
+ AF1 AMin3F1(AF1 x,AF1 y,AF1 z){return min(x,min(y,z));}
+ AF2 AMin3F2(AF2 x,AF2 y,AF2 z){return min(x,min(y,z));}
+ AF3 AMin3F3(AF3 x,AF3 y,AF3 z){return min(x,min(y,z));}
+ AF4 AMin3F4(AF4 x,AF4 y,AF4 z){return min(x,min(y,z));}
+//------------------------------------------------------------------------------------------------------------------------------
+ AU1 AMin3SU1(AU1 x,AU1 y,AU1 z){return AU1(min(ASU1(x),min(ASU1(y),ASU1(z))));}
+ AU2 AMin3SU2(AU2 x,AU2 y,AU2 z){return AU2(min(ASU2(x),min(ASU2(y),ASU2(z))));}
+ AU3 AMin3SU3(AU3 x,AU3 y,AU3 z){return AU3(min(ASU3(x),min(ASU3(y),ASU3(z))));}
+ AU4 AMin3SU4(AU4 x,AU4 y,AU4 z){return AU4(min(ASU4(x),min(ASU4(y),ASU4(z))));}
+//------------------------------------------------------------------------------------------------------------------------------
+ AU1 AMin3U1(AU1 x,AU1 y,AU1 z){return min(x,min(y,z));}
+ AU2 AMin3U2(AU2 x,AU2 y,AU2 z){return min(x,min(y,z));}
+ AU3 AMin3U3(AU3 x,AU3 y,AU3 z){return min(x,min(y,z));}
+ AU4 AMin3U4(AU4 x,AU4 y,AU4 z){return min(x,min(y,z));}
+//------------------------------------------------------------------------------------------------------------------------------
+ AU1 AMinSU1(AU1 a,AU1 b){return AU1(min(ASU1(a),ASU1(b)));}
+ AU2 AMinSU2(AU2 a,AU2 b){return AU2(min(ASU2(a),ASU2(b)));}
+ AU3 AMinSU3(AU3 a,AU3 b){return AU3(min(ASU3(a),ASU3(b)));}
+ AU4 AMinSU4(AU4 a,AU4 b){return AU4(min(ASU4(a),ASU4(b)));}
+//------------------------------------------------------------------------------------------------------------------------------
+ AF1 ANCosF1(AF1 x){return cos(x*AF1_(A_2PI));}
+ AF2 ANCosF2(AF2 x){return cos(x*AF2_(A_2PI));}
+ AF3 ANCosF3(AF3 x){return cos(x*AF3_(A_2PI));}
+ AF4 ANCosF4(AF4 x){return cos(x*AF4_(A_2PI));}
+//------------------------------------------------------------------------------------------------------------------------------
+ AF1 ANSinF1(AF1 x){return sin(x*AF1_(A_2PI));}
+ AF2 ANSinF2(AF2 x){return sin(x*AF2_(A_2PI));}
+ AF3 ANSinF3(AF3 x){return sin(x*AF3_(A_2PI));}
+ AF4 ANSinF4(AF4 x){return sin(x*AF4_(A_2PI));}
+//------------------------------------------------------------------------------------------------------------------------------
+ AF1 ARcpF1(AF1 x){return rcp(x);}
+ AF2 ARcpF2(AF2 x){return rcp(x);}
+ AF3 ARcpF3(AF3 x){return rcp(x);}
+ AF4 ARcpF4(AF4 x){return rcp(x);}
+//------------------------------------------------------------------------------------------------------------------------------
+ AF1 ARsqF1(AF1 x){return rsqrt(x);}
+ AF2 ARsqF2(AF2 x){return rsqrt(x);}
+ AF3 ARsqF3(AF3 x){return rsqrt(x);}
+ AF4 ARsqF4(AF4 x){return rsqrt(x);}
+//------------------------------------------------------------------------------------------------------------------------------
+ AF1 ASatF1(AF1 x){return saturate(x);}
+ AF2 ASatF2(AF2 x){return saturate(x);}
+ AF3 ASatF3(AF3 x){return saturate(x);}
+ AF4 ASatF4(AF4 x){return saturate(x);}
+//------------------------------------------------------------------------------------------------------------------------------
+ AU1 AShrSU1(AU1 a,AU1 b){return AU1(ASU1(a)>>ASU1(b));}
+ AU2 AShrSU2(AU2 a,AU2 b){return AU2(ASU2(a)>>ASU2(b));}
+ AU3 AShrSU3(AU3 a,AU3 b){return AU3(ASU3(a)>>ASU3(b));}
+ AU4 AShrSU4(AU4 a,AU4 b){return AU4(ASU4(a)>>ASU4(b));}
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+//_____________________________________________________________/\_______________________________________________________________
+//==============================================================================================================================
+//                                                          HLSL BYTE
+//==============================================================================================================================
+ #ifdef A_BYTE
+ #endif
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+//_____________________________________________________________/\_______________________________________________________________
+//==============================================================================================================================
+//                                                         HLSL DOUBLE
+//==============================================================================================================================
+ #ifdef A_DUBL
+  #ifdef A_HLSL_6_2
+   #define AD1 float64_t
+   #define AD2 float64_t2
+   #define AD3 float64_t3
+   #define AD4 float64_t4
+  #else
+   #define AD1 double
+   #define AD2 double2
+   #define AD3 double3
+   #define AD4 double4
+  #endif
+//------------------------------------------------------------------------------------------------------------------------------
+  AD1 AD1_x(AD1 a){return AD1(a);}
+  AD2 AD2_x(AD1 a){return AD2(a,a);}
+  AD3 AD3_x(AD1 a){return AD3(a,a,a);}
+  AD4 AD4_x(AD1 a){return AD4(a,a,a,a);}
+  #define AD1_(a) AD1_x(AD1(a))
+  #define AD2_(a) AD2_x(AD1(a))
+  #define AD3_(a) AD3_x(AD1(a))
+  #define AD4_(a) AD4_x(AD1(a))
+//==============================================================================================================================
+  AD1 AFractD1(AD1 a){return a-floor(a);}
+  AD2 AFractD2(AD2 a){return a-floor(a);}
+  AD3 AFractD3(AD3 a){return a-floor(a);}
+  AD4 AFractD4(AD4 a){return a-floor(a);}
+//------------------------------------------------------------------------------------------------------------------------------
+  AD1 ALerpD1(AD1 x,AD1 y,AD1 a){return lerp(x,y,a);}
+  AD2 ALerpD2(AD2 x,AD2 y,AD2 a){return lerp(x,y,a);}
+  AD3 ALerpD3(AD3 x,AD3 y,AD3 a){return lerp(x,y,a);}
+  AD4 ALerpD4(AD4 x,AD4 y,AD4 a){return lerp(x,y,a);}
+//------------------------------------------------------------------------------------------------------------------------------
+  AD1 ARcpD1(AD1 x){return rcp(x);}
+  AD2 ARcpD2(AD2 x){return rcp(x);}
+  AD3 ARcpD3(AD3 x){return rcp(x);}
+  AD4 ARcpD4(AD4 x){return rcp(x);}
+//------------------------------------------------------------------------------------------------------------------------------
+  AD1 ARsqD1(AD1 x){return rsqrt(x);}
+  AD2 ARsqD2(AD2 x){return rsqrt(x);}
+  AD3 ARsqD3(AD3 x){return rsqrt(x);}
+  AD4 ARsqD4(AD4 x){return rsqrt(x);}
+//------------------------------------------------------------------------------------------------------------------------------
+  AD1 ASatD1(AD1 x){return saturate(x);}
+  AD2 ASatD2(AD2 x){return saturate(x);}
+  AD3 ASatD3(AD3 x){return saturate(x);}
+  AD4 ASatD4(AD4 x){return saturate(x);}
+ #endif
+//==============================================================================================================================
+//                                                         HLSL WAVE
+//==============================================================================================================================
+ #ifdef A_WAVE
+  // Where 'x' must be a compile time literal.
+  AF1 AWaveXorF1(AF1 v,AU1 x){return WaveReadLaneAt(v,WaveGetLaneIndex()^x);}
+  AF2 AWaveXorF2(AF2 v,AU1 x){return WaveReadLaneAt(v,WaveGetLaneIndex()^x);}
+  AF3 AWaveXorF3(AF3 v,AU1 x){return WaveReadLaneAt(v,WaveGetLaneIndex()^x);}
+  AF4 AWaveXorF4(AF4 v,AU1 x){return WaveReadLaneAt(v,WaveGetLaneIndex()^x);}
+  AU1 AWaveXorU1(AU1 v,AU1 x){return WaveReadLaneAt(v,WaveGetLaneIndex()^x);}
+  AU2 AWaveXorU1(AU2 v,AU1 x){return WaveReadLaneAt(v,WaveGetLaneIndex()^x);}
+  AU3 AWaveXorU1(AU3 v,AU1 x){return WaveReadLaneAt(v,WaveGetLaneIndex()^x);}
+  AU4 AWaveXorU1(AU4 v,AU1 x){return WaveReadLaneAt(v,WaveGetLaneIndex()^x);}
+ #endif
+//==============================================================================================================================
+#endif
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+//_____________________________________________________________/\_______________________________________________________________
+//==============================================================================================================================
+//
+//
+//                                                          GPU COMMON
+//
+//
+//==============================================================================================================================
+#ifdef A_GPU
+ // Negative and positive infinity.
+ #define A_INFP_F AF1_AU1(0x7f800000u)
+ #define A_INFN_F AF1_AU1(0xff800000u)
+//------------------------------------------------------------------------------------------------------------------------------
+ // Copy sign from 's' to positive 'd'.
+ AF1 ACpySgnF1(AF1 d,AF1 s){return AF1_AU1(AU1_AF1(d)|(AU1_AF1(s)&AU1_(0x80000000u)));}
+ AF2 ACpySgnF2(AF2 d,AF2 s){return AF2_AU2(AU2_AF2(d)|(AU2_AF2(s)&AU2_(0x80000000u)));}
+ AF3 ACpySgnF3(AF3 d,AF3 s){return AF3_AU3(AU3_AF3(d)|(AU3_AF3(s)&AU3_(0x80000000u)));}
+ AF4 ACpySgnF4(AF4 d,AF4 s){return AF4_AU4(AU4_AF4(d)|(AU4_AF4(s)&AU4_(0x80000000u)));}
+//------------------------------------------------------------------------------------------------------------------------------
+ // Single operation to return (useful to create a mask to use in lerp for branch free logic),
+ //  m=NaN := 0
+ //  m>=0  := 0
+ //  m<0   := 1
+ // Uses the following useful floating point logic,
+ //  saturate(+a*(-INF)==-INF) := 0
+ //  saturate( 0*(-INF)== NaN) := 0
+ //  saturate(-a*(-INF)==+INF) := 1
+ AF1 ASignedF1(AF1 m){return ASatF1(m*AF1_(A_INFN_F));}
+ AF2 ASignedF2(AF2 m){return ASatF2(m*AF2_(A_INFN_F));}
+ AF3 ASignedF3(AF3 m){return ASatF3(m*AF3_(A_INFN_F));}
+ AF4 ASignedF4(AF4 m){return ASatF4(m*AF4_(A_INFN_F));}
+//------------------------------------------------------------------------------------------------------------------------------
+ AF1 AGtZeroF1(AF1 m){return ASatF1(m*AF1_(A_INFP_F));}
+ AF2 AGtZeroF2(AF2 m){return ASatF2(m*AF2_(A_INFP_F));}
+ AF3 AGtZeroF3(AF3 m){return ASatF3(m*AF3_(A_INFP_F));}
+ AF4 AGtZeroF4(AF4 m){return ASatF4(m*AF4_(A_INFP_F));}
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+//_____________________________________________________________/\_______________________________________________________________
+//==============================================================================================================================
+//                                                [FIS] FLOAT INTEGER SORTABLE
+//------------------------------------------------------------------------------------------------------------------------------
+// Float to integer sortable.
+//  - If sign bit=0, flip the sign bit (positives).
+//  - If sign bit=1, flip all bits     (negatives).
+// Integer sortable to float.
+//  - If sign bit=1, flip the sign bit (positives).
+//  - If sign bit=0, flip all bits     (negatives).
+// Has nice side effects.
+//  - Larger integers are more positive values.
+//  - Float zero is mapped to center of integers (so clear to integer zero is a nice default for atomic max usage).
+// Burns 3 ops for conversion {shift,or,xor}.
+//==============================================================================================================================
+ AU1 AFisToU1(AU1 x){return x^(( AShrSU1(x,AU1_(31)))|AU1_(0x80000000));}
+ AU1 AFisFromU1(AU1 x){return x^((~AShrSU1(x,AU1_(31)))|AU1_(0x80000000));}
+//------------------------------------------------------------------------------------------------------------------------------
+ // Just adjust high 16-bit value (useful when upper part of 32-bit word is a 16-bit float value).
+ AU1 AFisToHiU1(AU1 x){return x^(( AShrSU1(x,AU1_(15)))|AU1_(0x80000000));}
+ AU1 AFisFromHiU1(AU1 x){return x^((~AShrSU1(x,AU1_(15)))|AU1_(0x80000000));}
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+//_____________________________________________________________/\_______________________________________________________________
+//==============================================================================================================================
+//                                               [BUC] BYTE UNSIGNED CONVERSION
+//------------------------------------------------------------------------------------------------------------------------------
+// Designed to use the optimal conversion, enables the scaling to possibly be factored into other computation.
+// Works on a range of {0 to A_BUC_<32,16>}, for <32-bit, and 16-bit> respectively.
+//------------------------------------------------------------------------------------------------------------------------------
+// OPCODE NOTES
+// ============
+// GCN does not do UNORM or SNORM for bytes in opcodes.
+//  - V_CVT_F32_UBYTE{0,1,2,3} - Unsigned byte to float.
+//  - V_CVT_PKACC_U8_F32 - Float to unsigned byte (does bit-field insert into 32-bit integer).
+// V_PERM_B32 does byte packing with ability to zero fill bytes as well.
+//  - Can pull out byte values from two sources, and zero fill upper 8-bits of packed hi and lo.
+//------------------------------------------------------------------------------------------------------------------------------
+// BYTE : FLOAT - ABuc{0,1,2,3}{To,From}U1() - Designed for V_CVT_F32_UBYTE* and V_CVT_PKACCUM_U8_F32 ops.
+// ====   =====
+//    0 : 0
+//    1 : 1
+//     ...
+//  255 : 255
+//      : 256 (just outside the encoding range)
+//------------------------------------------------------------------------------------------------------------------------------
+// BYTE : FLOAT - ABuc{0,1,2,3}{To,From}U2() - Designed for 16-bit denormal tricks and V_PERM_B32.
+// ====   =====
+//    0 : 0
+//    1 : 1/512
+//    2 : 1/256
+//     ...
+//   64 : 1/8
+//  128 : 1/4
+//  255 : 255/512
+//      : 1/2 (just outside the encoding range)
+//------------------------------------------------------------------------------------------------------------------------------
+// OPTIMAL IMPLEMENTATIONS ON AMD ARCHITECTURES
+// ============================================
+// r=ABuc0FromU1(i)
+//   V_CVT_F32_UBYTE0 r,i
+// --------------------------------------------
+// r=ABuc0ToU1(d,i)
+//   V_CVT_PKACCUM_U8_F32 r,i,0,d
+// --------------------------------------------
+// d=ABuc0FromU2(i)
+//   Where 'k0' is an SGPR with 0x0E0A
+//   Where 'k1' is an SGPR with {32768.0} packed into the lower 16-bits
+//   V_PERM_B32 d,i.x,i.y,k0
+//   V_PK_FMA_F16 d,d,k1.x,0
+// --------------------------------------------
+// r=ABuc0ToU2(d,i)
+//   Where 'k0' is an SGPR with {1.0/32768.0} packed into the lower 16-bits
+//   Where 'k1' is an SGPR with 0x????
+//   Where 'k2' is an SGPR with 0x????
+//   V_PK_FMA_F16 i,i,k0.x,0
+//   V_PERM_B32 r.x,i,i,k1
+//   V_PERM_B32 r.y,i,i,k2
+//==============================================================================================================================
+ // Peak range for 32-bit and 16-bit operations.
+ #define A_BUC_32 (255.0)
+ #define A_BUC_16 (255.0/512.0)
+//==============================================================================================================================
+ #if 1
+  // Designed to be one V_CVT_PKACCUM_U8_F32.
+  // The extra min is required to pattern match to V_CVT_PKACCUM_U8_F32.
+  AU1 ABuc0ToU1(AU1 d,AF1 i){return (d&0xffffff00u)|((min(AU1(i),255u)    )&(0x000000ffu));}
+  AU1 ABuc1ToU1(AU1 d,AF1 i){return (d&0xffff00ffu)|((min(AU1(i),255u)<< 8)&(0x0000ff00u));}
+  AU1 ABuc2ToU1(AU1 d,AF1 i){return (d&0xff00ffffu)|((min(AU1(i),255u)<<16)&(0x00ff0000u));}
+  AU1 ABuc3ToU1(AU1 d,AF1 i){return (d&0x00ffffffu)|((min(AU1(i),255u)<<24)&(0xff000000u));}
+//------------------------------------------------------------------------------------------------------------------------------
+  // Designed to be one V_CVT_F32_UBYTE*.
+  AF1 ABuc0FromU1(AU1 i){return AF1((i    )&255u);}
+  AF1 ABuc1FromU1(AU1 i){return AF1((i>> 8)&255u);}
+  AF1 ABuc2FromU1(AU1 i){return AF1((i>>16)&255u);}
+  AF1 ABuc3FromU1(AU1 i){return AF1((i>>24)&255u);}
+ #endif
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+//_____________________________________________________________/\_______________________________________________________________
+//==============================================================================================================================
+//                                                 [BSC] BYTE SIGNED CONVERSION
+//------------------------------------------------------------------------------------------------------------------------------
+// Similar to [BUC].
+// Works on a range of {-/+ A_BSC_<32,16>}, for <32-bit, and 16-bit> respectively.
+//------------------------------------------------------------------------------------------------------------------------------
+// ENCODING (without zero-based encoding)
+// ========
+//   0 = unused (can be used to mean something else)
+//   1 = lowest value
+// 128 = exact zero center (zero based encoding
+// 255 = highest value
+//------------------------------------------------------------------------------------------------------------------------------
+// Zero-based [Zb] flips the MSB bit of the byte (making 128 "exact zero" actually zero).
+// This is useful if there is a desire for cleared values to decode as zero.
+//------------------------------------------------------------------------------------------------------------------------------
+// BYTE : FLOAT - ABsc{0,1,2,3}{To,From}U2() - Designed for 16-bit denormal tricks and V_PERM_B32.
+// ====   =====
+//    0 : -127/512 (unused)
+//    1 : -126/512
+//    2 : -125/512
+//     ...
+//  128 : 0
+//     ...
+//  255 : 127/512
+//      : 1/4 (just outside the encoding range)
+//==============================================================================================================================
+ // Peak range for 32-bit and 16-bit operations.
+ #define A_BSC_32 (127.0)
+ #define A_BSC_16 (127.0/512.0)
+//==============================================================================================================================
+ #if 1
+  AU1 ABsc0ToU1(AU1 d,AF1 i){return (d&0xffffff00u)|((min(AU1(i+128.0),255u)    )&(0x000000ffu));}
+  AU1 ABsc1ToU1(AU1 d,AF1 i){return (d&0xffff00ffu)|((min(AU1(i+128.0),255u)<< 8)&(0x0000ff00u));}
+  AU1 ABsc2ToU1(AU1 d,AF1 i){return (d&0xff00ffffu)|((min(AU1(i+128.0),255u)<<16)&(0x00ff0000u));}
+  AU1 ABsc3ToU1(AU1 d,AF1 i){return (d&0x00ffffffu)|((min(AU1(i+128.0),255u)<<24)&(0xff000000u));}
+//------------------------------------------------------------------------------------------------------------------------------
+  AU1 ABsc0ToZbU1(AU1 d,AF1 i){return ((d&0xffffff00u)|((min(AU1(trunc(i)+128.0),255u)    )&(0x000000ffu)))^0x00000080u;}
+  AU1 ABsc1ToZbU1(AU1 d,AF1 i){return ((d&0xffff00ffu)|((min(AU1(trunc(i)+128.0),255u)<< 8)&(0x0000ff00u)))^0x00008000u;}
+  AU1 ABsc2ToZbU1(AU1 d,AF1 i){return ((d&0xff00ffffu)|((min(AU1(trunc(i)+128.0),255u)<<16)&(0x00ff0000u)))^0x00800000u;}
+  AU1 ABsc3ToZbU1(AU1 d,AF1 i){return ((d&0x00ffffffu)|((min(AU1(trunc(i)+128.0),255u)<<24)&(0xff000000u)))^0x80000000u;}
+//------------------------------------------------------------------------------------------------------------------------------
+  AF1 ABsc0FromU1(AU1 i){return AF1((i    )&255u)-128.0;}
+  AF1 ABsc1FromU1(AU1 i){return AF1((i>> 8)&255u)-128.0;}
+  AF1 ABsc2FromU1(AU1 i){return AF1((i>>16)&255u)-128.0;}
+  AF1 ABsc3FromU1(AU1 i){return AF1((i>>24)&255u)-128.0;}
+//------------------------------------------------------------------------------------------------------------------------------
+  AF1 ABsc0FromZbU1(AU1 i){return AF1(((i    )&255u)^0x80u)-128.0;}
+  AF1 ABsc1FromZbU1(AU1 i){return AF1(((i>> 8)&255u)^0x80u)-128.0;}
+  AF1 ABsc2FromZbU1(AU1 i){return AF1(((i>>16)&255u)^0x80u)-128.0;}
+  AF1 ABsc3FromZbU1(AU1 i){return AF1(((i>>24)&255u)^0x80u)-128.0;}
+ #endif
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+//_____________________________________________________________/\_______________________________________________________________
+//==============================================================================================================================
+//                                                    FLOAT APPROXIMATIONS
+//------------------------------------------------------------------------------------------------------------------------------
+// Michal Drobot has an excellent presentation on these: "Low Level Optimizations For GCN",
+//  - Idea dates back to SGI, then to Quake 3, etc.
+//  - https://michaldrobot.files.wordpress.com/2014/05/gcn_alu_opt_digitaldragons2014.pdf
+//     - sqrt(x)=rsqrt(x)*x
+//     - rcp(x)=rsqrt(x)*rsqrt(x) for positive x
+//  - https://github.com/michaldrobot/ShaderFastLibs/blob/master/ShaderFastMathLib.h
+//------------------------------------------------------------------------------------------------------------------------------
+// These below are from perhaps less complete searching for optimal.
+// Used FP16 normal range for testing with +4096 32-bit step size for sampling error.
+// So these match up well with the half approximations.
+//==============================================================================================================================
+ AF1 APrxLoSqrtF1(AF1 a){return AF1_AU1((AU1_AF1(a)>>AU1_(1))+AU1_(0x1fbc4639));}
+ AF1 APrxLoRcpF1(AF1 a){return AF1_AU1(AU1_(0x7ef07ebb)-AU1_AF1(a));}
+ AF1 APrxMedRcpF1(AF1 a){AF1 b=AF1_AU1(AU1_(0x7ef19fff)-AU1_AF1(a));return b*(-b*a+AF1_(2.0));}
+ AF1 APrxLoRsqF1(AF1 a){return AF1_AU1(AU1_(0x5f347d74)-(AU1_AF1(a)>>AU1_(1)));}
+//------------------------------------------------------------------------------------------------------------------------------
+ AF2 APrxLoSqrtF2(AF2 a){return AF2_AU2((AU2_AF2(a)>>AU2_(1))+AU2_(0x1fbc4639));}
+ AF2 APrxLoRcpF2(AF2 a){return AF2_AU2(AU2_(0x7ef07ebb)-AU2_AF2(a));}
+ AF2 APrxMedRcpF2(AF2 a){AF2 b=AF2_AU2(AU2_(0x7ef19fff)-AU2_AF2(a));return b*(-b*a+AF2_(2.0));}
+ AF2 APrxLoRsqF2(AF2 a){return AF2_AU2(AU2_(0x5f347d74)-(AU2_AF2(a)>>AU2_(1)));}
+//------------------------------------------------------------------------------------------------------------------------------
+ AF3 APrxLoSqrtF3(AF3 a){return AF3_AU3((AU3_AF3(a)>>AU3_(1))+AU3_(0x1fbc4639));}
+ AF3 APrxLoRcpF3(AF3 a){return AF3_AU3(AU3_(0x7ef07ebb)-AU3_AF3(a));}
+ AF3 APrxMedRcpF3(AF3 a){AF3 b=AF3_AU3(AU3_(0x7ef19fff)-AU3_AF3(a));return b*(-b*a+AF3_(2.0));}
+ AF3 APrxLoRsqF3(AF3 a){return AF3_AU3(AU3_(0x5f347d74)-(AU3_AF3(a)>>AU3_(1)));}
+//------------------------------------------------------------------------------------------------------------------------------
+ AF4 APrxLoSqrtF4(AF4 a){return AF4_AU4((AU4_AF4(a)>>AU4_(1))+AU4_(0x1fbc4639));}
+ AF4 APrxLoRcpF4(AF4 a){return AF4_AU4(AU4_(0x7ef07ebb)-AU4_AF4(a));}
+ AF4 APrxMedRcpF4(AF4 a){AF4 b=AF4_AU4(AU4_(0x7ef19fff)-AU4_AF4(a));return b*(-b*a+AF4_(2.0));}
+ AF4 APrxLoRsqF4(AF4 a){return AF4_AU4(AU4_(0x5f347d74)-(AU4_AF4(a)>>AU4_(1)));}
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+//_____________________________________________________________/\_______________________________________________________________
+//==============================================================================================================================
+//                                                    PQ APPROXIMATIONS
+//------------------------------------------------------------------------------------------------------------------------------
+// PQ is very close to x^(1/8). The functions below Use the fast float approximation method to do
+// PQ<~>Gamma2 (4th power and fast 4th root) and PQ<~>Linear (8th power and fast 8th root). Maximum error is ~0.2%.
+//==============================================================================================================================
+// Helpers
+ AF1 Quart(AF1 a) { a = a * a; return a * a;}
+ AF1 Oct(AF1 a) { a = a * a; a = a * a; return a * a; }
+ AF2 Quart(AF2 a) { a = a * a; return a * a; }
+ AF2 Oct(AF2 a) { a = a * a; a = a * a; return a * a; }
+ AF3 Quart(AF3 a) { a = a * a; return a * a; }
+ AF3 Oct(AF3 a) { a = a * a; a = a * a; return a * a; }
+ AF4 Quart(AF4 a) { a = a * a; return a * a; }
+ AF4 Oct(AF4 a) { a = a * a; a = a * a; return a * a; }
+ //------------------------------------------------------------------------------------------------------------------------------
+ AF1 APrxPQToGamma2(AF1 a) { return Quart(a); }
+ AF1 APrxPQToLinear(AF1 a) { return Oct(a); }
+ AF1 APrxLoGamma2ToPQ(AF1 a) { return AF1_AU1((AU1_AF1(a) >> AU1_(2)) + AU1_(0x2F9A4E46)); }
+ AF1 APrxMedGamma2ToPQ(AF1 a) { AF1 b = AF1_AU1((AU1_AF1(a) >> AU1_(2)) + AU1_(0x2F9A4E46)); AF1 b4 = Quart(b); return b - b * (b4 - a) / (AF1_(4.0) * b4); }
+ AF1 APrxHighGamma2ToPQ(AF1 a) { return sqrt(sqrt(a)); }
+ AF1 APrxLoLinearToPQ(AF1 a) { return AF1_AU1((AU1_AF1(a) >> AU1_(3)) + AU1_(0x378D8723)); }
+ AF1 APrxMedLinearToPQ(AF1 a) { AF1 b = AF1_AU1((AU1_AF1(a) >> AU1_(3)) + AU1_(0x378D8723)); AF1 b8 = Oct(b); return b - b * (b8 - a) / (AF1_(8.0) * b8); }
+ AF1 APrxHighLinearToPQ(AF1 a) { return sqrt(sqrt(sqrt(a))); }
+ //------------------------------------------------------------------------------------------------------------------------------
+ AF2 APrxPQToGamma2(AF2 a) { return Quart(a); }
+ AF2 APrxPQToLinear(AF2 a) { return Oct(a); }
+ AF2 APrxLoGamma2ToPQ(AF2 a) { return AF2_AU2((AU2_AF2(a) >> AU2_(2)) + AU2_(0x2F9A4E46)); }
+ AF2 APrxMedGamma2ToPQ(AF2 a) { AF2 b = AF2_AU2((AU2_AF2(a) >> AU2_(2)) + AU2_(0x2F9A4E46)); AF2 b4 = Quart(b); return b - b * (b4 - a) / (AF1_(4.0) * b4); }
+ AF2 APrxHighGamma2ToPQ(AF2 a) { return sqrt(sqrt(a)); }
+ AF2 APrxLoLinearToPQ(AF2 a) { return AF2_AU2((AU2_AF2(a) >> AU2_(3)) + AU2_(0x378D8723)); }
+ AF2 APrxMedLinearToPQ(AF2 a) { AF2 b = AF2_AU2((AU2_AF2(a) >> AU2_(3)) + AU2_(0x378D8723)); AF2 b8 = Oct(b); return b - b * (b8 - a) / (AF1_(8.0) * b8); }
+ AF2 APrxHighLinearToPQ(AF2 a) { return sqrt(sqrt(sqrt(a))); }
+ //------------------------------------------------------------------------------------------------------------------------------
+ AF3 APrxPQToGamma2(AF3 a) { return Quart(a); }
+ AF3 APrxPQToLinear(AF3 a) { return Oct(a); }
+ AF3 APrxLoGamma2ToPQ(AF3 a) { return AF3_AU3((AU3_AF3(a) >> AU3_(2)) + AU3_(0x2F9A4E46)); }
+ AF3 APrxMedGamma2ToPQ(AF3 a) { AF3 b = AF3_AU3((AU3_AF3(a) >> AU3_(2)) + AU3_(0x2F9A4E46)); AF3 b4 = Quart(b); return b - b * (b4 - a) / (AF1_(4.0) * b4); }
+ AF3 APrxHighGamma2ToPQ(AF3 a) { return sqrt(sqrt(a)); }
+ AF3 APrxLoLinearToPQ(AF3 a) { return AF3_AU3((AU3_AF3(a) >> AU3_(3)) + AU3_(0x378D8723)); }
+ AF3 APrxMedLinearToPQ(AF3 a) { AF3 b = AF3_AU3((AU3_AF3(a) >> AU3_(3)) + AU3_(0x378D8723)); AF3 b8 = Oct(b); return b - b * (b8 - a) / (AF1_(8.0) * b8); }
+ AF3 APrxHighLinearToPQ(AF3 a) { return sqrt(sqrt(sqrt(a))); }
+ //------------------------------------------------------------------------------------------------------------------------------
+ AF4 APrxPQToGamma2(AF4 a) { return Quart(a); }
+ AF4 APrxPQToLinear(AF4 a) { return Oct(a); }
+ AF4 APrxLoGamma2ToPQ(AF4 a) { return AF4_AU4((AU4_AF4(a) >> AU4_(2)) + AU4_(0x2F9A4E46)); }
+ AF4 APrxMedGamma2ToPQ(AF4 a) { AF4 b = AF4_AU4((AU4_AF4(a) >> AU4_(2)) + AU4_(0x2F9A4E46)); AF4 b4 = Quart(b); return b - b * (b4 - a) / (AF1_(4.0) * b4); }
+ AF4 APrxHighGamma2ToPQ(AF4 a) { return sqrt(sqrt(a)); }
+ AF4 APrxLoLinearToPQ(AF4 a) { return AF4_AU4((AU4_AF4(a) >> AU4_(3)) + AU4_(0x378D8723)); }
+ AF4 APrxMedLinearToPQ(AF4 a) { AF4 b = AF4_AU4((AU4_AF4(a) >> AU4_(3)) + AU4_(0x378D8723)); AF4 b8 = Oct(b); return b - b * (b8 - a) / (AF1_(8.0) * b8); }
+ AF4 APrxHighLinearToPQ(AF4 a) { return sqrt(sqrt(sqrt(a))); }
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+//_____________________________________________________________/\_______________________________________________________________
+//==============================================================================================================================
+//                                                    PARABOLIC SIN & COS
+//------------------------------------------------------------------------------------------------------------------------------
+// Approximate answers to transcendental questions.
+//------------------------------------------------------------------------------------------------------------------------------
+//==============================================================================================================================
+ #if 1
+  // Valid input range is {-1 to 1} representing {0 to 2 pi}.
+  // Output range is {-1/4 to 1/4} representing {-1 to 1}.
+  AF1 APSinF1(AF1 x){return x*abs(x)-x;} // MAD.
+  AF2 APSinF2(AF2 x){return x*abs(x)-x;}
+  AF1 APCosF1(AF1 x){x=AFractF1(x*AF1_(0.5)+AF1_(0.75));x=x*AF1_(2.0)-AF1_(1.0);return APSinF1(x);} // 3x MAD, FRACT
+  AF2 APCosF2(AF2 x){x=AFractF2(x*AF2_(0.5)+AF2_(0.75));x=x*AF2_(2.0)-AF2_(1.0);return APSinF2(x);}
+  AF2 APSinCosF1(AF1 x){AF1 y=AFractF1(x*AF1_(0.5)+AF1_(0.75));y=y*AF1_(2.0)-AF1_(1.0);return APSinF2(AF2(x,y));}
+ #endif
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+//_____________________________________________________________/\_______________________________________________________________
+//==============================================================================================================================
+//                                                     [ZOL] ZERO ONE LOGIC
+//------------------------------------------------------------------------------------------------------------------------------
+// Conditional free logic designed for easy 16-bit packing, and backwards porting to 32-bit.
+//------------------------------------------------------------------------------------------------------------------------------
+// 0 := false
+// 1 := true
+//------------------------------------------------------------------------------------------------------------------------------
+// AndNot(x,y)   -> !(x&y) .... One op.
+// AndOr(x,y,z)  -> (x&y)|z ... One op.
+// GtZero(x)     -> x>0.0 ..... One op.
+// Sel(x,y,z)    -> x?y:z ..... Two ops, has no precision loss.
+// Signed(x)     -> x<0.0 ..... One op.
+// ZeroPass(x,y) -> x?0:y ..... Two ops, 'y' is a pass through safe for aliasing as integer.
+//------------------------------------------------------------------------------------------------------------------------------
+// OPTIMIZATION NOTES
+// ==================
+// - On Vega to use 2 constants in a packed op, pass in as one AW2 or one AH2 'k.xy' and use as 'k.xx' and 'k.yy'.
+//   For example 'a.xy*k.xx+k.yy'.
+//==============================================================================================================================
+ #if 1
+  AU1 AZolAndU1(AU1 x,AU1 y){return min(x,y);}
+  AU2 AZolAndU2(AU2 x,AU2 y){return min(x,y);}
+  AU3 AZolAndU3(AU3 x,AU3 y){return min(x,y);}
+  AU4 AZolAndU4(AU4 x,AU4 y){return min(x,y);}
+//------------------------------------------------------------------------------------------------------------------------------
+  AU1 AZolNotU1(AU1 x){return x^AU1_(1);}
+  AU2 AZolNotU2(AU2 x){return x^AU2_(1);}
+  AU3 AZolNotU3(AU3 x){return x^AU3_(1);}
+  AU4 AZolNotU4(AU4 x){return x^AU4_(1);}
+//------------------------------------------------------------------------------------------------------------------------------
+  AU1 AZolOrU1(AU1 x,AU1 y){return max(x,y);}
+  AU2 AZolOrU2(AU2 x,AU2 y){return max(x,y);}
+  AU3 AZolOrU3(AU3 x,AU3 y){return max(x,y);}
+  AU4 AZolOrU4(AU4 x,AU4 y){return max(x,y);}
+//==============================================================================================================================
+  AU1 AZolF1ToU1(AF1 x){return AU1(x);}
+  AU2 AZolF2ToU2(AF2 x){return AU2(x);}
+  AU3 AZolF3ToU3(AF3 x){return AU3(x);}
+  AU4 AZolF4ToU4(AF4 x){return AU4(x);}
+//------------------------------------------------------------------------------------------------------------------------------
+  // 2 ops, denormals don't work in 32-bit on PC (and if they are enabled, OMOD is disabled).
+  AU1 AZolNotF1ToU1(AF1 x){return AU1(AF1_(1.0)-x);}
+  AU2 AZolNotF2ToU2(AF2 x){return AU2(AF2_(1.0)-x);}
+  AU3 AZolNotF3ToU3(AF3 x){return AU3(AF3_(1.0)-x);}
+  AU4 AZolNotF4ToU4(AF4 x){return AU4(AF4_(1.0)-x);}
+//------------------------------------------------------------------------------------------------------------------------------
+  AF1 AZolU1ToF1(AU1 x){return AF1(x);}
+  AF2 AZolU2ToF2(AU2 x){return AF2(x);}
+  AF3 AZolU3ToF3(AU3 x){return AF3(x);}
+  AF4 AZolU4ToF4(AU4 x){return AF4(x);}
+//==============================================================================================================================
+  AF1 AZolAndF1(AF1 x,AF1 y){return min(x,y);}
+  AF2 AZolAndF2(AF2 x,AF2 y){return min(x,y);}
+  AF3 AZolAndF3(AF3 x,AF3 y){return min(x,y);}
+  AF4 AZolAndF4(AF4 x,AF4 y){return min(x,y);}
+//------------------------------------------------------------------------------------------------------------------------------
+  AF1 ASolAndNotF1(AF1 x,AF1 y){return (-x)*y+AF1_(1.0);}
+  AF2 ASolAndNotF2(AF2 x,AF2 y){return (-x)*y+AF2_(1.0);}
+  AF3 ASolAndNotF3(AF3 x,AF3 y){return (-x)*y+AF3_(1.0);}
+  AF4 ASolAndNotF4(AF4 x,AF4 y){return (-x)*y+AF4_(1.0);}
+//------------------------------------------------------------------------------------------------------------------------------
+  AF1 AZolAndOrF1(AF1 x,AF1 y,AF1 z){return ASatF1(x*y+z);}
+  AF2 AZolAndOrF2(AF2 x,AF2 y,AF2 z){return ASatF2(x*y+z);}
+  AF3 AZolAndOrF3(AF3 x,AF3 y,AF3 z){return ASatF3(x*y+z);}
+  AF4 AZolAndOrF4(AF4 x,AF4 y,AF4 z){return ASatF4(x*y+z);}
+//------------------------------------------------------------------------------------------------------------------------------
+  AF1 AZolGtZeroF1(AF1 x){return ASatF1(x*AF1_(A_INFP_F));}
+  AF2 AZolGtZeroF2(AF2 x){return ASatF2(x*AF2_(A_INFP_F));}
+  AF3 AZolGtZeroF3(AF3 x){return ASatF3(x*AF3_(A_INFP_F));}
+  AF4 AZolGtZeroF4(AF4 x){return ASatF4(x*AF4_(A_INFP_F));}
+//------------------------------------------------------------------------------------------------------------------------------
+  AF1 AZolNotF1(AF1 x){return AF1_(1.0)-x;}
+  AF2 AZolNotF2(AF2 x){return AF2_(1.0)-x;}
+  AF3 AZolNotF3(AF3 x){return AF3_(1.0)-x;}
+  AF4 AZolNotF4(AF4 x){return AF4_(1.0)-x;}
+//------------------------------------------------------------------------------------------------------------------------------
+  AF1 AZolOrF1(AF1 x,AF1 y){return max(x,y);}
+  AF2 AZolOrF2(AF2 x,AF2 y){return max(x,y);}
+  AF3 AZolOrF3(AF3 x,AF3 y){return max(x,y);}
+  AF4 AZolOrF4(AF4 x,AF4 y){return max(x,y);}
+//------------------------------------------------------------------------------------------------------------------------------
+  AF1 AZolSelF1(AF1 x,AF1 y,AF1 z){AF1 r=(-x)*z+z;return x*y+r;}
+  AF2 AZolSelF2(AF2 x,AF2 y,AF2 z){AF2 r=(-x)*z+z;return x*y+r;}
+  AF3 AZolSelF3(AF3 x,AF3 y,AF3 z){AF3 r=(-x)*z+z;return x*y+r;}
+  AF4 AZolSelF4(AF4 x,AF4 y,AF4 z){AF4 r=(-x)*z+z;return x*y+r;}
+//------------------------------------------------------------------------------------------------------------------------------
+  AF1 AZolSignedF1(AF1 x){return ASatF1(x*AF1_(A_INFN_F));}
+  AF2 AZolSignedF2(AF2 x){return ASatF2(x*AF2_(A_INFN_F));}
+  AF3 AZolSignedF3(AF3 x){return ASatF3(x*AF3_(A_INFN_F));}
+  AF4 AZolSignedF4(AF4 x){return ASatF4(x*AF4_(A_INFN_F));}
+//------------------------------------------------------------------------------------------------------------------------------
+  AF1 AZolZeroPassF1(AF1 x,AF1 y){return AF1_AU1((AU1_AF1(x)!=AU1_(0))?AU1_(0):AU1_AF1(y));}
+  AF2 AZolZeroPassF2(AF2 x,AF2 y){return AF2_AU2((AU2_AF2(x)!=AU2_(0))?AU2_(0):AU2_AF2(y));}
+  AF3 AZolZeroPassF3(AF3 x,AF3 y){return AF3_AU3((AU3_AF3(x)!=AU3_(0))?AU3_(0):AU3_AF3(y));}
+  AF4 AZolZeroPassF4(AF4 x,AF4 y){return AF4_AU4((AU4_AF4(x)!=AU4_(0))?AU4_(0):AU4_AF4(y));}
+ #endif
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+//_____________________________________________________________/\_______________________________________________________________
+//==============================================================================================================================
+//                                                      COLOR CONVERSIONS
+//------------------------------------------------------------------------------------------------------------------------------
+// These are all linear to/from some other space (where 'linear' has been shortened out of the function name).
+// So 'ToGamma' is 'LinearToGamma', and 'FromGamma' is 'LinearFromGamma'.
+// These are branch free implementations.
+// The AToSrgbF1() function is useful for stores for compute shaders for GPUs without hardware linear->sRGB store conversion.
+//------------------------------------------------------------------------------------------------------------------------------
+// TRANSFER FUNCTIONS
+// ==================
+// 709 ..... Rec709 used for some HDTVs
+// Gamma ... Typically 2.2 for some PC displays, or 2.4-2.5 for CRTs, or 2.2 FreeSync2 native
+// Pq ...... PQ native for HDR10
+// Srgb .... The sRGB output, typical of PC displays, useful for 10-bit output, or storing to 8-bit UNORM without SRGB type
+// Two ..... Gamma 2.0, fastest conversion (useful for intermediate pass approximations)
+// Three ... Gamma 3.0, less fast, but good for HDR.
+//------------------------------------------------------------------------------------------------------------------------------
+// KEEPING TO SPEC
+// ===============
+// Both Rec.709 and sRGB have a linear segment which as spec'ed would intersect the curved segment 2 times.
+//  (a.) For 8-bit sRGB, steps {0 to 10.3} are in the linear region (4% of the encoding range).
+//  (b.) For 8-bit  709, steps {0 to 20.7} are in the linear region (8% of the encoding range).
+// Also there is a slight step in the transition regions.
+// Precision of the coefficients in the spec being the likely cause.
+// Main usage case of the sRGB code is to do the linear->sRGB converstion in a compute shader before store.
+// This is to work around lack of hardware (typically only ROP does the conversion for free).
+// To "correct" the linear segment, would be to introduce error, because hardware decode of sRGB->linear is fixed (and free).
+// So this header keeps with the spec.
+// For linear->sRGB transforms, the linear segment in some respects reduces error, because rounding in that region is linear.
+// Rounding in the curved region in hardware (and fast software code) introduces error due to rounding in non-linear.
+//------------------------------------------------------------------------------------------------------------------------------
+// FOR PQ
+// ======
+// Both input and output is {0.0-1.0}, and where output 1.0 represents 10000.0 cd/m^2.
+// All constants are only specified to FP32 precision.
+// External PQ source reference,
+//  - https://github.com/ampas/aces-dev/blob/master/transforms/ctl/utilities/ACESlib.Utilities_Color.a1.0.1.ctl
+//------------------------------------------------------------------------------------------------------------------------------
+// PACKED VERSIONS
+// ===============
+// These are the A*H2() functions.
+// There is no PQ functions as FP16 seemed to not have enough precision for the conversion.
+// The remaining functions are "good enough" for 8-bit, and maybe 10-bit if not concerned about a few 1-bit errors.
+// Precision is lowest in the 709 conversion, higher in sRGB, higher still in Two and Gamma (when using 2.2 at least).
+//------------------------------------------------------------------------------------------------------------------------------
+// NOTES
+// =====
+// Could be faster for PQ conversions to be in ALU or a texture lookup depending on usage case.
+//==============================================================================================================================
+ #if 1
+  AF1 ATo709F1(AF1 c){AF3 j=AF3(0.018*4.5,4.5,0.45);AF2 k=AF2(1.099,-0.099);
+   return clamp(j.x  ,c*j.y  ,pow(c,j.z  )*k.x  +k.y  );}
+  AF2 ATo709F2(AF2 c){AF3 j=AF3(0.018*4.5,4.5,0.45);AF2 k=AF2(1.099,-0.099);
+   return clamp(j.xx ,c*j.yy ,pow(c,j.zz )*k.xx +k.yy );}
+  AF3 ATo709F3(AF3 c){AF3 j=AF3(0.018*4.5,4.5,0.45);AF2 k=AF2(1.099,-0.099);
+   return clamp(j.xxx,c*j.yyy,pow(c,j.zzz)*k.xxx+k.yyy);}
+//------------------------------------------------------------------------------------------------------------------------------
+  // Note 'rcpX' is '1/x', where the 'x' is what would be used in AFromGamma().
+  AF1 AToGammaF1(AF1 c,AF1 rcpX){return pow(c,AF1_(rcpX));}
+  AF2 AToGammaF2(AF2 c,AF1 rcpX){return pow(c,AF2_(rcpX));}
+  AF3 AToGammaF3(AF3 c,AF1 rcpX){return pow(c,AF3_(rcpX));}
+//------------------------------------------------------------------------------------------------------------------------------
+  AF1 AToPqF1(AF1 x){AF1 p=pow(x,AF1_(0.159302));
+   return pow((AF1_(0.835938)+AF1_(18.8516)*p)/(AF1_(1.0)+AF1_(18.6875)*p),AF1_(78.8438));}
+  AF2 AToPqF1(AF2 x){AF2 p=pow(x,AF2_(0.159302));
+   return pow((AF2_(0.835938)+AF2_(18.8516)*p)/(AF2_(1.0)+AF2_(18.6875)*p),AF2_(78.8438));}
+  AF3 AToPqF1(AF3 x){AF3 p=pow(x,AF3_(0.159302));
+   return pow((AF3_(0.835938)+AF3_(18.8516)*p)/(AF3_(1.0)+AF3_(18.6875)*p),AF3_(78.8438));}
+//------------------------------------------------------------------------------------------------------------------------------
+  AF1 AToSrgbF1(AF1 c){AF3 j=AF3(0.0031308*12.92,12.92,1.0/2.4);AF2 k=AF2(1.055,-0.055);
+   return clamp(j.x  ,c*j.y  ,pow(c,j.z  )*k.x  +k.y  );}
+  AF2 AToSrgbF2(AF2 c){AF3 j=AF3(0.0031308*12.92,12.92,1.0/2.4);AF2 k=AF2(1.055,-0.055);
+   return clamp(j.xx ,c*j.yy ,pow(c,j.zz )*k.xx +k.yy );}
+  AF3 AToSrgbF3(AF3 c){AF3 j=AF3(0.0031308*12.92,12.92,1.0/2.4);AF2 k=AF2(1.055,-0.055);
+   return clamp(j.xxx,c*j.yyy,pow(c,j.zzz)*k.xxx+k.yyy);}
+//------------------------------------------------------------------------------------------------------------------------------
+  AF1 AToTwoF1(AF1 c){return sqrt(c);}
+  AF2 AToTwoF2(AF2 c){return sqrt(c);}
+  AF3 AToTwoF3(AF3 c){return sqrt(c);}
+//------------------------------------------------------------------------------------------------------------------------------
+  AF1 AToThreeF1(AF1 c){return pow(c,AF1_(1.0/3.0));}
+  AF2 AToThreeF2(AF2 c){return pow(c,AF2_(1.0/3.0));}
+  AF3 AToThreeF3(AF3 c){return pow(c,AF3_(1.0/3.0));}
+ #endif
+//==============================================================================================================================
+ #if 1
+  // Unfortunately median won't work here.
+  AF1 AFrom709F1(AF1 c){AF3 j=AF3(0.081/4.5,1.0/4.5,1.0/0.45);AF2 k=AF2(1.0/1.099,0.099/1.099);
+   return AZolSelF1(AZolSignedF1(c-j.x  ),c*j.y  ,pow(c*k.x  +k.y  ,j.z  ));}
+  AF2 AFrom709F2(AF2 c){AF3 j=AF3(0.081/4.5,1.0/4.5,1.0/0.45);AF2 k=AF2(1.0/1.099,0.099/1.099);
+   return AZolSelF2(AZolSignedF2(c-j.xx ),c*j.yy ,pow(c*k.xx +k.yy ,j.zz ));}
+  AF3 AFrom709F3(AF3 c){AF3 j=AF3(0.081/4.5,1.0/4.5,1.0/0.45);AF2 k=AF2(1.0/1.099,0.099/1.099);
+   return AZolSelF3(AZolSignedF3(c-j.xxx),c*j.yyy,pow(c*k.xxx+k.yyy,j.zzz));}
+//------------------------------------------------------------------------------------------------------------------------------
+  AF1 AFromGammaF1(AF1 c,AF1 x){return pow(c,AF1_(x));}
+  AF2 AFromGammaF2(AF2 c,AF1 x){return pow(c,AF2_(x));}
+  AF3 AFromGammaF3(AF3 c,AF1 x){return pow(c,AF3_(x));}
+//------------------------------------------------------------------------------------------------------------------------------
+  AF1 AFromPqF1(AF1 x){AF1 p=pow(x,AF1_(0.0126833));
+   return pow(ASatF1(p-AF1_(0.835938))/(AF1_(18.8516)-AF1_(18.6875)*p),AF1_(6.27739));}
+  AF2 AFromPqF1(AF2 x){AF2 p=pow(x,AF2_(0.0126833));
+   return pow(ASatF2(p-AF2_(0.835938))/(AF2_(18.8516)-AF2_(18.6875)*p),AF2_(6.27739));}
+  AF3 AFromPqF1(AF3 x){AF3 p=pow(x,AF3_(0.0126833));
+   return pow(ASatF3(p-AF3_(0.835938))/(AF3_(18.8516)-AF3_(18.6875)*p),AF3_(6.27739));}
+//------------------------------------------------------------------------------------------------------------------------------
+  // Unfortunately median won't work here.
+  AF1 AFromSrgbF1(AF1 c){AF3 j=AF3(0.04045/12.92,1.0/12.92,2.4);AF2 k=AF2(1.0/1.055,0.055/1.055);
+   return AZolSelF1(AZolSignedF1(c-j.x  ),c*j.y  ,pow(c*k.x  +k.y  ,j.z  ));}
+  AF2 AFromSrgbF2(AF2 c){AF3 j=AF3(0.04045/12.92,1.0/12.92,2.4);AF2 k=AF2(1.0/1.055,0.055/1.055);
+   return AZolSelF2(AZolSignedF2(c-j.xx ),c*j.yy ,pow(c*k.xx +k.yy ,j.zz ));}
+  AF3 AFromSrgbF3(AF3 c){AF3 j=AF3(0.04045/12.92,1.0/12.92,2.4);AF2 k=AF2(1.0/1.055,0.055/1.055);
+   return AZolSelF3(AZolSignedF3(c-j.xxx),c*j.yyy,pow(c*k.xxx+k.yyy,j.zzz));}
+//------------------------------------------------------------------------------------------------------------------------------
+  AF1 AFromTwoF1(AF1 c){return c*c;}
+  AF2 AFromTwoF2(AF2 c){return c*c;}
+  AF3 AFromTwoF3(AF3 c){return c*c;}
+//------------------------------------------------------------------------------------------------------------------------------
+  AF1 AFromThreeF1(AF1 c){return c*c*c;}
+  AF2 AFromThreeF2(AF2 c){return c*c*c;}
+  AF3 AFromThreeF3(AF3 c){return c*c*c;}
+ #endif
+//==============================================================================================================================
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+//_____________________________________________________________/\_______________________________________________________________
+//==============================================================================================================================
+//                                                          CS REMAP
+//==============================================================================================================================
+ // Simple remap 64x1 to 8x8 with rotated 2x2 pixel quads in quad linear.
+ //  543210
+ //  ======
+ //  ..xxx.
+ //  yy...y
+ AU2 ARmp8x8(AU1 a){return AU2(ABfe(a,1u,3u),ABfiM(ABfe(a,3u,3u),a,1u));}
+//==============================================================================================================================
+ // More complex remap 64x1 to 8x8 which is necessary for 2D wave reductions.
+ //  543210
+ //  ======
+ //  .xx..x
+ //  y..yy.
+ // Details,
+ //  LANE TO 8x8 MAPPING
+ //  ===================
+ //  00 01 08 09 10 11 18 19
+ //  02 03 0a 0b 12 13 1a 1b
+ //  04 05 0c 0d 14 15 1c 1d
+ //  06 07 0e 0f 16 17 1e 1f
+ //  20 21 28 29 30 31 38 39
+ //  22 23 2a 2b 32 33 3a 3b
+ //  24 25 2c 2d 34 35 3c 3d
+ //  26 27 2e 2f 36 37 3e 3f
+ AU2 ARmpRed8x8(AU1 a){return AU2(ABfiM(ABfe(a,2u,3u),a,1u),ABfiM(ABfe(a,3u,3u),ABfe(a,1u,2u),2u));}
+//==============================================================================================================================
+#endif
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+//_____________________________________________________________/\_______________________________________________________________
+//==============================================================================================================================
+//
+//                                                          REFERENCE
+//
+//------------------------------------------------------------------------------------------------------------------------------
+// IEEE FLOAT RULES
+// ================
+//  - saturate(NaN)=0, saturate(-INF)=0, saturate(+INF)=1
+//  - {+/-}0 * {+/-}INF = NaN
+//  - -INF + (+INF) = NaN
+//  - {+/-}0 / {+/-}0 = NaN
+//  - {+/-}INF / {+/-}INF = NaN
+//  - a<(-0) := sqrt(a) = NaN (a=-0.0 won't NaN)
+//  - 0 == -0
+//  - 4/0 = +INF
+//  - 4/-0 = -INF
+//  - 4+INF = +INF
+//  - 4-INF = -INF
+//  - 4*(+INF) = +INF
+//  - 4*(-INF) = -INF
+//  - -4*(+INF) = -INF
+//  - sqrt(+INF) = +INF
+//------------------------------------------------------------------------------------------------------------------------------
+// FP16 ENCODING
+// =============
+// fedcba9876543210
+// ----------------
+// ......mmmmmmmmmm  10-bit mantissa (encodes 11-bit 0.5 to 1.0 except for denormals)
+// .eeeee..........  5-bit exponent
+// .00000..........  denormals
+// .00001..........  -14 exponent
+// .11110..........   15 exponent
+// .111110000000000  infinity
+// .11111nnnnnnnnnn  NaN with n!=0
+// s...............  sign
+//------------------------------------------------------------------------------------------------------------------------------
+// FP16/INT16 ALIASING DENORMAL
+// ============================
+// 11-bit unsigned integers alias with half float denormal/normal values,
+//     1 = 2^(-24) = 1/16777216 ....................... first denormal value
+//     2 = 2^(-23)
+//   ...
+//  1023 = 2^(-14)*(1-2^(-10)) = 2^(-14)*(1-1/1024) ... last denormal value
+//  1024 = 2^(-14) = 1/16384 .......................... first normal value that still maps to integers
+//  2047 .............................................. last normal value that still maps to integers
+// Scaling limits,
+//  2^15 = 32768 ...................................... largest power of 2 scaling
+// Largest pow2 conversion mapping is at *32768,
+//     1 : 2^(-9) = 1/512
+//     2 : 1/256
+//     4 : 1/128
+//     8 : 1/64
+//    16 : 1/32
+//    32 : 1/16
+//    64 : 1/8
+//   128 : 1/4
+//   256 : 1/2
+//   512 : 1
+//  1024 : 2
+//  2047 : a little less than 4
+//==============================================================================================================================
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+//_____________________________________________________________/\_______________________________________________________________
+//==============================================================================================================================
+//
+//
+//                                                     GPU/CPU PORTABILITY
+//
+//
+//------------------------------------------------------------------------------------------------------------------------------
+// This is the GPU implementation.
+// See the CPU implementation for docs.
+//==============================================================================================================================
+#ifdef A_GPU
+ #define A_TRUE true
+ #define A_FALSE false
+ #define A_STATIC
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+//_____________________________________________________________/\_______________________________________________________________
+//==============================================================================================================================
+//                                     VECTOR ARGUMENT/RETURN/INITIALIZATION PORTABILITY
+//==============================================================================================================================
+ #define retAD2 AD2
+ #define retAD3 AD3
+ #define retAD4 AD4
+ #define retAF2 AF2
+ #define retAF3 AF3
+ #define retAF4 AF4
+ #define retAL2 AL2
+ #define retAL3 AL3
+ #define retAL4 AL4
+ #define retAU2 AU2
+ #define retAU3 AU3
+ #define retAU4 AU4
+//------------------------------------------------------------------------------------------------------------------------------
+ #define inAD2 in AD2
+ #define inAD3 in AD3
+ #define inAD4 in AD4
+ #define inAF2 in AF2
+ #define inAF3 in AF3
+ #define inAF4 in AF4
+ #define inAL2 in AL2
+ #define inAL3 in AL3
+ #define inAL4 in AL4
+ #define inAU2 in AU2
+ #define inAU3 in AU3
+ #define inAU4 in AU4
+//------------------------------------------------------------------------------------------------------------------------------
+ #define inoutAD2 inout AD2
+ #define inoutAD3 inout AD3
+ #define inoutAD4 inout AD4
+ #define inoutAF2 inout AF2
+ #define inoutAF3 inout AF3
+ #define inoutAF4 inout AF4
+ #define inoutAL2 inout AL2
+ #define inoutAL3 inout AL3
+ #define inoutAL4 inout AL4
+ #define inoutAU2 inout AU2
+ #define inoutAU3 inout AU3
+ #define inoutAU4 inout AU4
+//------------------------------------------------------------------------------------------------------------------------------
+ #define outAD2 out AD2
+ #define outAD3 out AD3
+ #define outAD4 out AD4
+ #define outAF2 out AF2
+ #define outAF3 out AF3
+ #define outAF4 out AF4
+ #define outAL2 out AL2
+ #define outAL3 out AL3
+ #define outAL4 out AL4
+ #define outAU2 out AU2
+ #define outAU3 out AU3
+ #define outAU4 out AU4
+//------------------------------------------------------------------------------------------------------------------------------
+ #define varAD2(x) AD2 x
+ #define varAD3(x) AD3 x
+ #define varAD4(x) AD4 x
+ #define varAF2(x) AF2 x
+ #define varAF3(x) AF3 x
+ #define varAF4(x) AF4 x
+ #define varAL2(x) AL2 x
+ #define varAL3(x) AL3 x
+ #define varAL4(x) AL4 x
+ #define varAU2(x) AU2 x
+ #define varAU3(x) AU3 x
+ #define varAU4(x) AU4 x
+//------------------------------------------------------------------------------------------------------------------------------
+ #define initAD2(x,y) AD2(x,y)
+ #define initAD3(x,y,z) AD3(x,y,z)
+ #define initAD4(x,y,z,w) AD4(x,y,z,w)
+ #define initAF2(x,y) AF2(x,y)
+ #define initAF3(x,y,z) AF3(x,y,z)
+ #define initAF4(x,y,z,w) AF4(x,y,z,w)
+ #define initAL2(x,y) AL2(x,y)
+ #define initAL3(x,y,z) AL3(x,y,z)
+ #define initAL4(x,y,z,w) AL4(x,y,z,w)
+ #define initAU2(x,y) AU2(x,y)
+ #define initAU3(x,y,z) AU3(x,y,z)
+ #define initAU4(x,y,z,w) AU4(x,y,z,w)
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+//_____________________________________________________________/\_______________________________________________________________
+//==============================================================================================================================
+//                                                     SCALAR RETURN OPS
+//==============================================================================================================================
+ #define AAbsD1(a) abs(AD1(a))
+ #define AAbsF1(a) abs(AF1(a))
+//------------------------------------------------------------------------------------------------------------------------------
+ #define ACosD1(a) cos(AD1(a))
+ #define ACosF1(a) cos(AF1(a))
+//------------------------------------------------------------------------------------------------------------------------------
+ #define ADotD2(a,b) dot(AD2(a),AD2(b))
+ #define ADotD3(a,b) dot(AD3(a),AD3(b))
+ #define ADotD4(a,b) dot(AD4(a),AD4(b))
+ #define ADotF2(a,b) dot(AF2(a),AF2(b))
+ #define ADotF3(a,b) dot(AF3(a),AF3(b))
+ #define ADotF4(a,b) dot(AF4(a),AF4(b))
+//------------------------------------------------------------------------------------------------------------------------------
+ #define AExp2D1(a) exp2(AD1(a))
+ #define AExp2F1(a) exp2(AF1(a))
+//------------------------------------------------------------------------------------------------------------------------------
+ #define AFloorD1(a) floor(AD1(a))
+ #define AFloorF1(a) floor(AF1(a))
+//------------------------------------------------------------------------------------------------------------------------------
+ #define ALog2D1(a) log2(AD1(a))
+ #define ALog2F1(a) log2(AF1(a))
+//------------------------------------------------------------------------------------------------------------------------------
+ #define AMaxD1(a,b) max(a,b)
+ #define AMaxF1(a,b) max(a,b)
+ #define AMaxL1(a,b) max(a,b)
+ #define AMaxU1(a,b) max(a,b)
+//------------------------------------------------------------------------------------------------------------------------------
+ #define AMinD1(a,b) min(a,b)
+ #define AMinF1(a,b) min(a,b)
+ #define AMinL1(a,b) min(a,b)
+ #define AMinU1(a,b) min(a,b)
+//------------------------------------------------------------------------------------------------------------------------------
+ #define ASinD1(a) sin(AD1(a))
+ #define ASinF1(a) sin(AF1(a))
+//------------------------------------------------------------------------------------------------------------------------------
+ #define ASqrtD1(a) sqrt(AD1(a))
+ #define ASqrtF1(a) sqrt(AF1(a))
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+//_____________________________________________________________/\_______________________________________________________________
+//==============================================================================================================================
+//                                               SCALAR RETURN OPS - DEPENDENT
+//==============================================================================================================================
+ #define APowD1(a,b) pow(AD1(a),AF1(b))
+ #define APowF1(a,b) pow(AF1(a),AF1(b))
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+//_____________________________________________________________/\_______________________________________________________________
+//==============================================================================================================================
+//                                                         VECTOR OPS
+//------------------------------------------------------------------------------------------------------------------------------
+// These are added as needed for production or prototyping, so not necessarily a complete set.
+// They follow a convention of taking in a destination and also returning the destination value to increase utility.
+//==============================================================================================================================
+ #ifdef A_DUBL
+  AD2 opAAbsD2(outAD2 d,inAD2 a){d=abs(a);return d;}
+  AD3 opAAbsD3(outAD3 d,inAD3 a){d=abs(a);return d;}
+  AD4 opAAbsD4(outAD4 d,inAD4 a){d=abs(a);return d;}
+//------------------------------------------------------------------------------------------------------------------------------
+  AD2 opAAddD2(outAD2 d,inAD2 a,inAD2 b){d=a+b;return d;}
+  AD3 opAAddD3(outAD3 d,inAD3 a,inAD3 b){d=a+b;return d;}
+  AD4 opAAddD4(outAD4 d,inAD4 a,inAD4 b){d=a+b;return d;}
+//------------------------------------------------------------------------------------------------------------------------------
+  AD2 opAAddOneD2(outAD2 d,inAD2 a,AD1 b){d=a+AD2_(b);return d;}
+  AD3 opAAddOneD3(outAD3 d,inAD3 a,AD1 b){d=a+AD3_(b);return d;}
+  AD4 opAAddOneD4(outAD4 d,inAD4 a,AD1 b){d=a+AD4_(b);return d;}
+//------------------------------------------------------------------------------------------------------------------------------
+  AD2 opACpyD2(outAD2 d,inAD2 a){d=a;return d;}
+  AD3 opACpyD3(outAD3 d,inAD3 a){d=a;return d;}
+  AD4 opACpyD4(outAD4 d,inAD4 a){d=a;return d;}
+//------------------------------------------------------------------------------------------------------------------------------
+  AD2 opALerpD2(outAD2 d,inAD2 a,inAD2 b,inAD2 c){d=ALerpD2(a,b,c);return d;}
+  AD3 opALerpD3(outAD3 d,inAD3 a,inAD3 b,inAD3 c){d=ALerpD3(a,b,c);return d;}
+  AD4 opALerpD4(outAD4 d,inAD4 a,inAD4 b,inAD4 c){d=ALerpD4(a,b,c);return d;}
+//------------------------------------------------------------------------------------------------------------------------------
+  AD2 opALerpOneD2(outAD2 d,inAD2 a,inAD2 b,AD1 c){d=ALerpD2(a,b,AD2_(c));return d;}
+  AD3 opALerpOneD3(outAD3 d,inAD3 a,inAD3 b,AD1 c){d=ALerpD3(a,b,AD3_(c));return d;}
+  AD4 opALerpOneD4(outAD4 d,inAD4 a,inAD4 b,AD1 c){d=ALerpD4(a,b,AD4_(c));return d;}
+//------------------------------------------------------------------------------------------------------------------------------
+  AD2 opAMaxD2(outAD2 d,inAD2 a,inAD2 b){d=max(a,b);return d;}
+  AD3 opAMaxD3(outAD3 d,inAD3 a,inAD3 b){d=max(a,b);return d;}
+  AD4 opAMaxD4(outAD4 d,inAD4 a,inAD4 b){d=max(a,b);return d;}
+//------------------------------------------------------------------------------------------------------------------------------
+  AD2 opAMinD2(outAD2 d,inAD2 a,inAD2 b){d=min(a,b);return d;}
+  AD3 opAMinD3(outAD3 d,inAD3 a,inAD3 b){d=min(a,b);return d;}
+  AD4 opAMinD4(outAD4 d,inAD4 a,inAD4 b){d=min(a,b);return d;}
+//------------------------------------------------------------------------------------------------------------------------------
+  AD2 opAMulD2(outAD2 d,inAD2 a,inAD2 b){d=a*b;return d;}
+  AD3 opAMulD3(outAD3 d,inAD3 a,inAD3 b){d=a*b;return d;}
+  AD4 opAMulD4(outAD4 d,inAD4 a,inAD4 b){d=a*b;return d;}
+//------------------------------------------------------------------------------------------------------------------------------
+  AD2 opAMulOneD2(outAD2 d,inAD2 a,AD1 b){d=a*AD2_(b);return d;}
+  AD3 opAMulOneD3(outAD3 d,inAD3 a,AD1 b){d=a*AD3_(b);return d;}
+  AD4 opAMulOneD4(outAD4 d,inAD4 a,AD1 b){d=a*AD4_(b);return d;}
+//------------------------------------------------------------------------------------------------------------------------------
+  AD2 opANegD2(outAD2 d,inAD2 a){d=-a;return d;}
+  AD3 opANegD3(outAD3 d,inAD3 a){d=-a;return d;}
+  AD4 opANegD4(outAD4 d,inAD4 a){d=-a;return d;}
+//------------------------------------------------------------------------------------------------------------------------------
+  AD2 opARcpD2(outAD2 d,inAD2 a){d=ARcpD2(a);return d;}
+  AD3 opARcpD3(outAD3 d,inAD3 a){d=ARcpD3(a);return d;}
+  AD4 opARcpD4(outAD4 d,inAD4 a){d=ARcpD4(a);return d;}
+ #endif
+//==============================================================================================================================
+ AF2 opAAbsF2(outAF2 d,inAF2 a){d=abs(a);return d;}
+ AF3 opAAbsF3(outAF3 d,inAF3 a){d=abs(a);return d;}
+ AF4 opAAbsF4(outAF4 d,inAF4 a){d=abs(a);return d;}
+//------------------------------------------------------------------------------------------------------------------------------
+ AF2 opAAddF2(outAF2 d,inAF2 a,inAF2 b){d=a+b;return d;}
+ AF3 opAAddF3(outAF3 d,inAF3 a,inAF3 b){d=a+b;return d;}
+ AF4 opAAddF4(outAF4 d,inAF4 a,inAF4 b){d=a+b;return d;}
+//------------------------------------------------------------------------------------------------------------------------------
+ AF2 opAAddOneF2(outAF2 d,inAF2 a,AF1 b){d=a+AF2_(b);return d;}
+ AF3 opAAddOneF3(outAF3 d,inAF3 a,AF1 b){d=a+AF3_(b);return d;}
+ AF4 opAAddOneF4(outAF4 d,inAF4 a,AF1 b){d=a+AF4_(b);return d;}
+//------------------------------------------------------------------------------------------------------------------------------
+ AF2 opACpyF2(outAF2 d,inAF2 a){d=a;return d;}
+ AF3 opACpyF3(outAF3 d,inAF3 a){d=a;return d;}
+ AF4 opACpyF4(outAF4 d,inAF4 a){d=a;return d;}
+//------------------------------------------------------------------------------------------------------------------------------
+ AF2 opALerpF2(outAF2 d,inAF2 a,inAF2 b,inAF2 c){d=ALerpF2(a,b,c);return d;}
+ AF3 opALerpF3(outAF3 d,inAF3 a,inAF3 b,inAF3 c){d=ALerpF3(a,b,c);return d;}
+ AF4 opALerpF4(outAF4 d,inAF4 a,inAF4 b,inAF4 c){d=ALerpF4(a,b,c);return d;}
+//------------------------------------------------------------------------------------------------------------------------------
+ AF2 opALerpOneF2(outAF2 d,inAF2 a,inAF2 b,AF1 c){d=ALerpF2(a,b,AF2_(c));return d;}
+ AF3 opALerpOneF3(outAF3 d,inAF3 a,inAF3 b,AF1 c){d=ALerpF3(a,b,AF3_(c));return d;}
+ AF4 opALerpOneF4(outAF4 d,inAF4 a,inAF4 b,AF1 c){d=ALerpF4(a,b,AF4_(c));return d;}
+//------------------------------------------------------------------------------------------------------------------------------
+ AF2 opAMaxF2(outAF2 d,inAF2 a,inAF2 b){d=max(a,b);return d;}
+ AF3 opAMaxF3(outAF3 d,inAF3 a,inAF3 b){d=max(a,b);return d;}
+ AF4 opAMaxF4(outAF4 d,inAF4 a,inAF4 b){d=max(a,b);return d;}
+//------------------------------------------------------------------------------------------------------------------------------
+ AF2 opAMinF2(outAF2 d,inAF2 a,inAF2 b){d=min(a,b);return d;}
+ AF3 opAMinF3(outAF3 d,inAF3 a,inAF3 b){d=min(a,b);return d;}
+ AF4 opAMinF4(outAF4 d,inAF4 a,inAF4 b){d=min(a,b);return d;}
+//------------------------------------------------------------------------------------------------------------------------------
+ AF2 opAMulF2(outAF2 d,inAF2 a,inAF2 b){d=a*b;return d;}
+ AF3 opAMulF3(outAF3 d,inAF3 a,inAF3 b){d=a*b;return d;}
+ AF4 opAMulF4(outAF4 d,inAF4 a,inAF4 b){d=a*b;return d;}
+//------------------------------------------------------------------------------------------------------------------------------
+ AF2 opAMulOneF2(outAF2 d,inAF2 a,AF1 b){d=a*AF2_(b);return d;}
+ AF3 opAMulOneF3(outAF3 d,inAF3 a,AF1 b){d=a*AF3_(b);return d;}
+ AF4 opAMulOneF4(outAF4 d,inAF4 a,AF1 b){d=a*AF4_(b);return d;}
+//------------------------------------------------------------------------------------------------------------------------------
+ AF2 opANegF2(outAF2 d,inAF2 a){d=-a;return d;}
+ AF3 opANegF3(outAF3 d,inAF3 a){d=-a;return d;}
+ AF4 opANegF4(outAF4 d,inAF4 a){d=-a;return d;}
+//------------------------------------------------------------------------------------------------------------------------------
+ AF2 opARcpF2(outAF2 d,inAF2 a){d=ARcpF2(a);return d;}
+ AF3 opARcpF3(outAF3 d,inAF3 a){d=ARcpF3(a);return d;}
+ AF4 opARcpF4(outAF4 d,inAF4 a){d=ARcpF4(a);return d;}
+#endif
+
+#endif
+
+//_____________________________________________________________/\_______________________________________________________________
+//==============================================================================================================================
+//
+//                                 [CAS] FIDELITY FX - CONSTRAST ADAPTIVE SHARPENING 1.20190610
+//
+//==============================================================================================================================
+// LICENSE
+// =======
+// Copyright (c) 2017-2019 Advanced Micro Devices, Inc. All rights reserved.
+// -------
+// Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation
+// files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy,
+// modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the
+// Software is furnished to do so, subject to the following conditions:
+// -------
+// The above copyright notice and this permission notice shall be included in all copies or substantial portions of the
+// Software.
+// -------
+// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE
+// WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE AUTHORS OR
+// COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+// ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+//------------------------------------------------------------------------------------------------------------------------------
+#define CAS_AREA_LIMIT 4.0
+//------------------------------------------------------------------------------------------------------------------------------
+// Pass in output and input resolution in pixels.
+// This returns true if CAS supports scaling in the given configuration.
+AP1 CasSupportScaling(AF1 outX,AF1 outY,AF1 inX,AF1 inY){return ((outX*outY)*ARcpF1(inX*inY))<=CAS_AREA_LIMIT;}
+//==============================================================================================================================
+// Call to setup required constant values (works on CPU or GPU).
+#ifndef A_GPU
+A_STATIC void CasSetup(
+ outAU4 const0,
+ outAU4 const1,
+ AF1 sharpness, // 0 := default (lower ringing), 1 := maximum (higest ringing)
+ AF1 inputSizeInPixelsX,
+ AF1 inputSizeInPixelsY,
+ AF1 outputSizeInPixelsX,
+ AF1 outputSizeInPixelsY){
+  // Scaling terms.
+  const0[0]=AU1_AF1(inputSizeInPixelsX*ARcpF1(outputSizeInPixelsX));
+  const0[1]=AU1_AF1(inputSizeInPixelsY*ARcpF1(outputSizeInPixelsY));
+  const0[2]=AU1_AF1(AF1_(0.5)*inputSizeInPixelsX*ARcpF1(outputSizeInPixelsX)-AF1_(0.5));
+  const0[3]=AU1_AF1(AF1_(0.5)*inputSizeInPixelsY*ARcpF1(outputSizeInPixelsY)-AF1_(0.5));
+  // Sharpness value.
+  AF1 sharp=-ARcpF1(ALerpF1(8.0,5.0,ASatF1(sharpness)));
+  varAF2(hSharp)=initAF2(sharp,0.0);
+  const1[0]=AU1_AF1(sharp);
+  const1[1]=AU1_AH2_AF2(hSharp);
+  const1[2]=AU1_AF1(AF1_(8.0)*inputSizeInPixelsX*ARcpF1(outputSizeInPixelsX));
+  const1[3]=0;}
+#endif
+
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+//_____________________________________________________________/\_______________________________________________________________
+//==============================================================================================================================
+//                                                     NON-PACKED VERSION
+//==============================================================================================================================
+#ifdef A_GPU
+ AF3 CasLoad(ASU2 p) { return texelFetch(diffuseRect, p, 0).rgb; }
+ void CasInput(inout AF1 r,inout AF1 g,inout AF1 b)
+ {
+ }
+
+//------------------------------------------------------------------------------------------------------------------------------
+ void CasFilter(
+ out AF1 pixR, // Output values, non-vector so port between CasFilter() and CasFilterH() is easy.
+ out AF1 pixG,
+ out AF1 pixB,
+ AU2 ip, // Integer pixel position in output.
+ AU4 const0, // Constants generated by CasSetup().
+ AU4 const1,
+ AP1 noScaling){ // Must be a compile-time literal value, true = sharpen only (no resize).
+//------------------------------------------------------------------------------------------------------------------------------
+  // Debug a checker pattern of on/off tiles for visual inspection.
+  #ifdef CAS_DEBUG_CHECKER
+   if((((ip.x^ip.y)>>8u)&1u)==0u){AF3 pix0=CasLoad(ASU2(ip));
+    pixR=pix0.r;pixG=pix0.g;pixB=pix0.b;CasInput(pixR,pixG,pixB);return;}
+  #endif
+//------------------------------------------------------------------------------------------------------------------------------
+  // No scaling algorithm uses minimal 3x3 pixel neighborhood.
+  if(noScaling){
+   // a b c
+   // d e f
+   // g h i
+   ASU2 sp=ASU2(ip);
+   AF3 a=CasLoad(sp+ASU2(-1,-1));
+   AF3 b=CasLoad(sp+ASU2( 0,-1));
+   AF3 c=CasLoad(sp+ASU2( 1,-1));
+   AF3 d=CasLoad(sp+ASU2(-1, 0));
+   AF3 e=CasLoad(sp);
+   AF3 f=CasLoad(sp+ASU2( 1, 0));
+   AF3 g=CasLoad(sp+ASU2(-1, 1));
+   AF3 h=CasLoad(sp+ASU2( 0, 1));
+   AF3 i=CasLoad(sp+ASU2( 1, 1));
+   // Run optional input transform.
+   CasInput(a.r,a.g,a.b);
+   CasInput(b.r,b.g,b.b);
+   CasInput(c.r,c.g,c.b);
+   CasInput(d.r,d.g,d.b);
+   CasInput(e.r,e.g,e.b);
+   CasInput(f.r,f.g,f.b);
+   CasInput(g.r,g.g,g.b);
+   CasInput(h.r,h.g,h.b);
+   CasInput(i.r,i.g,i.b);
+   // Soft min and max.
+   //  a b c             b
+   //  d e f * 0.5  +  d e f * 0.5
+   //  g h i             h
+   // These are 2.0x bigger (factored out the extra multiply).
+   AF1 mnR=AMin3F1(AMin3F1(d.r,e.r,f.r),b.r,h.r);
+   AF1 mnG=AMin3F1(AMin3F1(d.g,e.g,f.g),b.g,h.g);
+   AF1 mnB=AMin3F1(AMin3F1(d.b,e.b,f.b),b.b,h.b);
+   #ifdef CAS_BETTER_DIAGONALS
+    AF1 mnR2=AMin3F1(AMin3F1(mnR,a.r,c.r),g.r,i.r);
+    AF1 mnG2=AMin3F1(AMin3F1(mnG,a.g,c.g),g.g,i.g);
+    AF1 mnB2=AMin3F1(AMin3F1(mnB,a.b,c.b),g.b,i.b);
+    mnR=mnR+mnR2;
+    mnG=mnG+mnG2;
+    mnB=mnB+mnB2;
+   #endif
+   AF1 mxR=AMax3F1(AMax3F1(d.r,e.r,f.r),b.r,h.r);
+   AF1 mxG=AMax3F1(AMax3F1(d.g,e.g,f.g),b.g,h.g);
+   AF1 mxB=AMax3F1(AMax3F1(d.b,e.b,f.b),b.b,h.b);
+   #ifdef CAS_BETTER_DIAGONALS
+    AF1 mxR2=AMax3F1(AMax3F1(mxR,a.r,c.r),g.r,i.r);
+    AF1 mxG2=AMax3F1(AMax3F1(mxG,a.g,c.g),g.g,i.g);
+    AF1 mxB2=AMax3F1(AMax3F1(mxB,a.b,c.b),g.b,i.b);
+    mxR=mxR+mxR2;
+    mxG=mxG+mxG2;
+    mxB=mxB+mxB2;
+   #endif
+   // Smooth minimum distance to signal limit divided by smooth max.
+   #ifdef CAS_GO_SLOWER
+    AF1 rcpMR=ARcpF1(mxR);
+    AF1 rcpMG=ARcpF1(mxG);
+    AF1 rcpMB=ARcpF1(mxB);
+   #else
+    AF1 rcpMR=APrxLoRcpF1(mxR);
+    AF1 rcpMG=APrxLoRcpF1(mxG);
+    AF1 rcpMB=APrxLoRcpF1(mxB);
+   #endif
+   #ifdef CAS_BETTER_DIAGONALS
+    AF1 ampR=ASatF1(min(mnR,AF1_(2.0)-mxR)*rcpMR);
+    AF1 ampG=ASatF1(min(mnG,AF1_(2.0)-mxG)*rcpMG);
+    AF1 ampB=ASatF1(min(mnB,AF1_(2.0)-mxB)*rcpMB);
+   #else
+    AF1 ampR=ASatF1(min(mnR,AF1_(1.0)-mxR)*rcpMR);
+    AF1 ampG=ASatF1(min(mnG,AF1_(1.0)-mxG)*rcpMG);
+    AF1 ampB=ASatF1(min(mnB,AF1_(1.0)-mxB)*rcpMB);
+   #endif
+   // Shaping amount of sharpening.
+   #ifdef CAS_GO_SLOWER
+    ampR=sqrt(ampR);
+    ampG=sqrt(ampG);
+    ampB=sqrt(ampB);
+   #else
+    ampR=APrxLoSqrtF1(ampR);
+    ampG=APrxLoSqrtF1(ampG);
+    ampB=APrxLoSqrtF1(ampB);
+   #endif
+   // Filter shape.
+   //  0 w 0
+   //  w 1 w
+   //  0 w 0
+   AF1 peak=AF1_AU1(const1.x);
+   AF1 wR=ampR*peak;
+   AF1 wG=ampG*peak;
+   AF1 wB=ampB*peak;
+   // Filter.
+   #ifndef CAS_SLOW
+    // Using green coef only, depending on dead code removal to strip out the extra overhead.
+    #ifdef CAS_GO_SLOWER
+     AF1 rcpWeight=ARcpF1(AF1_(1.0)+AF1_(4.0)*wG);
+    #else
+     AF1 rcpWeight=APrxMedRcpF1(AF1_(1.0)+AF1_(4.0)*wG);
+    #endif
+    pixR=ASatF1((b.r*wG+d.r*wG+f.r*wG+h.r*wG+e.r)*rcpWeight);
+    pixG=ASatF1((b.g*wG+d.g*wG+f.g*wG+h.g*wG+e.g)*rcpWeight);
+    pixB=ASatF1((b.b*wG+d.b*wG+f.b*wG+h.b*wG+e.b)*rcpWeight);
+   #else
+    #ifdef CAS_GO_SLOWER
+     AF1 rcpWeightR=ARcpF1(AF1_(1.0)+AF1_(4.0)*wR);
+     AF1 rcpWeightG=ARcpF1(AF1_(1.0)+AF1_(4.0)*wG);
+     AF1 rcpWeightB=ARcpF1(AF1_(1.0)+AF1_(4.0)*wB);
+    #else
+     AF1 rcpWeightR=APrxMedRcpF1(AF1_(1.0)+AF1_(4.0)*wR);
+     AF1 rcpWeightG=APrxMedRcpF1(AF1_(1.0)+AF1_(4.0)*wG);
+     AF1 rcpWeightB=APrxMedRcpF1(AF1_(1.0)+AF1_(4.0)*wB);
+    #endif
+    pixR=ASatF1((b.r*wR+d.r*wR+f.r*wR+h.r*wR+e.r)*rcpWeightR);
+    pixG=ASatF1((b.g*wG+d.g*wG+f.g*wG+h.g*wG+e.g)*rcpWeightG);
+    pixB=ASatF1((b.b*wB+d.b*wB+f.b*wB+h.b*wB+e.b)*rcpWeightB);
+   #endif
+   return;}
+//------------------------------------------------------------------------------------------------------------------------------
+  // Scaling algorithm adaptively interpolates between nearest 4 results of the non-scaling algorithm.
+  //  a b c d
+  //  e f g h
+  //  i j k l
+  //  m n o p
+  // Working these 4 results.
+  //  +-----+-----+
+  //  |     |     |
+  //  |  f..|..g  |
+  //  |  .  |  .  |
+  //  +-----+-----+
+  //  |  .  |  .  |
+  //  |  j..|..k  |
+  //  |     |     |
+  //  +-----+-----+
+  AF2 pp=AF2(ip)*AF2_AU2(const0.xy)+AF2_AU2(const0.zw);
+  AF2 fp=floor(pp);
+  pp-=fp;
+  ASU2 sp=ASU2(fp);
+  AF3 a=CasLoad(sp+ASU2(-1,-1));
+  AF3 b=CasLoad(sp+ASU2( 0,-1));
+  AF3 e=CasLoad(sp+ASU2(-1, 0));
+  AF3 f=CasLoad(sp);
+  AF3 c=CasLoad(sp+ASU2( 1,-1));
+  AF3 d=CasLoad(sp+ASU2( 2,-1));
+  AF3 g=CasLoad(sp+ASU2( 1, 0));
+  AF3 h=CasLoad(sp+ASU2( 2, 0));
+  AF3 i=CasLoad(sp+ASU2(-1, 1));
+  AF3 j=CasLoad(sp+ASU2( 0, 1));
+  AF3 m=CasLoad(sp+ASU2(-1, 2));
+  AF3 n=CasLoad(sp+ASU2( 0, 2));
+  AF3 k=CasLoad(sp+ASU2( 1, 1));
+  AF3 l=CasLoad(sp+ASU2( 2, 1));
+  AF3 o=CasLoad(sp+ASU2( 1, 2));
+  AF3 p=CasLoad(sp+ASU2( 2, 2));
+  // Run optional input transform.
+  CasInput(a.r,a.g,a.b);
+  CasInput(b.r,b.g,b.b);
+  CasInput(c.r,c.g,c.b);
+  CasInput(d.r,d.g,d.b);
+  CasInput(e.r,e.g,e.b);
+  CasInput(f.r,f.g,f.b);
+  CasInput(g.r,g.g,g.b);
+  CasInput(h.r,h.g,h.b);
+  CasInput(i.r,i.g,i.b);
+  CasInput(j.r,j.g,j.b);
+  CasInput(k.r,k.g,k.b);
+  CasInput(l.r,l.g,l.b);
+  CasInput(m.r,m.g,m.b);
+  CasInput(n.r,n.g,n.b);
+  CasInput(o.r,o.g,o.b);
+  CasInput(p.r,p.g,p.b);
+  // Soft min and max.
+  // These are 2.0x bigger (factored out the extra multiply).
+  //  a b c             b
+  //  e f g * 0.5  +  e f g * 0.5  [F]
+  //  i j k             j
+  AF1 mnfR=AMin3F1(AMin3F1(b.r,e.r,f.r),g.r,j.r);
+  AF1 mnfG=AMin3F1(AMin3F1(b.g,e.g,f.g),g.g,j.g);
+  AF1 mnfB=AMin3F1(AMin3F1(b.b,e.b,f.b),g.b,j.b);
+  #ifdef CAS_BETTER_DIAGONALS
+   AF1 mnfR2=AMin3F1(AMin3F1(mnfR,a.r,c.r),i.r,k.r);
+   AF1 mnfG2=AMin3F1(AMin3F1(mnfG,a.g,c.g),i.g,k.g);
+   AF1 mnfB2=AMin3F1(AMin3F1(mnfB,a.b,c.b),i.b,k.b);
+   mnfR=mnfR+mnfR2;
+   mnfG=mnfG+mnfG2;
+   mnfB=mnfB+mnfB2;
+  #endif
+  AF1 mxfR=AMax3F1(AMax3F1(b.r,e.r,f.r),g.r,j.r);
+  AF1 mxfG=AMax3F1(AMax3F1(b.g,e.g,f.g),g.g,j.g);
+  AF1 mxfB=AMax3F1(AMax3F1(b.b,e.b,f.b),g.b,j.b);
+  #ifdef CAS_BETTER_DIAGONALS
+   AF1 mxfR2=AMax3F1(AMax3F1(mxfR,a.r,c.r),i.r,k.r);
+   AF1 mxfG2=AMax3F1(AMax3F1(mxfG,a.g,c.g),i.g,k.g);
+   AF1 mxfB2=AMax3F1(AMax3F1(mxfB,a.b,c.b),i.b,k.b);
+   mxfR=mxfR+mxfR2;
+   mxfG=mxfG+mxfG2;
+   mxfB=mxfB+mxfB2;
+  #endif
+  //  b c d             c
+  //  f g h * 0.5  +  f g h * 0.5  [G]
+  //  j k l             k
+  AF1 mngR=AMin3F1(AMin3F1(c.r,f.r,g.r),h.r,k.r);
+  AF1 mngG=AMin3F1(AMin3F1(c.g,f.g,g.g),h.g,k.g);
+  AF1 mngB=AMin3F1(AMin3F1(c.b,f.b,g.b),h.b,k.b);
+  #ifdef CAS_BETTER_DIAGONALS
+   AF1 mngR2=AMin3F1(AMin3F1(mngR,b.r,d.r),j.r,l.r);
+   AF1 mngG2=AMin3F1(AMin3F1(mngG,b.g,d.g),j.g,l.g);
+   AF1 mngB2=AMin3F1(AMin3F1(mngB,b.b,d.b),j.b,l.b);
+   mngR=mngR+mngR2;
+   mngG=mngG+mngG2;
+   mngB=mngB+mngB2;
+  #endif
+  AF1 mxgR=AMax3F1(AMax3F1(c.r,f.r,g.r),h.r,k.r);
+  AF1 mxgG=AMax3F1(AMax3F1(c.g,f.g,g.g),h.g,k.g);
+  AF1 mxgB=AMax3F1(AMax3F1(c.b,f.b,g.b),h.b,k.b);
+  #ifdef CAS_BETTER_DIAGONALS
+   AF1 mxgR2=AMax3F1(AMax3F1(mxgR,b.r,d.r),j.r,l.r);
+   AF1 mxgG2=AMax3F1(AMax3F1(mxgG,b.g,d.g),j.g,l.g);
+   AF1 mxgB2=AMax3F1(AMax3F1(mxgB,b.b,d.b),j.b,l.b);
+   mxgR=mxgR+mxgR2;
+   mxgG=mxgG+mxgG2;
+   mxgB=mxgB+mxgB2;
+  #endif
+  //  e f g             f
+  //  i j k * 0.5  +  i j k * 0.5  [J]
+  //  m n o             n
+  AF1 mnjR=AMin3F1(AMin3F1(f.r,i.r,j.r),k.r,n.r);
+  AF1 mnjG=AMin3F1(AMin3F1(f.g,i.g,j.g),k.g,n.g);
+  AF1 mnjB=AMin3F1(AMin3F1(f.b,i.b,j.b),k.b,n.b);
+  #ifdef CAS_BETTER_DIAGONALS
+   AF1 mnjR2=AMin3F1(AMin3F1(mnjR,e.r,g.r),m.r,o.r);
+   AF1 mnjG2=AMin3F1(AMin3F1(mnjG,e.g,g.g),m.g,o.g);
+   AF1 mnjB2=AMin3F1(AMin3F1(mnjB,e.b,g.b),m.b,o.b);
+   mnjR=mnjR+mnjR2;
+   mnjG=mnjG+mnjG2;
+   mnjB=mnjB+mnjB2;
+  #endif
+  AF1 mxjR=AMax3F1(AMax3F1(f.r,i.r,j.r),k.r,n.r);
+  AF1 mxjG=AMax3F1(AMax3F1(f.g,i.g,j.g),k.g,n.g);
+  AF1 mxjB=AMax3F1(AMax3F1(f.b,i.b,j.b),k.b,n.b);
+  #ifdef CAS_BETTER_DIAGONALS
+   AF1 mxjR2=AMax3F1(AMax3F1(mxjR,e.r,g.r),m.r,o.r);
+   AF1 mxjG2=AMax3F1(AMax3F1(mxjG,e.g,g.g),m.g,o.g);
+   AF1 mxjB2=AMax3F1(AMax3F1(mxjB,e.b,g.b),m.b,o.b);
+   mxjR=mxjR+mxjR2;
+   mxjG=mxjG+mxjG2;
+   mxjB=mxjB+mxjB2;
+  #endif
+  //  f g h             g
+  //  j k l * 0.5  +  j k l * 0.5  [K]
+  //  n o p             o
+  AF1 mnkR=AMin3F1(AMin3F1(g.r,j.r,k.r),l.r,o.r);
+  AF1 mnkG=AMin3F1(AMin3F1(g.g,j.g,k.g),l.g,o.g);
+  AF1 mnkB=AMin3F1(AMin3F1(g.b,j.b,k.b),l.b,o.b);
+  #ifdef CAS_BETTER_DIAGONALS
+   AF1 mnkR2=AMin3F1(AMin3F1(mnkR,f.r,h.r),n.r,p.r);
+   AF1 mnkG2=AMin3F1(AMin3F1(mnkG,f.g,h.g),n.g,p.g);
+   AF1 mnkB2=AMin3F1(AMin3F1(mnkB,f.b,h.b),n.b,p.b);
+   mnkR=mnkR+mnkR2;
+   mnkG=mnkG+mnkG2;
+   mnkB=mnkB+mnkB2;
+  #endif
+  AF1 mxkR=AMax3F1(AMax3F1(g.r,j.r,k.r),l.r,o.r);
+  AF1 mxkG=AMax3F1(AMax3F1(g.g,j.g,k.g),l.g,o.g);
+  AF1 mxkB=AMax3F1(AMax3F1(g.b,j.b,k.b),l.b,o.b);
+  #ifdef CAS_BETTER_DIAGONALS
+   AF1 mxkR2=AMax3F1(AMax3F1(mxkR,f.r,h.r),n.r,p.r);
+   AF1 mxkG2=AMax3F1(AMax3F1(mxkG,f.g,h.g),n.g,p.g);
+   AF1 mxkB2=AMax3F1(AMax3F1(mxkB,f.b,h.b),n.b,p.b);
+   mxkR=mxkR+mxkR2;
+   mxkG=mxkG+mxkG2;
+   mxkB=mxkB+mxkB2;
+  #endif
+  // Smooth minimum distance to signal limit divided by smooth max.
+  #ifdef CAS_GO_SLOWER
+   AF1 rcpMfR=ARcpF1(mxfR);
+   AF1 rcpMfG=ARcpF1(mxfG);
+   AF1 rcpMfB=ARcpF1(mxfB);
+   AF1 rcpMgR=ARcpF1(mxgR);
+   AF1 rcpMgG=ARcpF1(mxgG);
+   AF1 rcpMgB=ARcpF1(mxgB);
+   AF1 rcpMjR=ARcpF1(mxjR);
+   AF1 rcpMjG=ARcpF1(mxjG);
+   AF1 rcpMjB=ARcpF1(mxjB);
+   AF1 rcpMkR=ARcpF1(mxkR);
+   AF1 rcpMkG=ARcpF1(mxkG);
+   AF1 rcpMkB=ARcpF1(mxkB);
+  #else
+   AF1 rcpMfR=APrxLoRcpF1(mxfR);
+   AF1 rcpMfG=APrxLoRcpF1(mxfG);
+   AF1 rcpMfB=APrxLoRcpF1(mxfB);
+   AF1 rcpMgR=APrxLoRcpF1(mxgR);
+   AF1 rcpMgG=APrxLoRcpF1(mxgG);
+   AF1 rcpMgB=APrxLoRcpF1(mxgB);
+   AF1 rcpMjR=APrxLoRcpF1(mxjR);
+   AF1 rcpMjG=APrxLoRcpF1(mxjG);
+   AF1 rcpMjB=APrxLoRcpF1(mxjB);
+   AF1 rcpMkR=APrxLoRcpF1(mxkR);
+   AF1 rcpMkG=APrxLoRcpF1(mxkG);
+   AF1 rcpMkB=APrxLoRcpF1(mxkB);
+  #endif
+  #ifdef CAS_BETTER_DIAGONALS
+   AF1 ampfR=ASatF1(min(mnfR,AF1_(2.0)-mxfR)*rcpMfR);
+   AF1 ampfG=ASatF1(min(mnfG,AF1_(2.0)-mxfG)*rcpMfG);
+   AF1 ampfB=ASatF1(min(mnfB,AF1_(2.0)-mxfB)*rcpMfB);
+   AF1 ampgR=ASatF1(min(mngR,AF1_(2.0)-mxgR)*rcpMgR);
+   AF1 ampgG=ASatF1(min(mngG,AF1_(2.0)-mxgG)*rcpMgG);
+   AF1 ampgB=ASatF1(min(mngB,AF1_(2.0)-mxgB)*rcpMgB);
+   AF1 ampjR=ASatF1(min(mnjR,AF1_(2.0)-mxjR)*rcpMjR);
+   AF1 ampjG=ASatF1(min(mnjG,AF1_(2.0)-mxjG)*rcpMjG);
+   AF1 ampjB=ASatF1(min(mnjB,AF1_(2.0)-mxjB)*rcpMjB);
+   AF1 ampkR=ASatF1(min(mnkR,AF1_(2.0)-mxkR)*rcpMkR);
+   AF1 ampkG=ASatF1(min(mnkG,AF1_(2.0)-mxkG)*rcpMkG);
+   AF1 ampkB=ASatF1(min(mnkB,AF1_(2.0)-mxkB)*rcpMkB);
+  #else
+   AF1 ampfR=ASatF1(min(mnfR,AF1_(1.0)-mxfR)*rcpMfR);
+   AF1 ampfG=ASatF1(min(mnfG,AF1_(1.0)-mxfG)*rcpMfG);
+   AF1 ampfB=ASatF1(min(mnfB,AF1_(1.0)-mxfB)*rcpMfB);
+   AF1 ampgR=ASatF1(min(mngR,AF1_(1.0)-mxgR)*rcpMgR);
+   AF1 ampgG=ASatF1(min(mngG,AF1_(1.0)-mxgG)*rcpMgG);
+   AF1 ampgB=ASatF1(min(mngB,AF1_(1.0)-mxgB)*rcpMgB);
+   AF1 ampjR=ASatF1(min(mnjR,AF1_(1.0)-mxjR)*rcpMjR);
+   AF1 ampjG=ASatF1(min(mnjG,AF1_(1.0)-mxjG)*rcpMjG);
+   AF1 ampjB=ASatF1(min(mnjB,AF1_(1.0)-mxjB)*rcpMjB);
+   AF1 ampkR=ASatF1(min(mnkR,AF1_(1.0)-mxkR)*rcpMkR);
+   AF1 ampkG=ASatF1(min(mnkG,AF1_(1.0)-mxkG)*rcpMkG);
+   AF1 ampkB=ASatF1(min(mnkB,AF1_(1.0)-mxkB)*rcpMkB);
+  #endif
+  // Shaping amount of sharpening.
+  #ifdef CAS_GO_SLOWER
+   ampfR=sqrt(ampfR);
+   ampfG=sqrt(ampfG);
+   ampfB=sqrt(ampfB);
+   ampgR=sqrt(ampgR);
+   ampgG=sqrt(ampgG);
+   ampgB=sqrt(ampgB);
+   ampjR=sqrt(ampjR);
+   ampjG=sqrt(ampjG);
+   ampjB=sqrt(ampjB);
+   ampkR=sqrt(ampkR);
+   ampkG=sqrt(ampkG);
+   ampkB=sqrt(ampkB);
+  #else
+   ampfR=APrxLoSqrtF1(ampfR);
+   ampfG=APrxLoSqrtF1(ampfG);
+   ampfB=APrxLoSqrtF1(ampfB);
+   ampgR=APrxLoSqrtF1(ampgR);
+   ampgG=APrxLoSqrtF1(ampgG);
+   ampgB=APrxLoSqrtF1(ampgB);
+   ampjR=APrxLoSqrtF1(ampjR);
+   ampjG=APrxLoSqrtF1(ampjG);
+   ampjB=APrxLoSqrtF1(ampjB);
+   ampkR=APrxLoSqrtF1(ampkR);
+   ampkG=APrxLoSqrtF1(ampkG);
+   ampkB=APrxLoSqrtF1(ampkB);
+  #endif
+  // Filter shape.
+  //  0 w 0
+  //  w 1 w
+  //  0 w 0
+  AF1 peak=AF1_AU1(const1.x);
+  AF1 wfR=ampfR*peak;
+  AF1 wfG=ampfG*peak;
+  AF1 wfB=ampfB*peak;
+  AF1 wgR=ampgR*peak;
+  AF1 wgG=ampgG*peak;
+  AF1 wgB=ampgB*peak;
+  AF1 wjR=ampjR*peak;
+  AF1 wjG=ampjG*peak;
+  AF1 wjB=ampjB*peak;
+  AF1 wkR=ampkR*peak;
+  AF1 wkG=ampkG*peak;
+  AF1 wkB=ampkB*peak;
+  // Blend between 4 results.
+  //  s t
+  //  u v
+  AF1 s=(AF1_(1.0)-pp.x)*(AF1_(1.0)-pp.y);
+  AF1 t=           pp.x *(AF1_(1.0)-pp.y);
+  AF1 u=(AF1_(1.0)-pp.x)*           pp.y ;
+  AF1 v=           pp.x *           pp.y ;
+  // Thin edges to hide bilinear interpolation (helps diagonals).
+  AF1 thinB=1.0/32.0;
+  #ifdef CAS_GO_SLOWER
+   s*=ARcpF1(thinB+(mxfG-mnfG));
+   t*=ARcpF1(thinB+(mxgG-mngG));
+   u*=ARcpF1(thinB+(mxjG-mnjG));
+   v*=ARcpF1(thinB+(mxkG-mnkG));
+  #else
+   s*=APrxLoRcpF1(thinB+(mxfG-mnfG));
+   t*=APrxLoRcpF1(thinB+(mxgG-mngG));
+   u*=APrxLoRcpF1(thinB+(mxjG-mnjG));
+   v*=APrxLoRcpF1(thinB+(mxkG-mnkG));
+  #endif
+  // Final weighting.
+  //    b c
+  //  e f g h
+  //  i j k l
+  //    n o
+  //  _____  _____  _____  _____
+  //         fs        gt
+  //
+  //  _____  _____  _____  _____
+  //  fs      s gt  fs  t     gt
+  //         ju        kv
+  //  _____  _____  _____  _____
+  //         fs        gt
+  //  ju      u kv  ju  v     kv
+  //  _____  _____  _____  _____
+  //
+  //         ju        kv
+  AF1 qbeR=wfR*s;
+  AF1 qbeG=wfG*s;
+  AF1 qbeB=wfB*s;
+  AF1 qchR=wgR*t;
+  AF1 qchG=wgG*t;
+  AF1 qchB=wgB*t;
+  AF1 qfR=wgR*t+wjR*u+s;
+  AF1 qfG=wgG*t+wjG*u+s;
+  AF1 qfB=wgB*t+wjB*u+s;
+  AF1 qgR=wfR*s+wkR*v+t;
+  AF1 qgG=wfG*s+wkG*v+t;
+  AF1 qgB=wfB*s+wkB*v+t;
+  AF1 qjR=wfR*s+wkR*v+u;
+  AF1 qjG=wfG*s+wkG*v+u;
+  AF1 qjB=wfB*s+wkB*v+u;
+  AF1 qkR=wgR*t+wjR*u+v;
+  AF1 qkG=wgG*t+wjG*u+v;
+  AF1 qkB=wgB*t+wjB*u+v;
+  AF1 qinR=wjR*u;
+  AF1 qinG=wjG*u;
+  AF1 qinB=wjB*u;
+  AF1 qloR=wkR*v;
+  AF1 qloG=wkG*v;
+  AF1 qloB=wkB*v;
+  // Filter.
+  #ifndef CAS_SLOW
+   // Using green coef only, depending on dead code removal to strip out the extra overhead.
+   #ifdef CAS_GO_SLOWER
+    AF1 rcpWG=ARcpF1(AF1_(2.0)*qbeG+AF1_(2.0)*qchG+AF1_(2.0)*qinG+AF1_(2.0)*qloG+qfG+qgG+qjG+qkG);
+   #else
+    AF1 rcpWG=APrxMedRcpF1(AF1_(2.0)*qbeG+AF1_(2.0)*qchG+AF1_(2.0)*qinG+AF1_(2.0)*qloG+qfG+qgG+qjG+qkG);
+   #endif
+   pixR=ASatF1((b.r*qbeG+e.r*qbeG+c.r*qchG+h.r*qchG+i.r*qinG+n.r*qinG+l.r*qloG+o.r*qloG+f.r*qfG+g.r*qgG+j.r*qjG+k.r*qkG)*rcpWG);
+   pixG=ASatF1((b.g*qbeG+e.g*qbeG+c.g*qchG+h.g*qchG+i.g*qinG+n.g*qinG+l.g*qloG+o.g*qloG+f.g*qfG+g.g*qgG+j.g*qjG+k.g*qkG)*rcpWG);
+   pixB=ASatF1((b.b*qbeG+e.b*qbeG+c.b*qchG+h.b*qchG+i.b*qinG+n.b*qinG+l.b*qloG+o.b*qloG+f.b*qfG+g.b*qgG+j.b*qjG+k.b*qkG)*rcpWG);
+  #else
+   #ifdef CAS_GO_SLOWER
+    AF1 rcpWR=ARcpF1(AF1_(2.0)*qbeR+AF1_(2.0)*qchR+AF1_(2.0)*qinR+AF1_(2.0)*qloR+qfR+qgR+qjR+qkR);
+    AF1 rcpWG=ARcpF1(AF1_(2.0)*qbeG+AF1_(2.0)*qchG+AF1_(2.0)*qinG+AF1_(2.0)*qloG+qfG+qgG+qjG+qkG);
+    AF1 rcpWB=ARcpF1(AF1_(2.0)*qbeB+AF1_(2.0)*qchB+AF1_(2.0)*qinB+AF1_(2.0)*qloB+qfB+qgB+qjB+qkB);
+   #else
+    AF1 rcpWR=APrxMedRcpF1(AF1_(2.0)*qbeR+AF1_(2.0)*qchR+AF1_(2.0)*qinR+AF1_(2.0)*qloR+qfR+qgR+qjR+qkR);
+    AF1 rcpWG=APrxMedRcpF1(AF1_(2.0)*qbeG+AF1_(2.0)*qchG+AF1_(2.0)*qinG+AF1_(2.0)*qloG+qfG+qgG+qjG+qkG);
+    AF1 rcpWB=APrxMedRcpF1(AF1_(2.0)*qbeB+AF1_(2.0)*qchB+AF1_(2.0)*qinB+AF1_(2.0)*qloB+qfB+qgB+qjB+qkB);
+   #endif
+   pixR=ASatF1((b.r*qbeR+e.r*qbeR+c.r*qchR+h.r*qchR+i.r*qinR+n.r*qinR+l.r*qloR+o.r*qloR+f.r*qfR+g.r*qgR+j.r*qjR+k.r*qkR)*rcpWR);
+   pixG=ASatF1((b.g*qbeG+e.g*qbeG+c.g*qchG+h.g*qchG+i.g*qinG+n.g*qinG+l.g*qloG+o.g*qloG+f.g*qfG+g.g*qgG+j.g*qjG+k.g*qkG)*rcpWG);
+   pixB=ASatF1((b.b*qbeB+e.b*qbeB+c.b*qchB+h.b*qchB+i.b*qinB+n.b*qinB+l.b*qloB+o.b*qloB+f.b*qfB+g.b*qgB+j.b*qjB+k.b*qkB)*rcpWB);
+  #endif
+ }
+#endif
+
+#ifdef A_GPU
+void main()
+{
+    vec4 diff = vec4(0.f);
+    uvec2 point = uvec2(vary_fragcoord * out_screen_res.xy);
+    CasFilter(diff.r, diff.g, diff.b, point, cas_param_0, cas_param_1, true);
+    diff.a = texture(diffuseRect, vary_fragcoord).a;
+    frag_color = diff;
+}
+#endif
diff --git a/indra/newview/app_settings/shaders/class1/deferred/SMAA.glsl b/indra/newview/app_settings/shaders/class1/deferred/SMAA.glsl
new file mode 100644
index 0000000000..fdb77cce6e
--- /dev/null
+++ b/indra/newview/app_settings/shaders/class1/deferred/SMAA.glsl
@@ -0,0 +1,1463 @@
+/**
+ * @file SMAA.glsl
+ *
+ * $LicenseInfo:firstyear=2024&license=viewerlgpl$
+ * Second Life Viewer Source Code
+ * Copyright (C) 2024, Linden Research, Inc.
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation;
+ * version 2.1 of the License only.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
+ *
+ * Linden Research, Inc., 945 Battery Street, San Francisco, CA  94111  USA
+ * $/LicenseInfo$
+ */
+
+#extension GL_ARB_texture_rectangle : enable
+#extension GL_ARB_shader_texture_lod : enable
+#extension GL_EXT_gpu_shader4 : enable
+
+/*[EXTRA_CODE_HERE]*/
+
+#ifdef VERTEX_SHADER
+    #define SMAA_INCLUDE_VS 1
+    #define SMAA_INCLUDE_PS 0
+#else
+    #define SMAA_INCLUDE_VS 0
+    #define SMAA_INCLUDE_PS 1
+#endif
+
+uniform vec4 SMAA_RT_METRICS;
+
+/**
+ * Copyright (C) 2013 Jorge Jimenez (jorge@iryoku.com)
+ * Copyright (C) 2013 Jose I. Echevarria (joseignacioechevarria@gmail.com)
+ * Copyright (C) 2013 Belen Masia (bmasia@unizar.es)
+ * Copyright (C) 2013 Fernando Navarro (fernandn@microsoft.com)
+ * Copyright (C) 2013 Diego Gutierrez (diegog@unizar.es)
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * this software and associated documentation files (the "Software"), to deal in
+ * the Software without restriction, including without limitation the rights to
+ * use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies
+ * of the Software, and to permit persons to whom the Software is furnished to
+ * do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software. As clarification, there
+ * is no requirement that the copyright notice and permission be included in
+ * binary distributions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+
+/**
+ *                  _______  ___  ___       ___           ___
+ *                 /       ||   \/   |     /   \         /   \
+ *                |   (---- |  \  /  |    /  ^  \       /  ^  \
+ *                 \   \    |  |\/|  |   /  /_\  \     /  /_\  \
+ *              ----)   |   |  |  |  |  /  _____  \   /  _____  \
+ *             |_______/    |__|  |__| /__/     \__\ /__/     \__\
+ *
+ *                               E N H A N C E D
+ *       S U B P I X E L   M O R P H O L O G I C A L   A N T I A L I A S I N G
+ *
+ *                         http://www.iryoku.com/smaa/
+ *
+ * Hi, welcome aboard!
+ *
+ * Here you'll find instructions to get the shader up and running as fast as
+ * possible.
+ *
+ * IMPORTANTE NOTICE: when updating, remember to update both this file and the
+ * precomputed textures! They may change from version to version.
+ *
+ * The shader has three passes, chained together as follows:
+ *
+ *                           |input|------------------�
+ *                              v                     |
+ *                    [ SMAA*EdgeDetection ]          |
+ *                              v                     |
+ *                          |edgesTex|                |
+ *                              v                     |
+ *              [ SMAABlendingWeightCalculation ]     |
+ *                              v                     |
+ *                          |blendTex|                |
+ *                              v                     |
+ *                [ SMAANeighborhoodBlending ] <------�
+ *                              v
+ *                           |output|
+ *
+ * Note that each [pass] has its own vertex and pixel shader. Remember to use
+ * oversized triangles instead of quads to avoid overshading along the
+ * diagonal.
+ *
+ * You've three edge detection methods to choose from: luma, color or depth.
+ * They represent different quality/performance and anti-aliasing/sharpness
+ * tradeoffs, so our recommendation is for you to choose the one that best
+ * suits your particular scenario:
+ *
+ * - Depth edge detection is usually the fastest but it may miss some edges.
+ *
+ * - Luma edge detection is usually more expensive than depth edge detection,
+ *   but catches visible edges that depth edge detection can miss.
+ *
+ * - Color edge detection is usually the most expensive one but catches
+ *   chroma-only edges.
+ *
+ * For quickstarters: just use luma edge detection.
+ *
+ * The general advice is to not rush the integration process and ensure each
+ * step is done correctly (don't try to integrate SMAA T2x with predicated edge
+ * detection from the start!). Ok then, let's go!
+ *
+ *  1. The first step is to create two RGBA temporal render targets for holding
+ *     |edgesTex| and |blendTex|.
+ *
+ *     In DX10 or DX11, you can use a RG render target for the edges texture.
+ *     In the case of NVIDIA GPUs, using RG render targets seems to actually be
+ *     slower.
+ *
+ *     On the Xbox 360, you can use the same render target for resolving both
+ *     |edgesTex| and |blendTex|, as they aren't needed simultaneously.
+ *
+ *  2. Both temporal render targets |edgesTex| and |blendTex| must be cleared
+ *     each frame. Do not forget to clear the alpha channel!
+ *
+ *  3. The next step is loading the two supporting precalculated textures,
+ *     'areaTex' and 'searchTex'. You'll find them in the 'Textures' folder as
+ *     C++ headers, and also as regular DDS files. They'll be needed for the
+ *     'SMAABlendingWeightCalculation' pass.
+ *
+ *     If you use the C++ headers, be sure to load them in the format specified
+ *     inside of them.
+ *
+ *     You can also compress 'areaTex' and 'searchTex' using BC5 and BC4
+ *     respectively, if you have that option in your content processor pipeline.
+ *     When compressing then, you get a non-perceptible quality decrease, and a
+ *     marginal performance increase.
+ *
+ *  4. All samplers must be set to linear filtering and clamp.
+ *
+ *     After you get the technique working, remember that 64-bit inputs have
+ *     half-rate linear filtering on GCN.
+ *
+ *     If SMAA is applied to 64-bit color buffers, switching to point filtering
+ *     when accesing them will increase the performance. Search for
+ *     'SMAASamplePoint' to see which textures may benefit from point
+ *     filtering, and where (which is basically the color input in the edge
+ *     detection and resolve passes).
+ *
+ *  5. All texture reads and buffer writes must be non-sRGB, with the exception
+ *     of the input read and the output write in
+ *     'SMAANeighborhoodBlending' (and only in this pass!). If sRGB reads in
+ *     this last pass are not possible, the technique will work anyway, but
+ *     will perform antialiasing in gamma space.
+ *
+ *     IMPORTANT: for best results the input read for the color/luma edge
+ *     detection should *NOT* be sRGB.
+ *
+ *  6. Before including SMAA.h you'll have to setup the render target metrics,
+ *     the target and any optional configuration defines. Optionally you can
+ *     use a preset.
+ *
+ *     You have the following targets available:
+ *         SMAA_HLSL_3
+ *         SMAA_HLSL_4
+ *         SMAA_HLSL_4_1
+ *         SMAA_GLSL_2 *
+ *         SMAA_GLSL_3 *
+ *         SMAA_GLSL_4 *
+ *
+ *         * (See SMAA_INCLUDE_VS and SMAA_INCLUDE_PS below).
+ *
+ *     And four presets:
+ *         SMAA_PRESET_LOW          (%60 of the quality)
+ *         SMAA_PRESET_MEDIUM       (%80 of the quality)
+ *         SMAA_PRESET_HIGH         (%95 of the quality)
+ *         SMAA_PRESET_ULTRA        (%99 of the quality)
+ *
+ *     For example:
+ *         #define SMAA_RT_METRICS float4(1.0 / 1280.0, 1.0 / 720.0, 1280.0, 720.0)
+ *         #define SMAA_HLSL_4
+ *         #define SMAA_PRESET_HIGH
+ *         #include "SMAA.h"
+ *
+ *     Note that SMAA_RT_METRICS doesn't need to be a macro, it can be a
+ *     uniform variable. The code is designed to minimize the impact of not
+ *     using a constant value, but it is still better to hardcode it.
+ *
+ *     Depending on how you encoded 'areaTex' and 'searchTex', you may have to
+ *     add (and customize) the following defines before including SMAA.h:
+ *          #define SMAA_AREATEX_SELECT(sample) sample.rg
+ *          #define SMAA_SEARCHTEX_SELECT(sample) sample.r
+ *
+ *     If your engine is already using porting macros, you can define
+ *     SMAA_CUSTOM_SL, and define the porting functions by yourself.
+ *
+ *  7. Then, you'll have to setup the passes as indicated in the scheme above.
+ *     You can take a look into SMAA.fx, to see how we did it for our demo.
+ *     Checkout the function wrappers, you may want to copy-paste them!
+ *
+ *  8. It's recommended to validate the produced |edgesTex| and |blendTex|.
+ *     You can use a screenshot from your engine to compare the |edgesTex|
+ *     and |blendTex| produced inside of the engine with the results obtained
+ *     with the reference demo.
+ *
+ *  9. After you get the last pass to work, it's time to optimize. You'll have
+ *     to initialize a stencil buffer in the first pass (discard is already in
+ *     the code), then mask execution by using it the second pass. The last
+ *     pass should be executed in all pixels.
+ *
+ *
+ * After this point you can choose to enable predicated thresholding,
+ * temporal supersampling and motion blur integration:
+ *
+ * a) If you want to use predicated thresholding, take a look into
+ *    SMAA_PREDICATION; you'll need to pass an extra texture in the edge
+ *    detection pass.
+ *
+ * b) If you want to enable temporal supersampling (SMAA T2x):
+ *
+ * 1. The first step is to render using subpixel jitters. I won't go into
+ *    detail, but it's as simple as moving each vertex position in the
+ *    vertex shader, you can check how we do it in our DX10 demo.
+ *
+ * 2. Then, you must setup the temporal resolve. You may want to take a look
+ *    into SMAAResolve for resolving 2x modes. After you get it working, you'll
+ *    probably see ghosting everywhere. But fear not, you can enable the
+ *    CryENGINE temporal reprojection by setting the SMAA_REPROJECTION macro.
+ *    Check out SMAA_DECODE_VELOCITY if your velocity buffer is encoded.
+ *
+ * 3. The next step is to apply SMAA to each subpixel jittered frame, just as
+ *    done for 1x.
+ *
+ * 4. At this point you should already have something usable, but for best
+ *    results the proper area textures must be set depending on current jitter.
+ *    For this, the parameter 'subsampleIndices' of
+ *    'SMAABlendingWeightCalculationPS' must be set as follows, for our T2x
+ *    mode:
+ *
+ *    @SUBSAMPLE_INDICES
+ *
+ *    | S# |  Camera Jitter   |  subsampleIndices    |
+ *    +----+------------------+---------------------+
+ *    |  0 |  ( 0.25, -0.25)  |  float4(1, 1, 1, 0)  |
+ *    |  1 |  (-0.25,  0.25)  |  float4(2, 2, 2, 0)  |
+ *
+ *    These jitter positions assume a bottom-to-top y axis. S# stands for the
+ *    sample number.
+ *
+ * More information about temporal supersampling here:
+ *    http://iryoku.com/aacourse/downloads/13-Anti-Aliasing-Methods-in-CryENGINE-3.pdf
+ *
+ * c) If you want to enable spatial multisampling (SMAA S2x):
+ *
+ * 1. The scene must be rendered using MSAA 2x. The MSAA 2x buffer must be
+ *    created with:
+ *      - DX10:     see below (*)
+ *      - DX10.1:   D3D10_STANDARD_MULTISAMPLE_PATTERN or
+ *      - DX11:     D3D11_STANDARD_MULTISAMPLE_PATTERN
+ *
+ *    This allows to ensure that the subsample order matches the table in
+ *    @SUBSAMPLE_INDICES.
+ *
+ *    (*) In the case of DX10, we refer the reader to:
+ *      - SMAA::detectMSAAOrder and
+ *      - SMAA::msaaReorder
+ *
+ *    These functions allow to match the standard multisample patterns by
+ *    detecting the subsample order for a specific GPU, and reordering
+ *    them appropriately.
+ *
+ * 2. A shader must be run to output each subsample into a separate buffer
+ *    (DX10 is required). You can use SMAASeparate for this purpose, or just do
+ *    it in an existing pass (for example, in the tone mapping pass, which has
+ *    the advantage of feeding tone mapped subsamples to SMAA, which will yield
+ *    better results).
+ *
+ * 3. The full SMAA 1x pipeline must be run for each separated buffer, storing
+ *    the results in the final buffer. The second run should alpha blend with
+ *    the existing final buffer using a blending factor of 0.5.
+ *    'subsampleIndices' must be adjusted as in the SMAA T2x case (see point
+ *    b).
+ *
+ * d) If you want to enable temporal supersampling on top of SMAA S2x
+ *    (which actually is SMAA 4x):
+ *
+ * 1. SMAA 4x consists on temporally jittering SMAA S2x, so the first step is
+ *    to calculate SMAA S2x for current frame. In this case, 'subsampleIndices'
+ *    must be set as follows:
+ *
+ *    | F# | S# |   Camera Jitter    |    Net Jitter     |   subsampleIndices   |
+ *    +----+----+--------------------+-------------------+----------------------+
+ *    |  0 |  0 |  ( 0.125,  0.125)  |  ( 0.375, -0.125) |  float4(5, 3, 1, 3)  |
+ *    |  0 |  1 |  ( 0.125,  0.125)  |  (-0.125,  0.375) |  float4(4, 6, 2, 3)  |
+ *    +----+----+--------------------+-------------------+----------------------+
+ *    |  1 |  2 |  (-0.125, -0.125)  |  ( 0.125, -0.375) |  float4(3, 5, 1, 4)  |
+ *    |  1 |  3 |  (-0.125, -0.125)  |  (-0.375,  0.125) |  float4(6, 4, 2, 4)  |
+ *
+ *    These jitter positions assume a bottom-to-top y axis. F# stands for the
+ *    frame number. S# stands for the sample number.
+ *
+ * 2. After calculating SMAA S2x for current frame (with the new subsample
+ *    indices), previous frame must be reprojected as in SMAA T2x mode (see
+ *    point b).
+ *
+ * e) If motion blur is used, you may want to do the edge detection pass
+ *    together with motion blur. This has two advantages:
+ *
+ * 1. Pixels under heavy motion can be omitted from the edge detection process.
+ *    For these pixels we can just store "no edge", as motion blur will take
+ *    care of them.
+ * 2. The center pixel tap is reused.
+ *
+ * Note that in this case depth testing should be used instead of stenciling,
+ * as we have to write all the pixels in the motion blur pass.
+ *
+ * That's it!
+ */
+
+//-----------------------------------------------------------------------------
+// SMAA Presets
+
+/**
+ * Note that if you use one of these presets, the following configuration
+ * macros will be ignored if set in the "Configurable Defines" section.
+ */
+
+#if defined(SMAA_PRESET_LOW)
+#define SMAA_THRESHOLD 0.15
+#define SMAA_MAX_SEARCH_STEPS 4
+#define SMAA_DISABLE_DIAG_DETECTION
+#define SMAA_DISABLE_CORNER_DETECTION
+#elif defined(SMAA_PRESET_MEDIUM)
+#define SMAA_THRESHOLD 0.1
+#define SMAA_MAX_SEARCH_STEPS 8
+#define SMAA_DISABLE_DIAG_DETECTION
+#define SMAA_DISABLE_CORNER_DETECTION
+#elif defined(SMAA_PRESET_HIGH)
+#define SMAA_THRESHOLD 0.1
+#define SMAA_MAX_SEARCH_STEPS 16
+#define SMAA_MAX_SEARCH_STEPS_DIAG 8
+#define SMAA_CORNER_ROUNDING 25
+#elif defined(SMAA_PRESET_ULTRA)
+#define SMAA_THRESHOLD 0.05
+#define SMAA_MAX_SEARCH_STEPS 32
+#define SMAA_MAX_SEARCH_STEPS_DIAG 16
+#define SMAA_CORNER_ROUNDING 25
+#endif
+
+//-----------------------------------------------------------------------------
+// Configurable Defines
+
+/**
+ * SMAA_THRESHOLD specifies the threshold or sensitivity to edges.
+ * Lowering this value you will be able to detect more edges at the expense of
+ * performance.
+ *
+ * Range: [0, 0.5]
+ *   0.1 is a reasonable value, and allows to catch most visible edges.
+ *   0.05 is a rather overkill value, that allows to catch 'em all.
+ *
+ *   If temporal supersampling is used, 0.2 could be a reasonable value, as low
+ *   contrast edges are properly filtered by just 2x.
+ */
+#ifndef SMAA_THRESHOLD
+#define SMAA_THRESHOLD 0.1
+#endif
+
+/**
+ * SMAA_DEPTH_THRESHOLD specifies the threshold for depth edge detection.
+ *
+ * Range: depends on the depth range of the scene.
+ */
+#ifndef SMAA_DEPTH_THRESHOLD
+#define SMAA_DEPTH_THRESHOLD (0.1 * SMAA_THRESHOLD)
+#endif
+
+/**
+ * SMAA_MAX_SEARCH_STEPS specifies the maximum steps performed in the
+ * horizontal/vertical pattern searches, at each side of the pixel.
+ *
+ * In number of pixels, it's actually the double. So the maximum line length
+ * perfectly handled by, for example 16, is 64 (by perfectly, we meant that
+ * longer lines won't look as good, but still antialiased).
+ *
+ * Range: [0, 112]
+ */
+#ifndef SMAA_MAX_SEARCH_STEPS
+#define SMAA_MAX_SEARCH_STEPS 16
+#endif
+
+/**
+ * SMAA_MAX_SEARCH_STEPS_DIAG specifies the maximum steps performed in the
+ * diagonal pattern searches, at each side of the pixel. In this case we jump
+ * one pixel at time, instead of two.
+ *
+ * Range: [0, 20]
+ *
+ * On high-end machines it is cheap (between a 0.8x and 0.9x slower for 16
+ * steps), but it can have a significant impact on older machines.
+ *
+ * Define SMAA_DISABLE_DIAG_DETECTION to disable diagonal processing.
+ */
+#ifndef SMAA_MAX_SEARCH_STEPS_DIAG
+#define SMAA_MAX_SEARCH_STEPS_DIAG 8
+#endif
+
+/**
+ * SMAA_CORNER_ROUNDING specifies how much sharp corners will be rounded.
+ *
+ * Range: [0, 100]
+ *
+ * Define SMAA_DISABLE_CORNER_DETECTION to disable corner processing.
+ */
+#ifndef SMAA_CORNER_ROUNDING
+#define SMAA_CORNER_ROUNDING 25
+#endif
+
+/**
+ * If there is an neighbor edge that has SMAA_LOCAL_CONTRAST_FACTOR times
+ * bigger contrast than current edge, current edge will be discarded.
+ *
+ * This allows to eliminate spurious crossing edges, and is based on the fact
+ * that, if there is too much contrast in a direction, that will hide
+ * perceptually contrast in the other neighbors.
+ */
+#ifndef SMAA_LOCAL_CONTRAST_ADAPTATION_FACTOR
+#define SMAA_LOCAL_CONTRAST_ADAPTATION_FACTOR 2.0
+#endif
+
+/**
+ * Predicated thresholding allows to better preserve texture details and to
+ * improve performance, by decreasing the number of detected edges using an
+ * additional buffer like the light accumulation buffer, object ids or even the
+ * depth buffer (the depth buffer usage may be limited to indoor or short range
+ * scenes).
+ *
+ * It locally decreases the luma or color threshold if an edge is found in an
+ * additional buffer (so the global threshold can be higher).
+ *
+ * This method was developed by Playstation EDGE MLAA team, and used in
+ * Killzone 3, by using the light accumulation buffer. More information here:
+ *     http://iryoku.com/aacourse/downloads/06-MLAA-on-PS3.pptx
+ */
+#ifndef SMAA_PREDICATION
+#define SMAA_PREDICATION 0
+#endif
+
+/**
+ * Threshold to be used in the additional predication buffer.
+ *
+ * Range: depends on the input, so you'll have to find the magic number that
+ * works for you.
+ */
+#ifndef SMAA_PREDICATION_THRESHOLD
+#define SMAA_PREDICATION_THRESHOLD 0.01
+#endif
+
+/**
+ * How much to scale the global threshold used for luma or color edge
+ * detection when using predication.
+ *
+ * Range: [1, 5]
+ */
+#ifndef SMAA_PREDICATION_SCALE
+#define SMAA_PREDICATION_SCALE 2.0
+#endif
+
+/**
+ * How much to locally decrease the threshold.
+ *
+ * Range: [0, 1]
+ */
+#ifndef SMAA_PREDICATION_STRENGTH
+#define SMAA_PREDICATION_STRENGTH 0.4
+#endif
+
+/**
+ * Temporal reprojection allows to remove ghosting artifacts when using
+ * temporal supersampling. We use the CryEngine 3 method which also introduces
+ * velocity weighting. This feature is of extreme importance for totally
+ * removing ghosting. More information here:
+ *    http://iryoku.com/aacourse/downloads/13-Anti-Aliasing-Methods-in-CryENGINE-3.pdf
+ *
+ * Note that you'll need to setup a velocity buffer for enabling reprojection.
+ * For static geometry, saving the previous depth buffer is a viable
+ * alternative.
+ */
+#ifndef SMAA_REPROJECTION
+#define SMAA_REPROJECTION 0
+#endif
+
+/**
+ * SMAA_REPROJECTION_WEIGHT_SCALE controls the velocity weighting. It allows to
+ * remove ghosting trails behind the moving object, which are not removed by
+ * just using reprojection. Using low values will exhibit ghosting, while using
+ * high values will disable temporal supersampling under motion.
+ *
+ * Behind the scenes, velocity weighting removes temporal supersampling when
+ * the velocity of the subsamples differs (meaning they are different objects).
+ *
+ * Range: [0, 80]
+ */
+#ifndef SMAA_REPROJECTION_WEIGHT_SCALE
+#define SMAA_REPROJECTION_WEIGHT_SCALE 30.0
+#endif
+
+/**
+ * On some compilers, discard and texture cannot be used in vertex shaders. Thus, they need
+ * to be compiled separately.
+ */
+#ifndef SMAA_INCLUDE_VS
+#define SMAA_INCLUDE_VS 1
+#endif
+#ifndef SMAA_INCLUDE_PS
+#define SMAA_INCLUDE_PS 1
+#endif
+
+//-----------------------------------------------------------------------------
+// Texture Access Defines
+
+#ifndef SMAA_AREATEX_SELECT
+#if defined(SMAA_HLSL_3)
+#define SMAA_AREATEX_SELECT(sample) sample.ra
+#else
+#define SMAA_AREATEX_SELECT(sample) sample.rg
+#endif
+#endif
+
+#ifndef SMAA_SEARCHTEX_SELECT
+#define SMAA_SEARCHTEX_SELECT(sample) sample.r
+#endif
+
+#ifndef SMAA_DECODE_VELOCITY
+#define SMAA_DECODE_VELOCITY(sample) sample.rg
+#endif
+
+//-----------------------------------------------------------------------------
+// Non-Configurable Defines
+
+#define SMAA_AREATEX_MAX_DISTANCE 16
+#define SMAA_AREATEX_MAX_DISTANCE_DIAG 20
+#define SMAA_AREATEX_PIXEL_SIZE (1.0 / float2(160.0, 560.0))
+#define SMAA_AREATEX_SUBTEX_SIZE (1.0 / 7.0)
+#define SMAA_SEARCHTEX_SIZE float2(66.0, 33.0)
+#define SMAA_SEARCHTEX_PACKED_SIZE float2(64.0, 16.0)
+#define SMAA_CORNER_ROUNDING_NORM (float(SMAA_CORNER_ROUNDING) / 100.0)
+
+//-----------------------------------------------------------------------------
+// Porting Functions
+
+#if defined(SMAA_HLSL_3)
+#ifndef SMAA_FLIP_Y
+#define SMAA_FLIP_Y 0
+#endif  // SMAA_FLIP_Y
+#define SMAATexture2D(tex) sampler2D tex
+#define SMAATexturePass2D(tex) tex
+#define SMAASampleLevelZero(tex, coord) tex2Dlod(tex, float4(coord, 0.0, 0.0))
+#define SMAASampleLevelZeroPoint(tex, coord) tex2Dlod(tex, float4(coord, 0.0, 0.0))
+#define SMAASampleLevelZeroOffset(tex, coord, offset) tex2Dlod(tex, float4(coord + offset * SMAA_RT_METRICS.xy, 0.0, 0.0))
+#define SMAASample(tex, coord) tex2D(tex, coord)
+#define SMAASamplePoint(tex, coord) tex2D(tex, coord)
+#define SMAASampleOffset(tex, coord, offset) tex2D(tex, coord + offset * SMAA_RT_METRICS.xy)
+#define SMAA_FLATTEN [flatten]
+#define SMAA_BRANCH [branch]
+#endif
+#if defined(SMAA_HLSL_4) || defined(SMAA_HLSL_4_1)
+#ifndef SMAA_FLIP_Y
+#define SMAA_FLIP_Y 0
+#endif  // SMAA_FLIP_Y
+SamplerState LinearSampler { Filter = MIN_MAG_LINEAR_MIP_POINT; AddressU = Clamp; AddressV = Clamp; };
+SamplerState PointSampler { Filter = MIN_MAG_MIP_POINT; AddressU = Clamp; AddressV = Clamp; };
+#define SMAATexture2D(tex) Texture2D tex
+#define SMAATexturePass2D(tex) tex
+#define SMAASampleLevelZero(tex, coord) tex.SampleLevel(LinearSampler, coord, 0)
+#define SMAASampleLevelZeroPoint(tex, coord) tex.SampleLevel(PointSampler, coord, 0)
+#define SMAASampleLevelZeroOffset(tex, coord, offset) tex.SampleLevel(LinearSampler, coord, 0, offset)
+#define SMAASample(tex, coord) tex.Sample(LinearSampler, coord)
+#define SMAASamplePoint(tex, coord) tex.Sample(PointSampler, coord)
+#define SMAASampleOffset(tex, coord, offset) tex.Sample(LinearSampler, coord, offset)
+#define SMAA_FLATTEN [flatten]
+#define SMAA_BRANCH [branch]
+#define SMAATexture2DMS2(tex) Texture2DMS<float4, 2> tex
+#define SMAALoad(tex, pos, sample) tex.Load(pos, sample)
+#if defined(SMAA_HLSL_4_1)
+#define SMAAGather(tex, coord) tex.Gather(LinearSampler, coord, 0)
+#endif
+#endif
+
+#if defined(SMAA_GLSL_2) || defined(SMAA_GLSL_3) || defined(SMAA_GLSL_4)
+#ifndef SMAA_FLIP_Y
+#define SMAA_FLIP_Y 1
+#endif  // SMAA_FLIP_Y
+
+#define SMAATexture2D(tex) sampler2D tex
+#define SMAATexturePass2D(tex) tex
+#if defined(SMAA_GLSL_2)
+#define SMAASampleLevelZero(tex, coord) texture2DLod(tex, coord, 0.0)
+#define SMAASampleLevelZeroPoint(tex, coord) texture2DLod(tex, coord, 0.0)
+#define SMAASampleLevelZeroOffset(tex, coord, offset) texture2DLodOffset(tex, coord, 0.0, offset)
+#define SMAASample(tex, coord) texture2D(tex, coord)
+#define SMAASamplePoint(tex, coord) texture2D(tex, coord)
+#define SMAASampleOffset(tex, coord, offset) texture2D(tex, coord, offset)
+#else
+#define SMAASampleLevelZero(tex, coord) textureLod(tex, coord, 0.0)
+#define SMAASampleLevelZeroPoint(tex, coord) textureLod(tex, coord, 0.0)
+#define SMAASampleLevelZeroOffset(tex, coord, offset) textureLodOffset(tex, coord, 0.0, offset)
+#define SMAASample(tex, coord) texture(tex, coord)
+#define SMAASamplePoint(tex, coord) texture(tex, coord)
+#define SMAASampleOffset(tex, coord, offset) textureOffset(tex, coord, offset)
+#endif
+#define SMAA_FLATTEN
+#define SMAA_BRANCH
+#define lerp(a, b, t) mix(a, b, t)
+#define saturate(a) clamp(a, 0.0, 1.0)
+#if defined(SMAA_GLSL_4)
+#define mad(a, b, c) fma(a, b, c)
+#define SMAAGather(tex, coord) textureGather(tex, coord)
+#else
+#define mad(a, b, c) (a * b + c)
+#endif
+#if defined(SMAA_GLSL_3) || defined(SMAA_GLSL_4)
+#define SMAATexture2DMS2(tex) sampler2DMS tex
+#define SMAALoad(tex, pos, sample) texelFetch(tex, pos, sample)
+#endif
+#define float2 vec2
+#define float3 vec3
+#define float4 vec4
+#define int2 ivec2
+#define int3 ivec3
+#define int4 ivec4
+#define bool2 bvec2
+#define bool3 bvec3
+#define bool4 bvec4
+#endif
+
+#if !defined(SMAA_HLSL_3) && !defined(SMAA_HLSL_4) && !defined(SMAA_HLSL_4_1) && !defined(SMAA_GLSL_2) && !defined(SMAA_GLSL_3) && !defined(SMAA_GLSL_4) && !defined(SMAA_CUSTOM_SL)
+#error you must define the shading language: SMAA_HLSL_*, SMAA_GLSL_* or SMAA_CUSTOM_SL
+#endif
+
+
+#if SMAA_FLIP_Y
+
+#define API_V_DIR(v) -(v)
+#define API_V_COORD(v) (1.0 - v)
+#define API_V_BELOW(v1, v2) v1 < v2
+#define API_V_ABOVE(v1, v2) v1 > v2
+
+#else  // VULKAN_FLIP
+
+#define API_V_DIR(v) v
+#define API_V_COORD(v) v
+#define API_V_BELOW(v1, v2) v1 > v2
+#define API_V_ABOVE(v1, v2) v1 < v2
+
+#endif  // VULKAN_FLIP
+
+
+//-----------------------------------------------------------------------------
+// Misc functions
+
+#if SMAA_INCLUDE_PS
+/**
+ * Gathers current pixel, and the top-left neighbors.
+ */
+float3 SMAAGatherNeighbours(float2 texcoord,
+                            float4 offset[3],
+                            SMAATexture2D(tex)) {
+    #ifdef SMAAGather
+
+    #if SMAA_FLIP_Y
+    return SMAAGather(tex, texcoord + SMAA_RT_METRICS.xy * float2(-0.5,  0.5)).zwy;
+    #else  // SMAA_FLIP_Y
+    return SMAAGather(tex, texcoord + SMAA_RT_METRICS.xy * float2(-0.5, -0.5)).grb;
+    #endif  // SMAA_FLIP_Y
+
+    #else  // SMAAGather
+    float P = SMAASamplePoint(tex, texcoord).r;
+    float Pleft = SMAASamplePoint(tex, offset[0].xy).r;
+    float Ptop  = SMAASamplePoint(tex, offset[0].zw).r;
+    return float3(P, Pleft, Ptop);
+    #endif
+}
+
+/**
+ * Adjusts the threshold by means of predication.
+ */
+float2 SMAACalculatePredicatedThreshold(float2 texcoord,
+                                        float4 offset[3],
+                                        SMAATexture2D(predicationTex)) {
+    float3 neighbours = SMAAGatherNeighbours(texcoord, offset, SMAATexturePass2D(predicationTex));
+    float2 delta = abs(neighbours.xx - neighbours.yz);
+    float2 edges = step(SMAA_PREDICATION_THRESHOLD, delta);
+    return SMAA_PREDICATION_SCALE * SMAA_THRESHOLD * (1.0 - SMAA_PREDICATION_STRENGTH * edges);
+}
+
+#endif  // SMAA_INCLUDE_PS
+
+/**
+ * Conditional move:
+ */
+void SMAAMovc(bool2 cond, inout float2 variable, float2 value) {
+    SMAA_FLATTEN if (cond.x) variable.x = value.x;
+    SMAA_FLATTEN if (cond.y) variable.y = value.y;
+}
+
+void SMAAMovc(bool4 cond, inout float4 variable, float4 value) {
+    SMAAMovc(cond.xy, variable.xy, value.xy);
+    SMAAMovc(cond.zw, variable.zw, value.zw);
+}
+
+
+#if SMAA_INCLUDE_VS
+//-----------------------------------------------------------------------------
+// Vertex Shaders
+
+/**
+ * Edge Detection Vertex Shader
+ */
+void SMAAEdgeDetectionVS(float2 texcoord,
+                         out float4 offset[3]) {
+    offset[0] = mad(SMAA_RT_METRICS.xyxy, float4(-1.0, 0.0, 0.0, API_V_DIR(-1.0)), texcoord.xyxy);
+    offset[1] = mad(SMAA_RT_METRICS.xyxy, float4( 1.0, 0.0, 0.0, API_V_DIR(1.0)), texcoord.xyxy);
+    offset[2] = mad(SMAA_RT_METRICS.xyxy, float4(-2.0, 0.0, 0.0, API_V_DIR(-2.0)), texcoord.xyxy);
+}
+
+/**
+ * Blend Weight Calculation Vertex Shader
+ */
+void SMAABlendingWeightCalculationVS(float2 texcoord,
+                                     out float2 pixcoord,
+                                     out float4 offset[3]) {
+    pixcoord = texcoord * SMAA_RT_METRICS.zw;
+
+    // We will use these offsets for the searches later on (see @PSEUDO_GATHER4):
+    offset[0] = mad(SMAA_RT_METRICS.xyxy, float4(-0.25, API_V_DIR(-0.125),  1.25, API_V_DIR(-0.125)), texcoord.xyxy);
+    offset[1] = mad(SMAA_RT_METRICS.xyxy, float4(-0.125, API_V_DIR(-0.25), -0.125,  API_V_DIR(1.25)), texcoord.xyxy);
+
+    // And these for the searches, they indicate the ends of the loops:
+    offset[2] = mad(SMAA_RT_METRICS.xxyy,
+                    float4(-2.0, 2.0, API_V_DIR(-2.0), API_V_DIR(2.0)) * float(SMAA_MAX_SEARCH_STEPS),
+                    float4(offset[0].xz, offset[1].yw));
+}
+
+/**
+ * Neighborhood Blending Vertex Shader
+ */
+void SMAANeighborhoodBlendingVS(float2 texcoord,
+                                out float4 offset) {
+    offset = mad(SMAA_RT_METRICS.xyxy, float4( 1.0, 0.0, 0.0, API_V_DIR(1.0)), texcoord.xyxy);
+}
+#endif // SMAA_INCLUDE_VS
+
+#if SMAA_INCLUDE_PS
+//-----------------------------------------------------------------------------
+// Edge Detection Pixel Shaders (First Pass)
+
+/**
+ * Luma Edge Detection
+ *
+ * IMPORTANT NOTICE: luma edge detection requires gamma-corrected colors, and
+ * thus 'colorTex' should be a non-sRGB texture.
+ */
+float2 SMAALumaEdgeDetectionPS(float2 texcoord,
+                               float4 offset[3],
+                               SMAATexture2D(colorTex)
+                               #if SMAA_PREDICATION
+                               , SMAATexture2D(predicationTex)
+                               #endif
+                               ) {
+    // Calculate the threshold:
+    #if SMAA_PREDICATION
+    float2 threshold = SMAACalculatePredicatedThreshold(texcoord, offset, SMAATexturePass2D(predicationTex));
+    #else
+    float2 threshold = float2(SMAA_THRESHOLD, SMAA_THRESHOLD);
+    #endif
+
+    // Calculate lumas:
+    float3 weights = float3(0.2126, 0.7152, 0.0722);
+    float L = dot(SMAASamplePoint(colorTex, texcoord).rgb, weights);
+
+    float Lleft = dot(SMAASamplePoint(colorTex, offset[0].xy).rgb, weights);
+    float Ltop  = dot(SMAASamplePoint(colorTex, offset[0].zw).rgb, weights);
+
+    // We do the usual threshold:
+    float4 delta;
+    delta.xy = abs(L - float2(Lleft, Ltop));
+    float2 edges = step(threshold, delta.xy);
+
+    // Then discard if there is no edge:
+    if (dot(edges, float2(1.0, 1.0)) == 0.0)
+        discard;
+
+    // Calculate right and bottom deltas:
+    float Lright = dot(SMAASamplePoint(colorTex, offset[1].xy).rgb, weights);
+    float Lbottom  = dot(SMAASamplePoint(colorTex, offset[1].zw).rgb, weights);
+    delta.zw = abs(L - float2(Lright, Lbottom));
+
+    // Calculate the maximum delta in the direct neighborhood:
+    float2 maxDelta = max(delta.xy, delta.zw);
+
+    // Calculate left-left and top-top deltas:
+    float Lleftleft = dot(SMAASamplePoint(colorTex, offset[2].xy).rgb, weights);
+    float Ltoptop = dot(SMAASamplePoint(colorTex, offset[2].zw).rgb, weights);
+    delta.zw = abs(float2(Lleft, Ltop) - float2(Lleftleft, Ltoptop));
+
+    // Calculate the final maximum delta:
+    maxDelta = max(maxDelta.xy, delta.zw);
+    float finalDelta = max(maxDelta.x, maxDelta.y);
+
+    // Local contrast adaptation:
+    edges.xy *= step(finalDelta, SMAA_LOCAL_CONTRAST_ADAPTATION_FACTOR * delta.xy);
+
+    return edges;
+}
+
+/**
+ * Color Edge Detection
+ *
+ * IMPORTANT NOTICE: color edge detection requires gamma-corrected colors, and
+ * thus 'colorTex' should be a non-sRGB texture.
+ */
+float2 SMAAColorEdgeDetectionPS(float2 texcoord,
+                                float4 offset[3],
+                                SMAATexture2D(colorTex)
+                                #if SMAA_PREDICATION
+                                , SMAATexture2D(predicationTex)
+                                #endif
+                                ) {
+    // Calculate the threshold:
+    #if SMAA_PREDICATION
+    float2 threshold = SMAACalculatePredicatedThreshold(texcoord, offset, predicationTex);
+    #else
+    float2 threshold = float2(SMAA_THRESHOLD, SMAA_THRESHOLD);
+    #endif
+
+    // Calculate color deltas:
+    float4 delta;
+    float3 C = SMAASamplePoint(colorTex, texcoord).rgb;
+
+    float3 Cleft = SMAASamplePoint(colorTex, offset[0].xy).rgb;
+    float3 t = abs(C - Cleft);
+    delta.x = max(max(t.r, t.g), t.b);
+
+    float3 Ctop  = SMAASamplePoint(colorTex, offset[0].zw).rgb;
+    t = abs(C - Ctop);
+    delta.y = max(max(t.r, t.g), t.b);
+
+    // We do the usual threshold:
+    float2 edges = step(threshold, delta.xy);
+
+    // Then discard if there is no edge:
+    if (dot(edges, float2(1.0, 1.0)) == 0.0)
+        discard;
+
+    // Calculate right and bottom deltas:
+    float3 Cright = SMAASamplePoint(colorTex, offset[1].xy).rgb;
+    t = abs(C - Cright);
+    delta.z = max(max(t.r, t.g), t.b);
+
+    float3 Cbottom  = SMAASamplePoint(colorTex, offset[1].zw).rgb;
+    t = abs(C - Cbottom);
+    delta.w = max(max(t.r, t.g), t.b);
+
+    // Calculate the maximum delta in the direct neighborhood:
+    float2 maxDelta = max(delta.xy, delta.zw);
+
+    // Calculate left-left and top-top deltas:
+    float3 Cleftleft  = SMAASamplePoint(colorTex, offset[2].xy).rgb;
+    t = abs(C - Cleftleft);
+    delta.z = max(max(t.r, t.g), t.b);
+
+    float3 Ctoptop = SMAASamplePoint(colorTex, offset[2].zw).rgb;
+    t = abs(C - Ctoptop);
+    delta.w = max(max(t.r, t.g), t.b);
+
+    // Calculate the final maximum delta:
+    maxDelta = max(maxDelta.xy, delta.zw);
+    float finalDelta = max(maxDelta.x, maxDelta.y);
+
+    // Local contrast adaptation:
+    edges.xy *= step(finalDelta, SMAA_LOCAL_CONTRAST_ADAPTATION_FACTOR * delta.xy);
+
+    return edges;
+}
+
+/**
+ * Depth Edge Detection
+ */
+float2 SMAADepthEdgeDetectionPS(float2 texcoord,
+                                float4 offset[3],
+                                SMAATexture2D(depthTex)) {
+    float3 neighbours = SMAAGatherNeighbours(texcoord, offset, SMAATexturePass2D(depthTex));
+    float2 delta = abs(neighbours.xx - float2(neighbours.y, neighbours.z));
+    float2 edges = step(SMAA_DEPTH_THRESHOLD, delta);
+
+    if (dot(edges, float2(1.0, 1.0)) == 0.0)
+        discard;
+
+    return edges;
+}
+
+//-----------------------------------------------------------------------------
+// Diagonal Search Functions
+
+#if !defined(SMAA_DISABLE_DIAG_DETECTION)
+
+/**
+ * Allows to decode two binary values from a bilinear-filtered access.
+ */
+float2 SMAADecodeDiagBilinearAccess(float2 e) {
+    // Bilinear access for fetching 'e' have a 0.25 offset, and we are
+    // interested in the R and G edges:
+    //
+    // +---G---+-------+
+    // |   x o R   x   |
+    // +-------+-------+
+    //
+    // Then, if one of these edge is enabled:
+    //   Red:   (0.75 * X + 0.25 * 1) => 0.25 or 1.0
+    //   Green: (0.75 * 1 + 0.25 * X) => 0.75 or 1.0
+    //
+    // This function will unpack the values (mad + mul + round):
+    // wolframalpha.com: round(x * abs(5 * x - 5 * 0.75)) plot 0 to 1
+    e.r = e.r * abs(5.0 * e.r - 5.0 * 0.75);
+    return round(e);
+}
+
+float4 SMAADecodeDiagBilinearAccess(float4 e) {
+    e.rb = e.rb * abs(5.0 * e.rb - 5.0 * 0.75);
+    return round(e);
+}
+
+/**
+ * These functions allows to perform diagonal pattern searches.
+ */
+float2 SMAASearchDiag1(SMAATexture2D(edgesTex), float2 texcoord, float2 dir, out float2 e) {
+    dir.y = API_V_DIR(dir.y);
+    float4 coord = float4(texcoord, -1.0, 1.0);
+    float3 t = float3(SMAA_RT_METRICS.xy, 1.0);
+    while (coord.z < float(SMAA_MAX_SEARCH_STEPS_DIAG - 1) &&
+           coord.w > 0.9) {
+        coord.xyz = mad(t, float3(dir, 1.0), coord.xyz);
+        e = SMAASampleLevelZero(edgesTex, coord.xy).rg;
+        coord.w = dot(e, float2(0.5, 0.5));
+    }
+    return coord.zw;
+}
+
+float2 SMAASearchDiag2(SMAATexture2D(edgesTex), float2 texcoord, float2 dir, out float2 e) {
+    dir.y = API_V_DIR(dir.y);
+    float4 coord = float4(texcoord, -1.0, 1.0);
+    coord.x += 0.25 * SMAA_RT_METRICS.x; // See @SearchDiag2Optimization
+    float3 t = float3(SMAA_RT_METRICS.xy, 1.0);
+    while (coord.z < float(SMAA_MAX_SEARCH_STEPS_DIAG - 1) &&
+           coord.w > 0.9) {
+        coord.xyz = mad(t, float3(dir, 1.0), coord.xyz);
+
+        // @SearchDiag2Optimization
+        // Fetch both edges at once using bilinear filtering:
+        e = SMAASampleLevelZero(edgesTex, coord.xy).rg;
+        e = SMAADecodeDiagBilinearAccess(e);
+
+        // Non-optimized version:
+        // e.g = SMAASampleLevelZero(edgesTex, coord.xy).g;
+        // e.r = SMAASampleLevelZeroOffset(edgesTex, coord.xy, int2(1, 0)).r;
+
+        coord.w = dot(e, float2(0.5, 0.5));
+    }
+    return coord.zw;
+}
+
+/**
+ * Similar to SMAAArea, this calculates the area corresponding to a certain
+ * diagonal distance and crossing edges 'e'.
+ */
+float2 SMAAAreaDiag(SMAATexture2D(areaTex), float2 dist, float2 e, float offset) {
+    float2 texcoord = mad(float2(SMAA_AREATEX_MAX_DISTANCE_DIAG, SMAA_AREATEX_MAX_DISTANCE_DIAG), e, dist);
+
+    // We do a scale and bias for mapping to texel space:
+    texcoord = mad(SMAA_AREATEX_PIXEL_SIZE, texcoord, 0.5 * SMAA_AREATEX_PIXEL_SIZE);
+
+    // Diagonal areas are on the second half of the texture:
+    texcoord.x += 0.5;
+
+    // Move to proper place, according to the subpixel offset:
+    texcoord.y += SMAA_AREATEX_SUBTEX_SIZE * offset;
+
+    texcoord.y = API_V_COORD(texcoord.y);
+
+    // Do it!
+    return SMAA_AREATEX_SELECT(SMAASampleLevelZero(areaTex, texcoord));
+}
+
+/**
+ * This searches for diagonal patterns and returns the corresponding weights.
+ */
+float2 SMAACalculateDiagWeights(SMAATexture2D(edgesTex), SMAATexture2D(areaTex), float2 texcoord, float2 e, float4 subsampleIndices) {
+    float2 weights = float2(0.0, 0.0);
+
+    // Search for the line ends:
+    float4 d;
+    float2 end;
+    if (e.r > 0.0) {
+        d.xz = SMAASearchDiag1(SMAATexturePass2D(edgesTex), texcoord, float2(-1.0,  1.0), end);
+        d.x += float(end.y > 0.9);
+    } else
+        d.xz = float2(0.0, 0.0);
+    d.yw = SMAASearchDiag1(SMAATexturePass2D(edgesTex), texcoord, float2(1.0, -1.0), end);
+
+    SMAA_BRANCH
+    if (d.x + d.y > 2.0) { // d.x + d.y + 1 > 3
+        // Fetch the crossing edges:
+        float4 coords = mad(float4(-d.x + 0.25, API_V_DIR(d.x), d.y, API_V_DIR(-d.y - 0.25)), SMAA_RT_METRICS.xyxy, texcoord.xyxy);
+        float4 c;
+        c.xy = SMAASampleLevelZeroOffset(edgesTex, coords.xy, int2(-1,  0)).rg;
+        c.zw = SMAASampleLevelZeroOffset(edgesTex, coords.zw, int2( 1,  0)).rg;
+        c.yxwz = SMAADecodeDiagBilinearAccess(c.xyzw);
+
+        // Non-optimized version:
+        // float4 coords = mad(float4(-d.x, d.x, d.y, -d.y), SMAA_RT_METRICS.xyxy, texcoord.xyxy);
+        // float4 c;
+        // c.x = SMAASampleLevelZeroOffset(edgesTex, coords.xy, int2(-1,  0)).g;
+        // c.y = SMAASampleLevelZeroOffset(edgesTex, coords.xy, int2( 0,  0)).r;
+        // c.z = SMAASampleLevelZeroOffset(edgesTex, coords.zw, int2( 1,  0)).g;
+        // c.w = SMAASampleLevelZeroOffset(edgesTex, coords.zw, int2( 1, -1)).r;
+
+        // Merge crossing edges at each side into a single value:
+        float2 cc = mad(float2(2.0, 2.0), c.xz, c.yw);
+
+        // Remove the crossing edge if we didn't found the end of the line:
+        SMAAMovc(bool2(step(0.9, d.zw)), cc, float2(0.0, 0.0));
+
+        // Fetch the areas for this line:
+        weights += SMAAAreaDiag(SMAATexturePass2D(areaTex), d.xy, cc, subsampleIndices.z);
+    }
+
+    // Search for the line ends:
+    d.xz = SMAASearchDiag2(SMAATexturePass2D(edgesTex), texcoord, float2(-1.0, -1.0), end);
+    if (SMAASampleLevelZeroOffset(edgesTex, texcoord, int2(1, 0)).r > 0.0) {
+        d.yw = SMAASearchDiag2(SMAATexturePass2D(edgesTex), texcoord, float2(1.0, 1.0), end);
+        d.y += float(end.y > 0.9);
+    } else
+        d.yw = float2(0.0, 0.0);
+
+    SMAA_BRANCH
+    if (d.x + d.y > 2.0) { // d.x + d.y + 1 > 3
+        // Fetch the crossing edges:
+        float4 coords = mad(float4(-d.x, API_V_DIR(-d.x), d.y, API_V_DIR(d.y)), SMAA_RT_METRICS.xyxy, texcoord.xyxy);
+        float4 c;
+        c.x  = SMAASampleLevelZeroOffset(edgesTex, coords.xy, int2(-1,  0)).g;
+        c.y  = SMAASampleLevelZeroOffset(edgesTex, coords.xy, int2( 0, API_V_DIR(-1))).r;
+        c.zw = SMAASampleLevelZeroOffset(edgesTex, coords.zw, int2( 1,  0)).gr;
+        float2 cc = mad(float2(2.0, 2.0), c.xz, c.yw);
+
+        // Remove the crossing edge if we didn't found the end of the line:
+        SMAAMovc(bool2(step(0.9, d.zw)), cc, float2(0.0, 0.0));
+
+        // Fetch the areas for this line:
+        weights += SMAAAreaDiag(SMAATexturePass2D(areaTex), d.xy, cc, subsampleIndices.w).gr;
+    }
+
+    return weights;
+}
+#endif
+
+//-----------------------------------------------------------------------------
+// Horizontal/Vertical Search Functions
+
+/**
+ * This allows to determine how much length should we add in the last step
+ * of the searches. It takes the bilinearly interpolated edge (see
+ * @PSEUDO_GATHER4), and adds 0, 1 or 2, depending on which edges and
+ * crossing edges are active.
+ */
+float SMAASearchLength(SMAATexture2D(searchTex), float2 e, float offset) {
+    // The texture is flipped vertically, with left and right cases taking half
+    // of the space horizontally:
+    float2 scale = SMAA_SEARCHTEX_SIZE * float2(0.5, -1.0);
+    float2 bias = SMAA_SEARCHTEX_SIZE * float2(offset, 1.0);
+
+    // Scale and bias to access texel centers:
+    scale += float2(-1.0,  1.0);
+    bias  += float2( 0.5, -0.5);
+
+    // Convert from pixel coordinates to texcoords:
+    // (We use SMAA_SEARCHTEX_PACKED_SIZE because the texture is cropped)
+    scale *= 1.0 / SMAA_SEARCHTEX_PACKED_SIZE;
+    bias *= 1.0 / SMAA_SEARCHTEX_PACKED_SIZE;
+
+    float2 coord = mad(scale, e, bias);
+    coord.y = API_V_COORD(coord.y);
+
+    // Lookup the search texture:
+    return SMAA_SEARCHTEX_SELECT(SMAASampleLevelZero(searchTex, coord));
+}
+
+/**
+ * Horizontal/vertical search functions for the 2nd pass.
+ */
+float SMAASearchXLeft(SMAATexture2D(edgesTex), SMAATexture2D(searchTex), float2 texcoord, float end) {
+    /**
+     * @PSEUDO_GATHER4
+     * This texcoord has been offset by (-0.25, -0.125) in the vertex shader to
+     * sample between edge, thus fetching four edges in a row.
+     * Sampling with different offsets in each direction allows to disambiguate
+     * which edges are active from the four fetched ones.
+     */
+    float2 e = float2(0.0, 1.0);
+    while (texcoord.x > end &&
+           e.g > 0.8281 && // Is there some edge not activated?
+           e.r == 0.0) { // Or is there a crossing edge that breaks the line?
+        e = SMAASampleLevelZero(edgesTex, texcoord).rg;
+        texcoord = mad(-float2(2.0, 0.0), SMAA_RT_METRICS.xy, texcoord);
+    }
+
+    float offset = mad(-(255.0 / 127.0), SMAASearchLength(SMAATexturePass2D(searchTex), e, 0.0), 3.25);
+    return mad(SMAA_RT_METRICS.x, offset, texcoord.x);
+
+    // Non-optimized version:
+    // We correct the previous (-0.25, -0.125) offset we applied:
+    // texcoord.x += 0.25 * SMAA_RT_METRICS.x;
+
+    // The searches are bias by 1, so adjust the coords accordingly:
+    // texcoord.x += SMAA_RT_METRICS.x;
+
+    // Disambiguate the length added by the last step:
+    // texcoord.x += 2.0 * SMAA_RT_METRICS.x; // Undo last step
+    // texcoord.x -= SMAA_RT_METRICS.x * (255.0 / 127.0) * SMAASearchLength(SMAATexturePass2D(searchTex), e, 0.0);
+    // return mad(SMAA_RT_METRICS.x, offset, texcoord.x);
+}
+
+float SMAASearchXRight(SMAATexture2D(edgesTex), SMAATexture2D(searchTex), float2 texcoord, float end) {
+    float2 e = float2(0.0, 1.0);
+    while (texcoord.x < end &&
+           e.g > 0.8281 && // Is there some edge not activated?
+           e.r == 0.0) { // Or is there a crossing edge that breaks the line?
+        e = SMAASampleLevelZero(edgesTex, texcoord).rg;
+        texcoord = mad(float2(2.0, 0.0), SMAA_RT_METRICS.xy, texcoord);
+    }
+    float offset = mad(-(255.0 / 127.0), SMAASearchLength(SMAATexturePass2D(searchTex), e, 0.5), 3.25);
+    return mad(-SMAA_RT_METRICS.x, offset, texcoord.x);
+}
+
+float SMAASearchYUp(SMAATexture2D(edgesTex), SMAATexture2D(searchTex), float2 texcoord, float end) {
+    float2 e = float2(1.0, 0.0);
+    while (API_V_BELOW(texcoord.y, end) &&
+           e.r > 0.8281 && // Is there some edge not activated?
+           e.g == 0.0) { // Or is there a crossing edge that breaks the line?
+        e = SMAASampleLevelZero(edgesTex, texcoord).rg;
+        texcoord = mad(-float2(0.0, API_V_DIR(2.0)), SMAA_RT_METRICS.xy, texcoord);
+    }
+    float offset = mad(-(255.0 / 127.0), SMAASearchLength(SMAATexturePass2D(searchTex), e.gr, 0.0), 3.25);
+    return mad(SMAA_RT_METRICS.y, API_V_DIR(offset), texcoord.y);
+}
+
+float SMAASearchYDown(SMAATexture2D(edgesTex), SMAATexture2D(searchTex), float2 texcoord, float end) {
+    float2 e = float2(1.0, 0.0);
+    while (API_V_ABOVE(texcoord.y, end) &&
+           e.r > 0.8281 && // Is there some edge not activated?
+           e.g == 0.0) { // Or is there a crossing edge that breaks the line?
+        e = SMAASampleLevelZero(edgesTex, texcoord).rg;
+        texcoord = mad(float2(0.0, API_V_DIR(2.0)), SMAA_RT_METRICS.xy, texcoord);
+    }
+    float offset = mad(-(255.0 / 127.0), SMAASearchLength(SMAATexturePass2D(searchTex), e.gr, 0.5), 3.25);
+    return mad(-SMAA_RT_METRICS.y, API_V_DIR(offset), texcoord.y);
+}
+
+/**
+ * Ok, we have the distance and both crossing edges. So, what are the areas
+ * at each side of current edge?
+ */
+float2 SMAAArea(SMAATexture2D(areaTex), float2 dist, float e1, float e2, float offset) {
+    // Rounding prevents precision errors of bilinear filtering:
+    float2 texcoord = mad(float2(SMAA_AREATEX_MAX_DISTANCE, SMAA_AREATEX_MAX_DISTANCE), round(4.0 * float2(e1, e2)), dist);
+
+    // We do a scale and bias for mapping to texel space:
+    texcoord = mad(SMAA_AREATEX_PIXEL_SIZE, texcoord, 0.5 * SMAA_AREATEX_PIXEL_SIZE);
+
+    // Move to proper place, according to the subpixel offset:
+    texcoord.y = mad(SMAA_AREATEX_SUBTEX_SIZE, offset, texcoord.y);
+
+    texcoord.y = API_V_COORD(texcoord.y);
+
+    // Do it!
+    return SMAA_AREATEX_SELECT(SMAASampleLevelZero(areaTex, texcoord));
+}
+
+//-----------------------------------------------------------------------------
+// Corner Detection Functions
+
+void SMAADetectHorizontalCornerPattern(SMAATexture2D(edgesTex), inout float2 weights, float4 texcoord, float2 d) {
+    #if !defined(SMAA_DISABLE_CORNER_DETECTION)
+    float2 leftRight = step(d.xy, d.yx);
+    float2 rounding = (1.0 - SMAA_CORNER_ROUNDING_NORM) * leftRight;
+
+    rounding /= leftRight.x + leftRight.y; // Reduce blending for pixels in the center of a line.
+
+    float2 factor = float2(1.0, 1.0);
+    factor.x -= rounding.x * SMAASampleLevelZeroOffset(edgesTex, texcoord.xy, int2(0,  API_V_DIR(1))).r;
+    factor.x -= rounding.y * SMAASampleLevelZeroOffset(edgesTex, texcoord.zw, int2(1,  API_V_DIR(1))).r;
+    factor.y -= rounding.x * SMAASampleLevelZeroOffset(edgesTex, texcoord.xy, int2(0, API_V_DIR(-2))).r;
+    factor.y -= rounding.y * SMAASampleLevelZeroOffset(edgesTex, texcoord.zw, int2(1, API_V_DIR(-2))).r;
+
+    weights *= saturate(factor);
+    #endif
+}
+
+void SMAADetectVerticalCornerPattern(SMAATexture2D(edgesTex), inout float2 weights, float4 texcoord, float2 d) {
+    #if !defined(SMAA_DISABLE_CORNER_DETECTION)
+    float2 leftRight = step(d.xy, d.yx);
+    float2 rounding = (1.0 - SMAA_CORNER_ROUNDING_NORM) * leftRight;
+
+    rounding /= leftRight.x + leftRight.y;
+
+    float2 factor = float2(1.0, 1.0);
+    factor.x -= rounding.x * SMAASampleLevelZeroOffset(edgesTex, texcoord.xy, int2( 1, 0)).g;
+    factor.x -= rounding.y * SMAASampleLevelZeroOffset(edgesTex, texcoord.zw, int2( 1, API_V_DIR(1))).g;
+    factor.y -= rounding.x * SMAASampleLevelZeroOffset(edgesTex, texcoord.xy, int2(-2, 0)).g;
+    factor.y -= rounding.y * SMAASampleLevelZeroOffset(edgesTex, texcoord.zw, int2(-2, API_V_DIR(1))).g;
+
+    weights *= saturate(factor);
+    #endif
+}
+
+//-----------------------------------------------------------------------------
+// Blending Weight Calculation Pixel Shader (Second Pass)
+
+float4 SMAABlendingWeightCalculationPS(float2 texcoord,
+                                       float2 pixcoord,
+                                       float4 offset[3],
+                                       SMAATexture2D(edgesTex),
+                                       SMAATexture2D(areaTex),
+                                       SMAATexture2D(searchTex),
+                                       float4 subsampleIndices) { // Just pass zero for SMAA 1x, see @SUBSAMPLE_INDICES.
+    float4 weights = float4(0.0, 0.0, 0.0, 0.0);
+
+    float2 e = SMAASample(edgesTex, texcoord).rg;
+
+    SMAA_BRANCH
+    if (e.g > 0.0) { // Edge at north
+        #if !defined(SMAA_DISABLE_DIAG_DETECTION)
+        // Diagonals have both north and west edges, so searching for them in
+        // one of the boundaries is enough.
+        weights.rg = SMAACalculateDiagWeights(SMAATexturePass2D(edgesTex), SMAATexturePass2D(areaTex), texcoord, e, subsampleIndices);
+
+        // We give priority to diagonals, so if we find a diagonal we skip
+        // horizontal/vertical processing.
+        SMAA_BRANCH
+        if (weights.r == -weights.g) { // weights.r + weights.g == 0.0
+        #endif
+
+        float2 d;
+
+        // Find the distance to the left:
+        float3 coords;
+        coords.x = SMAASearchXLeft(SMAATexturePass2D(edgesTex), SMAATexturePass2D(searchTex), offset[0].xy, offset[2].x);
+        coords.y = offset[1].y; // offset[1].y = texcoord.y - 0.25 * SMAA_RT_METRICS.y (@CROSSING_OFFSET)
+        d.x = coords.x;
+
+        // Now fetch the left crossing edges, two at a time using bilinear
+        // filtering. Sampling at -0.25 (see @CROSSING_OFFSET) enables to
+        // discern what value each edge has:
+        float e1 = SMAASampleLevelZero(edgesTex, coords.xy).r;
+
+        // Find the distance to the right:
+        coords.z = SMAASearchXRight(SMAATexturePass2D(edgesTex), SMAATexturePass2D(searchTex), offset[0].zw, offset[2].y);
+        d.y = coords.z;
+
+        // We want the distances to be in pixel units (doing this here allow to
+        // better interleave arithmetic and memory accesses):
+        d = abs(round(mad(SMAA_RT_METRICS.zz, d, -pixcoord.xx)));
+
+        // SMAAArea below needs a sqrt, as the areas texture is compressed
+        // quadratically:
+        float2 sqrt_d = sqrt(d);
+
+        // Fetch the right crossing edges:
+        float e2 = SMAASampleLevelZeroOffset(edgesTex, coords.zy, int2(1, 0)).r;
+
+        // Ok, we know how this pattern looks like, now it is time for getting
+        // the actual area:
+        weights.rg = SMAAArea(SMAATexturePass2D(areaTex), sqrt_d, e1, e2, subsampleIndices.y);
+
+        // Fix corners:
+        coords.y = texcoord.y;
+        SMAADetectHorizontalCornerPattern(SMAATexturePass2D(edgesTex), weights.rg, coords.xyzy, d);
+
+        #if !defined(SMAA_DISABLE_DIAG_DETECTION)
+        } else
+            e.r = 0.0; // Skip vertical processing.
+        #endif
+    }
+
+    SMAA_BRANCH
+    if (e.r > 0.0) { // Edge at west
+        float2 d;
+
+        // Find the distance to the top:
+        float3 coords;
+        coords.y = SMAASearchYUp(SMAATexturePass2D(edgesTex), SMAATexturePass2D(searchTex), offset[1].xy, offset[2].z);
+        coords.x = offset[0].x; // offset[1].x = texcoord.x - 0.25 * SMAA_RT_METRICS.x;
+        d.x = coords.y;
+
+        // Fetch the top crossing edges:
+        float e1 = SMAASampleLevelZero(edgesTex, coords.xy).g;
+
+        // Find the distance to the bottom:
+        coords.z = SMAASearchYDown(SMAATexturePass2D(edgesTex), SMAATexturePass2D(searchTex), offset[1].zw, offset[2].w);
+        d.y = coords.z;
+
+        // We want the distances to be in pixel units:
+        d = abs(round(mad(SMAA_RT_METRICS.ww, d, -pixcoord.yy)));
+
+        // SMAAArea below needs a sqrt, as the areas texture is compressed
+        // quadratically:
+        float2 sqrt_d = sqrt(d);
+
+        // Fetch the bottom crossing edges:
+        float e2 = SMAASampleLevelZeroOffset(edgesTex, coords.xz, int2(0, API_V_DIR(1))).g;
+
+        // Get the area for this direction:
+        weights.ba = SMAAArea(SMAATexturePass2D(areaTex), sqrt_d, e1, e2, subsampleIndices.x);
+
+        // Fix corners:
+        coords.x = texcoord.x;
+        SMAADetectVerticalCornerPattern(SMAATexturePass2D(edgesTex), weights.ba, coords.xyxz, d);
+    }
+
+    return weights;
+}
+
+//-----------------------------------------------------------------------------
+// Neighborhood Blending Pixel Shader (Third Pass)
+
+float4 SMAANeighborhoodBlendingPS(float2 texcoord,
+                                  float4 offset,
+                                  SMAATexture2D(colorTex),
+                                  SMAATexture2D(blendTex)
+                                  #if SMAA_REPROJECTION
+                                  , SMAATexture2D(velocityTex)
+                                  #endif
+                                  ) {
+    // Fetch the blending weights for current pixel:
+    float4 a;
+    a.x = SMAASample(blendTex, offset.xy).a; // Right
+    a.y = SMAASample(blendTex, offset.zw).g; // Top
+    a.wz = SMAASample(blendTex, texcoord).xz; // Bottom / Left
+
+    // Is there any blending weight with a value greater than 0.0?
+    SMAA_BRANCH
+    if (dot(a, float4(1.0, 1.0, 1.0, 1.0)) < 1e-5) {
+        float4 color = SMAASampleLevelZero(colorTex, texcoord);
+
+        #if SMAA_REPROJECTION
+        float2 velocity = SMAA_DECODE_VELOCITY(SMAASampleLevelZero(velocityTex, texcoord));
+
+        // Pack velocity into the alpha channel:
+        color.a = sqrt(5.0 * length(velocity));
+        #endif
+
+        return color;
+    } else {
+        bool h = max(a.x, a.z) > max(a.y, a.w); // max(horizontal) > max(vertical)
+
+        // Calculate the blending offsets:
+        float4 blendingOffset = float4(0.0, API_V_DIR(a.y), 0.0, API_V_DIR(a.w));
+        float2 blendingWeight = a.yw;
+        SMAAMovc(bool4(h, h, h, h), blendingOffset, float4(a.x, 0.0, a.z, 0.0));
+        SMAAMovc(bool2(h, h), blendingWeight, a.xz);
+        blendingWeight /= dot(blendingWeight, float2(1.0, 1.0));
+
+        // Calculate the texture coordinates:
+        float4 blendingCoord = mad(blendingOffset, float4(SMAA_RT_METRICS.xy, -SMAA_RT_METRICS.xy), texcoord.xyxy);
+
+        // We exploit bilinear filtering to mix current pixel with the chosen
+        // neighbor:
+        float4 color = blendingWeight.x * SMAASampleLevelZero(colorTex, blendingCoord.xy);
+        color += blendingWeight.y * SMAASampleLevelZero(colorTex, blendingCoord.zw);
+
+        #if SMAA_REPROJECTION
+        // Antialias velocity for proper reprojection in a later stage:
+        float2 velocity = blendingWeight.x * SMAA_DECODE_VELOCITY(SMAASampleLevelZero(velocityTex, blendingCoord.xy));
+        velocity += blendingWeight.y * SMAA_DECODE_VELOCITY(SMAASampleLevelZero(velocityTex, blendingCoord.zw));
+
+        // Pack velocity into the alpha channel:
+        color.a = sqrt(5.0 * length(velocity));
+        #endif
+
+        return color;
+    }
+}
+
+//-----------------------------------------------------------------------------
+// Temporal Resolve Pixel Shader (Optional Pass)
+
+float4 SMAAResolvePS(float2 texcoord,
+                     SMAATexture2D(currentColorTex),
+                     SMAATexture2D(previousColorTex)
+                     #if SMAA_REPROJECTION
+                     , SMAATexture2D(velocityTex)
+                     #endif
+                     ) {
+    #if SMAA_REPROJECTION
+    // Velocity is assumed to be calculated for motion blur, so we need to
+    // inverse it for reprojection:
+    float2 velocity = -SMAA_DECODE_VELOCITY(SMAASamplePoint(velocityTex, texcoord).rg);
+
+    // Fetch current pixel:
+    float4 current = SMAASamplePoint(currentColorTex, texcoord);
+
+    // Reproject current coordinates and fetch previous pixel:
+    float4 previous = SMAASamplePoint(previousColorTex, texcoord + velocity);
+
+    // Attenuate the previous pixel if the velocity is different:
+    float delta = abs(current.a * current.a - previous.a * previous.a) / 5.0;
+    float weight = 0.5 * saturate(1.0 - sqrt(delta) * SMAA_REPROJECTION_WEIGHT_SCALE);
+
+    // Blend the pixels according to the calculated weight:
+    return lerp(current, previous, weight);
+    #else
+    // Just blend the pixels:
+    float4 current = SMAASamplePoint(currentColorTex, texcoord);
+    float4 previous = SMAASamplePoint(previousColorTex, texcoord);
+    return lerp(current, previous, 0.5);
+    #endif
+}
+
+//-----------------------------------------------------------------------------
+// Separate Multisamples Pixel Shader (Optional Pass)
+
+#ifdef SMAALoad
+void SMAASeparatePS(float4 position,
+                    float2 texcoord,
+                    out float4 target0,
+                    out float4 target1,
+                    SMAATexture2DMS2(colorTexMS)) {
+    int2 pos = int2(position.xy);
+    target0 = SMAALoad(colorTexMS, pos, 0);
+    target1 = SMAALoad(colorTexMS, pos, 1);
+}
+#endif
+
+//-----------------------------------------------------------------------------
+#endif // SMAA_INCLUDE_PS
diff --git a/indra/newview/app_settings/shaders/class1/deferred/SMAABlendWeightsF.glsl b/indra/newview/app_settings/shaders/class1/deferred/SMAABlendWeightsF.glsl
new file mode 100644
index 0000000000..3332c5f58f
--- /dev/null
+++ b/indra/newview/app_settings/shaders/class1/deferred/SMAABlendWeightsF.glsl
@@ -0,0 +1,57 @@
+/**
+ * @file SMAABlendWeightsF.glsl
+ *
+ * $LicenseInfo:firstyear=2024&license=viewerlgpl$
+ * Second Life Viewer Source Code
+ * Copyright (C) 2024, Linden Research, Inc.
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation;
+ * version 2.1 of the License only.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
+ *
+ * Linden Research, Inc., 945 Battery Street, San Francisco, CA  94111  USA
+ * $/LicenseInfo$
+ */
+
+/*[EXTRA_CODE_HERE]*/
+
+out vec4 frag_color;
+
+in vec2 vary_texcoord0;
+in vec2 vary_pixcoord;
+in vec4 vary_offset[3];
+
+uniform sampler2D edgesTex;
+uniform sampler2D areaTex;
+uniform sampler2D searchTex;
+
+vec4 SMAABlendingWeightCalculationPS(vec2 texcoord,
+                                       vec2 pixcoord,
+                                       vec4 offset[3],
+                                       sampler2D edgesTex,
+                                       sampler2D areaTex,
+                                       sampler2D searchTex,
+                                       vec4 subsampleIndices);
+
+void main()
+{
+    frag_color = SMAABlendingWeightCalculationPS(vary_texcoord0,
+                                                 vary_pixcoord,
+                                                 vary_offset,
+                                                 edgesTex,
+                                                 areaTex,
+                                                 searchTex,
+                                                 vec4(0.0)
+                                                 );
+}
+
diff --git a/indra/newview/app_settings/shaders/class1/deferred/SMAABlendWeightsV.glsl b/indra/newview/app_settings/shaders/class1/deferred/SMAABlendWeightsV.glsl
new file mode 100644
index 0000000000..52f85ef30c
--- /dev/null
+++ b/indra/newview/app_settings/shaders/class1/deferred/SMAABlendWeightsV.glsl
@@ -0,0 +1,51 @@
+/**
+ * @file SMAABlendWeightsV.glsl
+ *
+ * $LicenseInfo:firstyear=2024&license=viewerlgpl$
+ * Second Life Viewer Source Code
+ * Copyright (C) 2024, Linden Research, Inc.
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation;
+ * version 2.1 of the License only.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
+ *
+ * Linden Research, Inc., 945 Battery Street, San Francisco, CA  94111  USA
+ * $/LicenseInfo$
+ */
+
+/*[EXTRA_CODE_HERE]*/
+
+uniform mat4 modelview_projection_matrix;
+
+in vec3 position;
+
+out vec2 vary_texcoord0;
+out vec2 vary_pixcoord;
+out vec4 vary_offset[3];
+
+#define float4 vec4
+#define float2 vec2
+void SMAABlendingWeightCalculationVS(float2 texcoord,
+                                     out float2 pixcoord,
+                                     out float4 offset[3]);
+
+void main()
+{
+    gl_Position = vec4(position.xyz, 1.0);
+    vary_texcoord0 = (gl_Position.xy*0.5+0.5);
+
+    SMAABlendingWeightCalculationVS(vary_texcoord0,
+                                    vary_pixcoord,
+                                    vary_offset);
+}
+
diff --git a/indra/newview/app_settings/shaders/class1/deferred/SMAAEdgeDetectF.glsl b/indra/newview/app_settings/shaders/class1/deferred/SMAAEdgeDetectF.glsl
new file mode 100644
index 0000000000..0a8cd4a4ea
--- /dev/null
+++ b/indra/newview/app_settings/shaders/class1/deferred/SMAAEdgeDetectF.glsl
@@ -0,0 +1,59 @@
+/**
+ * @file SMAAEdgeDetectF.glsl
+ * $LicenseInfo:firstyear=2024&license=viewerlgpl$
+ * Second Life Viewer Source Code
+ * Copyright (C) 2024, Linden Research, Inc.
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation;
+ * version 2.1 of the License only.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
+ *
+ * Linden Research, Inc., 945 Battery Street, San Francisco, CA  94111  USA
+ * $/LicenseInfo$
+ */
+
+/*[EXTRA_CODE_HERE]*/
+
+out vec4 frag_color;
+
+in vec2 vary_texcoord0;
+in vec4 vary_offset[3];
+
+uniform sampler2D diffuseRect;
+#if SMAA_PREDICATION
+uniform sampler2D predicationTex;
+#endif
+
+#define float4 vec4
+#define float2 vec2
+#define SMAATexture2D(tex) sampler2D tex
+
+float2 SMAAColorEdgeDetectionPS(float2 texcoord,
+                                float4 offset[3],
+                                SMAATexture2D(colorTex)
+                                #if SMAA_PREDICATION
+                                , SMAATexture2D(predicationTex)
+                                #endif
+                                );
+
+void main()
+{
+    vec2 val = SMAAColorEdgeDetectionPS(vary_texcoord0,
+                                          vary_offset,
+                                          diffuseRect
+                                          #if SMAA_PREDICATION
+                                          , predicationTex
+                                          #endif
+                                          );
+    frag_color = float4(val,0.0,0.0);
+}
diff --git a/indra/newview/app_settings/shaders/class1/deferred/SMAAEdgeDetectV.glsl b/indra/newview/app_settings/shaders/class1/deferred/SMAAEdgeDetectV.glsl
new file mode 100644
index 0000000000..7c0184bfc4
--- /dev/null
+++ b/indra/newview/app_settings/shaders/class1/deferred/SMAAEdgeDetectV.glsl
@@ -0,0 +1,45 @@
+/**
+ * @file SMAAEdgeDetectV.glsl
+ * $LicenseInfo:firstyear=2024&license=viewerlgpl$
+ * Second Life Viewer Source Code
+ * Copyright (C) 2024, Linden Research, Inc.
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation;
+ * version 2.1 of the License only.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
+ *
+ * Linden Research, Inc., 945 Battery Street, San Francisco, CA  94111  USA
+ * $/LicenseInfo$
+ */
+
+/*[EXTRA_CODE_HERE]*/
+
+uniform mat4 modelview_projection_matrix;
+
+in vec3 position;
+
+out vec2 vary_texcoord0;
+out vec4 vary_offset[3];
+
+#define float4 vec4
+#define float2 vec2
+void SMAAEdgeDetectionVS(float2 texcoord,
+                         out float4 offset[3]);
+
+void main()
+{
+    gl_Position = vec4(position.xyz, 1.0);
+    vary_texcoord0 = (gl_Position.xy*0.5+0.5);
+
+    SMAAEdgeDetectionVS(vary_texcoord0, vary_offset);
+}
diff --git a/indra/newview/app_settings/shaders/class1/deferred/SMAANeighborhoodBlendF.glsl b/indra/newview/app_settings/shaders/class1/deferred/SMAANeighborhoodBlendF.glsl
new file mode 100644
index 0000000000..3276405447
--- /dev/null
+++ b/indra/newview/app_settings/shaders/class1/deferred/SMAANeighborhoodBlendF.glsl
@@ -0,0 +1,63 @@
+/**
+ * @file SMAANeighborhoodBlendF.glsl
+ *
+ * $LicenseInfo:firstyear=2024&license=viewerlgpl$
+ * Second Life Viewer Source Code
+ * Copyright (C) 2024, Linden Research, Inc.
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation;
+ * version 2.1 of the License only.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
+ *
+ * Linden Research, Inc., 945 Battery Street, San Francisco, CA  94111  USA
+ * $/LicenseInfo$
+ */
+
+/*[EXTRA_CODE_HERE]*/
+
+out vec4 frag_color;
+
+in vec2 vary_texcoord0;
+in vec4 vary_offset;
+
+uniform sampler2D diffuseRect;
+uniform sampler2D blendTex;
+#if SMAA_REPROJECTION
+uniform sampler2D velocityTex;
+#endif
+
+#define float4 vec4
+#define float2 vec2
+#define SMAATexture2D(tex) sampler2D tex
+
+float4 SMAANeighborhoodBlendingPS(float2 texcoord,
+                                  float4 offset,
+                                  SMAATexture2D(colorTex),
+                                  SMAATexture2D(blendTex)
+                                  #if SMAA_REPROJECTION
+                                  , SMAATexture2D(velocityTex)
+                                  #endif
+                                  );
+
+void main()
+{
+    frag_color = SMAANeighborhoodBlendingPS(vary_texcoord0,
+                                            vary_offset,
+                                            diffuseRect,
+                                            blendTex
+                                            #if SMAA_REPROJECTION
+                                            , velocityTex
+                                            #endif
+                                            );
+}
+
diff --git a/indra/newview/app_settings/shaders/class1/deferred/SMAANeighborhoodBlendV.glsl b/indra/newview/app_settings/shaders/class1/deferred/SMAANeighborhoodBlendV.glsl
new file mode 100644
index 0000000000..7ea1ac61e3
--- /dev/null
+++ b/indra/newview/app_settings/shaders/class1/deferred/SMAANeighborhoodBlendV.glsl
@@ -0,0 +1,47 @@
+/**
+ * @file SMAANeighborhoodBlendV.glsl
+ *
+ * $LicenseInfo:firstyear=2024&license=viewerlgpl$
+ * Second Life Viewer Source Code
+ * Copyright (C) 2024, Linden Research, Inc.
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation;
+ * version 2.1 of the License only.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
+ *
+ * Linden Research, Inc., 945 Battery Street, San Francisco, CA  94111  USA
+ * $/LicenseInfo$
+ */
+
+/*[EXTRA_CODE_HERE]*/
+
+uniform mat4 modelview_projection_matrix;
+
+in vec3 position;
+
+out vec2 vary_texcoord0;
+out vec4 vary_offset;
+
+#define float4 vec4
+#define float2 vec2
+void SMAANeighborhoodBlendingVS(float2 texcoord,
+                                out float4 offset);
+
+void main()
+{
+    gl_Position = vec4(position.xyz, 1.0);
+    vary_texcoord0 = (gl_Position.xy*0.5+0.5);
+
+    SMAANeighborhoodBlendingVS(vary_texcoord0, vary_offset);
+}
+
diff --git a/indra/newview/app_settings/shaders/class1/deferred/deferredUtil.glsl b/indra/newview/app_settings/shaders/class1/deferred/deferredUtil.glsl
index ab0e4fd4d8..3319e32cdf 100644
--- a/indra/newview/app_settings/shaders/class1/deferred/deferredUtil.glsl
+++ b/indra/newview/app_settings/shaders/class1/deferred/deferredUtil.glsl
@@ -99,10 +99,13 @@ void calcHalfVectors(vec3 lv, vec3 n, vec3 v,
 {
     l  = normalize(lv);
     h  = normalize(l + v);
-    nh = clamp(dot(n, h), 0.0, 1.0);
-    nl = clamp(dot(n, l), 0.0, 1.0);
-    nv = clamp(dot(n, v), 0.0, 1.0);
-    vh = clamp(dot(v, h), 0.0, 1.0);
+
+    // lower bound to avoid divide by zero
+    float eps = 0.000001;
+    nh = clamp(dot(n, h), eps, 1.0);
+    nl = clamp(dot(n, l), eps, 1.0);
+    nv = clamp(dot(n, v), eps, 1.0);
+    vh = clamp(dot(v, h), eps, 1.0);
 
     lightDist = length(lv);
 }
diff --git a/indra/newview/app_settings/shaders/class1/deferred/fxaaF.glsl b/indra/newview/app_settings/shaders/class1/deferred/fxaaF.glsl
index 94dac7e5a9..655cb1ea97 100644
--- a/indra/newview/app_settings/shaders/class1/deferred/fxaaF.glsl
+++ b/indra/newview/app_settings/shaders/class1/deferred/fxaaF.glsl
@@ -32,7 +32,7 @@ out vec4 frag_color;
 
 #define FXAA_PC 1
 //#define FXAA_GLSL_130 1
-#define FXAA_QUALITY__PRESET 12
+//#define FXAA_QUALITY__PRESET 12
 
 /*============================================================================
 
@@ -256,6 +256,10 @@ A. Or use FXAA_GREEN_AS_LUMA.
     #define FXAA_GLSL_130 0
 #endif
 /*--------------------------------------------------------------------------*/
+#ifndef FXAA_GLSL_400
+    #define FXAA_GLSL_400 0
+#endif
+/*--------------------------------------------------------------------------*/
 #ifndef FXAA_HLSL_3
     #define FXAA_HLSL_3 0
 #endif
@@ -342,8 +346,8 @@ A. Or use FXAA_GREEN_AS_LUMA.
     // 1 = API supports gather4 on alpha channel.
     // 0 = API does not support gather4 on alpha channel.
     //
-    #if (FXAA_GLSL_130 == 0)
-        #define FXAA_GATHER4_ALPHA 0
+    #if (FXAA_GLSL_400 == 1)
+        #define FXAA_GATHER4_ALPHA 1
     #endif
     #if (FXAA_HLSL_5 == 1)
         #define FXAA_GATHER4_ALPHA 1
@@ -652,7 +656,7 @@ NOTE the other tuning knobs are now in the shader function inputs!
                                 API PORTING
 
 ============================================================================*/
-#if (FXAA_GLSL_120 == 1) || (FXAA_GLSL_130 == 1)
+#if (FXAA_GLSL_120 == 1) || (FXAA_GLSL_130 == 1) || (FXAA_GLSL_400 == 1)
     #define FxaaBool bool
     #define FxaaDiscard discard
     #define FxaaFloat float
@@ -714,6 +718,16 @@ NOTE the other tuning knobs are now in the shader function inputs!
     #endif
 #endif
 /*--------------------------------------------------------------------------*/
+#if (FXAA_GLSL_400 == 1)
+    // Requires "#version 400" or better
+    #define FxaaTexTop(t, p) textureLod(t, p, 0.0)
+    #define FxaaTexOff(t, p, o, r) textureLodOffset(t, p, 0.0, o)
+    #define FxaaTexAlpha4(t, p) textureGather(t, p, 3)
+    #define FxaaTexOffAlpha4(t, p, o) textureGatherOffset(t, p, o, 3)
+    #define FxaaTexGreen4(t, p) textureGather(t, p, 1)
+    #define FxaaTexOffGreen4(t, p, o) textureGatherOffset(t, p, o, 1)
+#endif
+/*--------------------------------------------------------------------------*/
 #if (FXAA_HLSL_3 == 1) || (FXAA_360 == 1) || (FXAA_PS3 == 1)
     #define FxaaInt2 float2
     #define FxaaTex sampler2D
diff --git a/indra/newview/app_settings/shaders/class1/deferred/normgenF.glsl b/indra/newview/app_settings/shaders/class1/deferred/normgenF.glsl
index 902746366d..607a8c6ef6 100644
--- a/indra/newview/app_settings/shaders/class1/deferred/normgenF.glsl
+++ b/indra/newview/app_settings/shaders/class1/deferred/normgenF.glsl
@@ -25,24 +25,53 @@
 
 /*[EXTRA_CODE_HERE]*/
 
+
+
+// generate a normal map using an approximation of the old emboss bump map "brightness/darkness" technique
+// srcMap is a source color image, output should be a normal
+
 out vec4 frag_color;
 
-uniform sampler2D alphaMap;
+uniform sampler2D srcMap;
 
 in vec2 vary_texcoord0;
 
 uniform float stepX;
 uniform float stepY;
 uniform float norm_scale;
+uniform int bump_code;
+
+#define BE_BRIGHTNESS 1
+#define BE_DARKNESS 2
+
+// get luminance or inverse luminance depending on bump_code
+float getBumpValue(vec2 texcoord)
+{
+    vec3 c = texture(srcMap, texcoord).rgb;
+
+    vec3 WEIGHT = vec3(0.2995, 0.5875, 0.1145);
+
+    float l = dot(c, WEIGHT);
+
+    if (bump_code == BE_DARKNESS)
+    {
+        l = 1.0 - l;
+    }
+
+    return l;
+}
+
 
 void main()
 {
-    float c = texture(alphaMap, vary_texcoord0).r;
+    float c = getBumpValue(vary_texcoord0);
+
+    float scaler = 512.0;
 
-    vec3 right = vec3(norm_scale, 0, (texture(alphaMap, vary_texcoord0+vec2(stepX, 0)).r-c)*255);
-    vec3 left = vec3(-norm_scale, 0, (texture(alphaMap, vary_texcoord0-vec2(stepX, 0)).r-c)*255);
-    vec3 up = vec3(0, -norm_scale, (texture(alphaMap, vary_texcoord0-vec2(0, stepY)).r-c)*255);
-    vec3 down = vec3(0, norm_scale, (texture(alphaMap, vary_texcoord0+vec2(0, stepY)).r-c)*255);
+    vec3 right = vec3(norm_scale, 0, (getBumpValue(vary_texcoord0+vec2(stepX, 0))-c)*scaler);
+    vec3 left = vec3(-norm_scale, 0, (getBumpValue(vary_texcoord0-vec2(stepX, 0))-c)*scaler);
+    vec3 up = vec3(0, -norm_scale, (getBumpValue(vary_texcoord0-vec2(0, stepY))-c)*scaler);
+    vec3 down = vec3(0, norm_scale, (getBumpValue(vary_texcoord0+vec2(0, stepY))-c)*scaler);
 
     vec3 norm = cross(right, down) + cross(down, left) + cross(left,up) + cross(up, right);
 
diff --git a/indra/newview/app_settings/shaders/class1/deferred/pbralphaV.glsl b/indra/newview/app_settings/shaders/class1/deferred/pbralphaV.glsl
index ae179d3f37..66adf50fa9 100644
--- a/indra/newview/app_settings/shaders/class1/deferred/pbralphaV.glsl
+++ b/indra/newview/app_settings/shaders/class1/deferred/pbralphaV.glsl
@@ -69,7 +69,7 @@ flat out float vary_sign;
 out vec3 vary_normal;
 
 vec2 texture_transform(vec2 vertex_texcoord, vec4[2] khr_gltf_transform, mat4 sl_animation_transform);
-vec3 tangent_space_transform(vec4 vertex_tangent, vec3 vertex_normal, vec4[2] khr_gltf_transform, mat4 sl_animation_transform);
+vec4 tangent_space_transform(vec4 vertex_tangent, vec3 vertex_normal, vec4[2] khr_gltf_transform, mat4 sl_animation_transform);
 
 
 void main()
@@ -103,8 +103,9 @@ void main()
 
     n = normalize(n);
 
-    vary_tangent = normalize(tangent_space_transform(vec4(t, tangent.w), n, texture_normal_transform, texture_matrix0));
-    vary_sign = tangent.w;
+    vec4 transformed_tangent = tangent_space_transform(vec4(t, tangent.w), n, texture_normal_transform, texture_matrix0);
+    vary_tangent = normalize(transformed_tangent.xyz);
+    vary_sign = transformed_tangent.w;
     vary_normal = n;
 
     vertex_color = diffuse_color;
diff --git a/indra/newview/app_settings/shaders/class1/deferred/pbropaqueV.glsl b/indra/newview/app_settings/shaders/class1/deferred/pbropaqueV.glsl
index fd020afd57..0ad9bf5e4b 100644
--- a/indra/newview/app_settings/shaders/class1/deferred/pbropaqueV.glsl
+++ b/indra/newview/app_settings/shaders/class1/deferred/pbropaqueV.glsl
@@ -63,7 +63,7 @@ out vec3 vary_normal;
 out vec3 vary_position;
 
 vec2 texture_transform(vec2 vertex_texcoord, vec4[2] khr_gltf_transform, mat4 sl_animation_transform);
-vec3 tangent_space_transform(vec4 vertex_tangent, vec3 vertex_normal, vec4[2] khr_gltf_transform, mat4 sl_animation_transform);
+vec4 tangent_space_transform(vec4 vertex_tangent, vec3 vertex_normal, vec4[2] khr_gltf_transform, mat4 sl_animation_transform);
 
 void main()
 {
@@ -97,8 +97,9 @@ void main()
 
     n = normalize(n);
 
-    vary_tangent = normalize(tangent_space_transform(vec4(t, tangent.w), n, texture_normal_transform, texture_matrix0));
-    vary_sign = tangent.w;
+    vec4 transformed_tangent = tangent_space_transform(vec4(t, tangent.w), n, texture_normal_transform, texture_matrix0);
+    vary_tangent = normalize(transformed_tangent.xyz);
+    vary_sign = transformed_tangent.w;
     vary_normal = n;
 
     vertex_color = diffuse_color;
diff --git a/indra/newview/app_settings/shaders/class1/deferred/pbrterrainF.glsl b/indra/newview/app_settings/shaders/class1/deferred/pbrterrainF.glsl
index 2cb7ff196b..410c447c64 100644
--- a/indra/newview/app_settings/shaders/class1/deferred/pbrterrainF.glsl
+++ b/indra/newview/app_settings/shaders/class1/deferred/pbrterrainF.glsl
@@ -30,6 +30,9 @@
 #define TERRAIN_PBR_DETAIL_NORMAL -2
 #define TERRAIN_PBR_DETAIL_METALLIC_ROUGHNESS -3
 
+#define TERRAIN_PAINT_TYPE_HEIGHTMAP_WITH_NOISE 0
+#define TERRAIN_PAINT_TYPE_PBR_PAINTMAP 1
+
 #if TERRAIN_PLANAR_TEXTURE_SAMPLE_COUNT == 3
 #define TerrainCoord vec4[3]
 #elif TERRAIN_PLANAR_TEXTURE_SAMPLE_COUNT == 1
@@ -48,6 +51,7 @@ struct TerrainMix
 };
 
 TerrainMix get_terrain_mix_weights(float alpha1, float alpha2, float alphaFinal);
+TerrainMix get_terrain_usage_from_weight3(vec3 weight3);
 
 struct PBRMix
 {
@@ -75,6 +79,9 @@ PBRMix terrain_sample_and_multiply_pbr(
 #endif
 #if (TERRAIN_PBR_DETAIL >= TERRAIN_PBR_DETAIL_NORMAL)
     , sampler2D tex_vNt
+#if TERRAIN_PLANAR_TEXTURE_SAMPLE_COUNT == 3
+    , float transform_sign
+#endif
 #endif
 #if (TERRAIN_PBR_DETAIL >= TERRAIN_PBR_DETAIL_EMISSIVE)
     , sampler2D tex_emissive
@@ -94,7 +101,11 @@ PBRMix mix_pbr(PBRMix mix1, PBRMix mix2, float mix2_weight);
 
 out vec4 frag_data[4];
 
+#if TERRAIN_PAINT_TYPE == TERRAIN_PAINT_TYPE_HEIGHTMAP_WITH_NOISE
 uniform sampler2D alpha_ramp;
+#elif TERRAIN_PAINT_TYPE == TERRAIN_PAINT_TYPE_PBR_PAINTMAP
+uniform sampler2D paint_map;
+#endif
 
 // https://registry.khronos.org/glTF/specs/2.0/glTF-2.0.html#additional-textures
 uniform sampler2D detail_0_base_color;
@@ -130,19 +141,25 @@ uniform vec3[4] emissiveColors;
 #endif
 uniform vec4 minimum_alphas; // PBR alphaMode: MASK, See: mAlphaCutoff, setAlphaCutoff()
 
-#if TERRAIN_PLANAR_TEXTURE_SAMPLE_COUNT == 3
-in vec4[10] vary_coords;
-#elif TERRAIN_PLANAR_TEXTURE_SAMPLE_COUNT == 1
-in vec4[2] vary_coords;
-#endif
 in vec3 vary_position;
 in vec3 vary_normal;
 #if (TERRAIN_PBR_DETAIL >= TERRAIN_PBR_DETAIL_NORMAL)
 in vec3 vary_tangents[4];
-flat in float vary_sign;
+flat in float vary_signs[4];
 #endif
+
+// vary_texcoord* are used for terrain composition, vary_coords are used for terrain UVs
+#if TERRAIN_PAINT_TYPE == TERRAIN_PAINT_TYPE_HEIGHTMAP_WITH_NOISE
 in vec4 vary_texcoord0;
 in vec4 vary_texcoord1;
+#elif TERRAIN_PAINT_TYPE == TERRAIN_PAINT_TYPE_PBR_PAINTMAP
+in vec2 vary_texcoord;
+#endif
+#if TERRAIN_PLANAR_TEXTURE_SAMPLE_COUNT == 3
+in vec4[10] vary_coords;
+#elif TERRAIN_PLANAR_TEXTURE_SAMPLE_COUNT == 1
+in vec4[2] vary_coords;
+#endif
 
 void mirrorClip(vec3 position);
 
@@ -150,11 +167,11 @@ float terrain_mix(TerrainMix tm, vec4 tms4);
 
 #if (TERRAIN_PBR_DETAIL >= TERRAIN_PBR_DETAIL_NORMAL)
 // from mikktspace.com
-vec3 mikktspace(vec3 vNt, vec3 vT)
+vec3 mikktspace(vec3 vNt, vec3 vT, float sign)
 {
     vec3 vN = vary_normal;
 
-    vec3 vB = vary_sign * cross(vN, vT);
+    vec3 vB = sign * cross(vN, vT);
     vec3 tnorm = normalize( vNt.x * vT + vNt.y * vB + vNt.z * vN );
 
     tnorm *= gl_FrontFacing ? 1.0 : -1.0;
@@ -168,11 +185,16 @@ void main()
     // Make sure we clip the terrain if we're in a mirror.
     mirrorClip(vary_position);
 
+    TerrainMix tm;
+#if TERRAIN_PAINT_TYPE == TERRAIN_PAINT_TYPE_HEIGHTMAP_WITH_NOISE
     float alpha1 = texture(alpha_ramp, vary_texcoord0.zw).a;
     float alpha2 = texture(alpha_ramp,vary_texcoord1.xy).a;
     float alphaFinal = texture(alpha_ramp, vary_texcoord1.zw).a;
 
-    TerrainMix tm = get_terrain_mix_weights(alpha1, alpha2, alphaFinal);
+    tm = get_terrain_mix_weights(alpha1, alpha2, alphaFinal);
+#elif TERRAIN_PAINT_TYPE == TERRAIN_PAINT_TYPE_PBR_PAINTMAP
+    tm = get_terrain_usage_from_weight3(texture(paint_map, vary_texcoord).xyz);
+#endif
 
 #if (TERRAIN_PBR_DETAIL >= TERRAIN_PBR_DETAIL_OCCLUSION)
     // RGB = Occlusion, Roughness, Metal
@@ -216,6 +238,9 @@ void main()
 #endif
 #if (TERRAIN_PBR_DETAIL >= TERRAIN_PBR_DETAIL_NORMAL)
             , detail_0_normal
+#if TERRAIN_PLANAR_TEXTURE_SAMPLE_COUNT == 3
+            , vary_signs[0]
+#endif
 #endif
 #if (TERRAIN_PBR_DETAIL >= TERRAIN_PBR_DETAIL_EMISSIVE)
             , detail_0_emissive
@@ -231,7 +256,7 @@ void main()
 #endif
         );
 #if (TERRAIN_PBR_DETAIL >= TERRAIN_PBR_DETAIL_NORMAL)
-        mix2.vNt = mikktspace(mix2.vNt, vary_tangents[0]);
+        mix2.vNt = mikktspace(mix2.vNt, vary_tangents[0], vary_signs[0]);
 #endif
         pbr_mix = mix_pbr(pbr_mix, mix2, tm.weight.x);
         break;
@@ -258,6 +283,9 @@ void main()
 #endif
 #if (TERRAIN_PBR_DETAIL >= TERRAIN_PBR_DETAIL_NORMAL)
             , detail_1_normal
+#if TERRAIN_PLANAR_TEXTURE_SAMPLE_COUNT == 3
+            , vary_signs[1]
+#endif
 #endif
 #if (TERRAIN_PBR_DETAIL >= TERRAIN_PBR_DETAIL_EMISSIVE)
             , detail_1_emissive
@@ -273,7 +301,7 @@ void main()
 #endif
         );
 #if (TERRAIN_PBR_DETAIL >= TERRAIN_PBR_DETAIL_NORMAL)
-        mix2.vNt = mikktspace(mix2.vNt, vary_tangents[1]);
+        mix2.vNt = mikktspace(mix2.vNt, vary_tangents[1], vary_signs[1]);
 #endif
         pbr_mix = mix_pbr(pbr_mix, mix2, tm.weight.y);
         break;
@@ -300,6 +328,9 @@ void main()
 #endif
 #if (TERRAIN_PBR_DETAIL >= TERRAIN_PBR_DETAIL_NORMAL)
             , detail_2_normal
+#if TERRAIN_PLANAR_TEXTURE_SAMPLE_COUNT == 3
+            , vary_signs[2]
+#endif
 #endif
 #if (TERRAIN_PBR_DETAIL >= TERRAIN_PBR_DETAIL_EMISSIVE)
             , detail_2_emissive
@@ -315,7 +346,7 @@ void main()
 #endif
         );
 #if (TERRAIN_PBR_DETAIL >= TERRAIN_PBR_DETAIL_NORMAL)
-        mix2.vNt = mikktspace(mix2.vNt, vary_tangents[2]);
+        mix2.vNt = mikktspace(mix2.vNt, vary_tangents[2], vary_signs[2]);
 #endif
         pbr_mix = mix_pbr(pbr_mix, mix2, tm.weight.z);
         break;
@@ -342,6 +373,9 @@ void main()
 #endif
 #if (TERRAIN_PBR_DETAIL >= TERRAIN_PBR_DETAIL_NORMAL)
             , detail_3_normal
+#if TERRAIN_PLANAR_TEXTURE_SAMPLE_COUNT == 3
+            , vary_signs[3]
+#endif
 #endif
 #if (TERRAIN_PBR_DETAIL >= TERRAIN_PBR_DETAIL_EMISSIVE)
             , detail_3_emissive
@@ -357,7 +391,7 @@ void main()
 #endif
         );
 #if (TERRAIN_PBR_DETAIL >= TERRAIN_PBR_DETAIL_NORMAL)
-        mix2.vNt = mikktspace(mix2.vNt, vary_tangents[3]);
+        mix2.vNt = mikktspace(mix2.vNt, vary_tangents[3], vary_signs[3]);
 #endif
         pbr_mix = mix_pbr(pbr_mix, mix2, tm.weight.w);
         break;
diff --git a/indra/newview/app_settings/shaders/class1/deferred/pbrterrainUtilF.glsl b/indra/newview/app_settings/shaders/class1/deferred/pbrterrainUtilF.glsl
index 1ae9efe544..dc43007dca 100644
--- a/indra/newview/app_settings/shaders/class1/deferred/pbrterrainUtilF.glsl
+++ b/indra/newview/app_settings/shaders/class1/deferred/pbrterrainUtilF.glsl
@@ -51,7 +51,12 @@
 #define TERRAIN_PBR_DETAIL_NORMAL -2
 #define TERRAIN_PBR_DETAIL_METALLIC_ROUGHNESS -3
 
+#define TERRAIN_PAINT_TYPE_HEIGHTMAP_WITH_NOISE 0
+#define TERRAIN_PAINT_TYPE_PBR_PAINTMAP 1
+
+#if TERRAIN_PLANAR_TEXTURE_SAMPLE_COUNT == 3
 in vec3 vary_vertex_normal;
+#endif
 
 vec3 srgb_to_linear(vec3 c);
 
@@ -202,6 +207,45 @@ TerrainMix get_terrain_mix_weights(float alpha1, float alpha2, float alphaFinal)
     return tm;
 }
 
+// A paintmap weight applier for 4 swatches. The input saves a channel by not
+// storing swatch 1, and assuming the weights of the 4 swatches add to 1.
+// The components of weight3 should be between 0 and 1
+// The sum of the components of weight3 should be between 0 and 1
+TerrainMix get_terrain_usage_from_weight3(vec3 weight3)
+{
+    // These steps ensure the output weights add to between 0 and 1
+    weight3.xyz = max(vec3(0.0), weight3.xyz);
+    weight3.xyz /= max(1.0, weight3.x + weight3.y + weight3.z);
+
+    TerrainMix tm;
+
+    // Extract the first weight from the other weights
+    tm.weight.x = 1.0 - (weight3.x + weight3.y + weight3.z);
+    tm.weight.yzw = weight3.xyz;
+    ivec4 usage = max(ivec4(0), ivec4(ceil(tm.weight)));
+
+    tm.type = (usage.x * MIX_X) |
+              (usage.y * MIX_Y) |
+              (usage.z * MIX_Z) |
+              (usage.w * MIX_W);
+    return tm;
+}
+
+// Inverse of get_terrain_usage_from_weight3, excluding usage flags
+// The components of weight should be between 0 and 1
+// The sum of the components of weight should be 1
+vec3 get_weight3_from_terrain_weight(vec4 weight)
+{
+    // These steps ensure the input weights add to 1
+    weight = max(vec4(0.0), weight);
+    weight.x += 1.0 - sign(weight.x + weight.y + weight.z + weight.w);
+    weight /= weight.x + weight.y + weight.z + weight.w;
+
+    // Then return the input weights with the first weight truncated
+    return weight.yzw;
+}
+
+#if TERRAIN_PLANAR_TEXTURE_SAMPLE_COUNT == 3
 TerrainTriplanar _t_triplanar()
 {
     float sharpness = TERRAIN_TRIPLANAR_BLEND_FACTOR;
@@ -219,6 +263,8 @@ TerrainTriplanar _t_triplanar()
               ((usage.z) * SAMPLE_Z);
     return tw;
 }
+#endif
+
 
 // Assume weights add to 1
 float terrain_mix(TerrainMix tm, vec4 tms4)
@@ -256,11 +302,12 @@ vec3 _t_normal_post_1(vec3 vNt0, float sign_or_zero)
 }
 
 // Triplanar-specific normal texture fixes
-vec3 _t_normal_post_x(vec3 vNt0)
+vec3 _t_normal_post_x(vec3 vNt0, float tangent_sign)
 {
     vec3 vNt_x = _t_normal_post_1(vNt0, sign(vary_vertex_normal.x));
     // *HACK: Transform normals according to orientation of the UVs
     vNt_x.xy = vec2(-vNt_x.y, vNt_x.x);
+    vNt_x.xy *= tangent_sign;
     return vNt_x;
 }
 vec3 _t_normal_post_y(vec3 vNt0)
@@ -285,6 +332,7 @@ PBRMix terrain_sample_pbr(
 #endif
 #if (TERRAIN_PBR_DETAIL >= TERRAIN_PBR_DETAIL_NORMAL)
     , sampler2D tex_vNt
+    , float tangent_sign
 #endif
 #if (TERRAIN_PBR_DETAIL >= TERRAIN_PBR_DETAIL_EMISSIVE)
     , sampler2D tex_emissive
@@ -314,7 +362,7 @@ PBRMix terrain_sample_pbr(
             );
 #if (TERRAIN_PBR_DETAIL >= TERRAIN_PBR_DETAIL_NORMAL)
         // Triplanar-specific normal texture fix
-        mix_x.vNt = _t_normal_post_x(mix_x.vNt);
+        mix_x.vNt = _t_normal_post_x(mix_x.vNt, tangent_sign);
 #endif
         mix = mix_pbr(mix, mix_x, tw.weight.x);
         break;
@@ -420,6 +468,9 @@ PBRMix terrain_sample_and_multiply_pbr(
 #endif
 #if (TERRAIN_PBR_DETAIL >= TERRAIN_PBR_DETAIL_NORMAL)
     , sampler2D tex_vNt
+#if TERRAIN_PLANAR_TEXTURE_SAMPLE_COUNT == 3
+    , float tangent_sign
+#endif
 #endif
 #if (TERRAIN_PBR_DETAIL >= TERRAIN_PBR_DETAIL_EMISSIVE)
     , sampler2D tex_emissive
@@ -446,6 +497,9 @@ PBRMix terrain_sample_and_multiply_pbr(
 #endif
 #if (TERRAIN_PBR_DETAIL >= TERRAIN_PBR_DETAIL_NORMAL)
         , tex_vNt
+#if TERRAIN_PLANAR_TEXTURE_SAMPLE_COUNT == 3
+        , tangent_sign
+#endif
 #endif
 #if (TERRAIN_PBR_DETAIL >= TERRAIN_PBR_DETAIL_EMISSIVE)
         , tex_emissive
diff --git a/indra/newview/app_settings/shaders/class1/deferred/pbrterrainV.glsl b/indra/newview/app_settings/shaders/class1/deferred/pbrterrainV.glsl
index f8e826bbdb..6791a22a76 100644
--- a/indra/newview/app_settings/shaders/class1/deferred/pbrterrainV.glsl
+++ b/indra/newview/app_settings/shaders/class1/deferred/pbrterrainV.glsl
@@ -28,31 +28,47 @@
 #define TERRAIN_PBR_DETAIL_NORMAL -2
 #define TERRAIN_PBR_DETAIL_METALLIC_ROUGHNESS -3
 
+#define TERRAIN_PAINT_TYPE_HEIGHTMAP_WITH_NOISE 0
+#define TERRAIN_PAINT_TYPE_PBR_PAINTMAP 1
+
 uniform mat3 normal_matrix;
 uniform mat4 texture_matrix0;
 uniform mat4 modelview_matrix;
 uniform mat4 modelview_projection_matrix;
+#if TERRAIN_PAINT_TYPE == TERRAIN_PAINT_TYPE_PBR_PAINTMAP
+uniform float region_scale;
+#endif
 
 in vec3 position;
 in vec3 normal;
 in vec4 tangent;
 in vec4 diffuse_color;
+#if TERRAIN_PAINT_TYPE == TERRAIN_PAINT_TYPE_HEIGHTMAP_WITH_NOISE
 in vec2 texcoord1;
+#endif
 
-out vec3 vary_vertex_normal; // Used by pbrterrainUtilF.glsl
+out vec3 vary_position;
 out vec3 vary_normal;
+#if TERRAIN_PLANAR_TEXTURE_SAMPLE_COUNT == 3
+out vec3 vary_vertex_normal; // Used by pbrterrainUtilF.glsl
+#endif
 #if (TERRAIN_PBR_DETAIL >= TERRAIN_PBR_DETAIL_NORMAL)
 out vec3 vary_tangents[4];
-flat out float vary_sign;
+flat out float vary_signs[4];
 #endif
+
+// vary_texcoord* are used for terrain composition, vary_coords are used for terrain UVs
+#if TERRAIN_PAINT_TYPE == TERRAIN_PAINT_TYPE_HEIGHTMAP_WITH_NOISE
 out vec4 vary_texcoord0;
 out vec4 vary_texcoord1;
+#elif TERRAIN_PAINT_TYPE == TERRAIN_PAINT_TYPE_PBR_PAINTMAP
+out vec2 vary_texcoord;
+#endif
 #if TERRAIN_PLANAR_TEXTURE_SAMPLE_COUNT == 3
 out vec4[10] vary_coords;
 #elif TERRAIN_PLANAR_TEXTURE_SAMPLE_COUNT == 1
 out vec4[2] vary_coords;
 #endif
-out vec3 vary_position;
 
 // *HACK: Each material uses only one texture transform, but the KHR texture
 // transform spec allows handling texture transforms separately for each
@@ -60,7 +76,7 @@ out vec3 vary_position;
 uniform vec4[5] terrain_texture_transforms;
 
 vec2 terrain_texture_transform(vec2 vertex_texcoord, vec4[2] khr_gltf_transform);
-vec3 terrain_tangent_space_transform(vec4 vertex_tangent, vec3 vertex_normal, vec4[2] khr_gltf_transform);
+vec4 terrain_tangent_space_transform(vec4 vertex_tangent, vec3 vertex_normal, vec4[2] khr_gltf_transform);
 
 void main()
 {
@@ -69,43 +85,52 @@ void main()
     vary_position = (modelview_matrix*vec4(position.xyz, 1.0)).xyz;
 
     vec3 n = normal_matrix * normal;
+#if TERRAIN_PLANAR_TEXTURE_SAMPLE_COUNT == 3
     vary_vertex_normal = normal;
+#endif
     vec3 t = normal_matrix * tangent.xyz;
 
 #if (TERRAIN_PBR_DETAIL >= TERRAIN_PBR_DETAIL_NORMAL)
     {
         vec4[2] ttt;
+        vec4 transformed_tangent;
         // material 1
         ttt[0].xyz = terrain_texture_transforms[0].xyz;
         ttt[1].x = terrain_texture_transforms[0].w;
         ttt[1].y = terrain_texture_transforms[1].x;
-        vary_tangents[0] = normalize(terrain_tangent_space_transform(vec4(t, tangent.w), n, ttt));
+        transformed_tangent = terrain_tangent_space_transform(vec4(t, tangent.w), n, ttt);
+        vary_tangents[0] = normalize(transformed_tangent.xyz);
+        vary_signs[0] = transformed_tangent.w;
         // material 2
         ttt[0].xyz = terrain_texture_transforms[1].yzw;
         ttt[1].xy = terrain_texture_transforms[2].xy;
-        vary_tangents[1] = normalize(terrain_tangent_space_transform(vec4(t, tangent.w), n, ttt));
+        transformed_tangent = terrain_tangent_space_transform(vec4(t, tangent.w), n, ttt);
+        vary_tangents[1] = normalize(transformed_tangent.xyz);
+        vary_signs[1] = transformed_tangent.w;
         // material 3
         ttt[0].xy = terrain_texture_transforms[2].zw;
         ttt[0].z = terrain_texture_transforms[3].x;
         ttt[1].xy = terrain_texture_transforms[3].yz;
-        vary_tangents[2] = normalize(terrain_tangent_space_transform(vec4(t, tangent.w), n, ttt));
+        transformed_tangent = terrain_tangent_space_transform(vec4(t, tangent.w), n, ttt);
+        vary_tangents[2] = normalize(transformed_tangent.xyz);
+        vary_signs[2] = transformed_tangent.w;
         // material 4
         ttt[0].x = terrain_texture_transforms[3].w;
         ttt[0].yz = terrain_texture_transforms[4].xy;
         ttt[1].xy = terrain_texture_transforms[4].zw;
-        vary_tangents[3] = normalize(terrain_tangent_space_transform(vec4(t, tangent.w), n, ttt));
+        transformed_tangent = terrain_tangent_space_transform(vec4(t, tangent.w), n, ttt);
+        vary_tangents[3] = normalize(transformed_tangent.xyz);
+        vary_signs[3] = transformed_tangent.w;
     }
-
-    vary_sign = tangent.w;
 #endif
     vary_normal = normalize(n);
 
     // Transform and pass tex coords
     {
         vec4[2] ttt;
+#define transform_xy()             terrain_texture_transform(position.xy,               ttt)
 #if TERRAIN_PLANAR_TEXTURE_SAMPLE_COUNT == 3
 // Don't care about upside-down (transform_xy_flipped())
-#define transform_xy()             terrain_texture_transform(position.xy,               ttt)
 #define transform_yz()             terrain_texture_transform(position.yz,               ttt)
 #define transform_negx_z()         terrain_texture_transform(position.xz * vec2(-1, 1), ttt)
 #define transform_yz_flipped()     terrain_texture_transform(position.yz * vec2(-1, 1), ttt)
@@ -150,26 +175,30 @@ void main()
         ttt[0].xyz = terrain_texture_transforms[0].xyz;
         ttt[1].x = terrain_texture_transforms[0].w;
         ttt[1].y = terrain_texture_transforms[1].x;
-        vary_coords[0].xy = terrain_texture_transform(position.xy, ttt);
+        vary_coords[0].xy = transform_xy();
         // material 2
         ttt[0].xyz = terrain_texture_transforms[1].yzw;
         ttt[1].xy = terrain_texture_transforms[2].xy;
-        vary_coords[0].zw = terrain_texture_transform(position.xy, ttt);
+        vary_coords[0].zw = transform_xy();
         // material 3
         ttt[0].xy = terrain_texture_transforms[2].zw;
         ttt[0].z = terrain_texture_transforms[3].x;
         ttt[1].xy = terrain_texture_transforms[3].yz;
-        vary_coords[1].xy = terrain_texture_transform(position.xy, ttt);
+        vary_coords[1].xy = transform_xy();
         // material 4
         ttt[0].x = terrain_texture_transforms[3].w;
         ttt[0].yz = terrain_texture_transforms[4].xy;
         ttt[1].xy = terrain_texture_transforms[4].zw;
-        vary_coords[1].zw = terrain_texture_transform(position.xy, ttt);
+        vary_coords[1].zw = transform_xy();
 #endif
     }
 
-    vec4 tc = vec4(texcoord1,0,1);
+#if TERRAIN_PAINT_TYPE == TERRAIN_PAINT_TYPE_HEIGHTMAP_WITH_NOISE
+    vec2 tc = texcoord1.xy;
     vary_texcoord0.zw = tc.xy;
     vary_texcoord1.xy = tc.xy-vec2(2.0, 0.0);
     vary_texcoord1.zw = tc.xy-vec2(1.0, 0.0);
+#elif TERRAIN_PAINT_TYPE == TERRAIN_PAINT_TYPE_PBR_PAINTMAP
+    vary_texcoord = position.xy / region_scale;
+#endif
 }
diff --git a/indra/newview/app_settings/shaders/class1/deferred/postDeferredGammaCorrect.glsl b/indra/newview/app_settings/shaders/class1/deferred/postDeferredGammaCorrect.glsl
index a0eb6cfbb8..befd2ae6da 100644
--- a/indra/newview/app_settings/shaders/class1/deferred/postDeferredGammaCorrect.glsl
+++ b/indra/newview/app_settings/shaders/class1/deferred/postDeferredGammaCorrect.glsl
@@ -28,141 +28,13 @@
 out vec4 frag_color;
 
 uniform sampler2D diffuseRect;
-uniform sampler2D exposureMap;
 
+uniform float gamma;
 uniform vec2 screen_res;
 in vec2 vary_fragcoord;
 
 vec3 linear_to_srgb(vec3 cl);
 
-//===============================================================
-// tone mapping taken from Khronos sample implementation
-//===============================================================
-
-// sRGB => XYZ => D65_2_D60 => AP1 => RRT_SAT
-const mat3 ACESInputMat = mat3
-(
-    0.59719, 0.07600, 0.02840,
-    0.35458, 0.90834, 0.13383,
-    0.04823, 0.01566, 0.83777
-);
-
-
-// ODT_SAT => XYZ => D60_2_D65 => sRGB
-const mat3 ACESOutputMat = mat3
-(
-    1.60475, -0.10208, -0.00327,
-    -0.53108,  1.10813, -0.07276,
-    -0.07367, -0.00605,  1.07602
-);
-
-// ACES tone map (faster approximation)
-// see: https://knarkowicz.wordpress.com/2016/01/06/aces-filmic-tone-mapping-curve/
-vec3 toneMapACES_Narkowicz(vec3 color)
-{
-    const float A = 2.51;
-    const float B = 0.03;
-    const float C = 2.43;
-    const float D = 0.59;
-    const float E = 0.14;
-    return clamp((color * (A * color + B)) / (color * (C * color + D) + E), 0.0, 1.0);
-}
-
-
-// ACES filmic tone map approximation
-// see https://github.com/TheRealMJP/BakingLab/blob/master/BakingLab/ACES.hlsl
-vec3 RRTAndODTFit(vec3 color)
-{
-    vec3 a = color * (color + 0.0245786) - 0.000090537;
-    vec3 b = color * (0.983729 * color + 0.4329510) + 0.238081;
-    return a / b;
-}
-
-
-// tone mapping
-vec3 toneMapACES_Hill(vec3 color)
-{
-    color = ACESInputMat * color;
-
-    // Apply RRT and ODT
-    color = RRTAndODTFit(color);
-
-    color = ACESOutputMat * color;
-
-    // Clamp to [0, 1]
-    color = clamp(color, 0.0, 1.0);
-
-    return color;
-}
-
-uniform float exposure;
-uniform float gamma;
-uniform float aces_mix;
-
-vec3 toneMap(vec3 color)
-{
-#ifndef NO_POST
-    float exp_scale = texture(exposureMap, vec2(0.5,0.5)).r;
-
-    color *= exposure * exp_scale;
-
-    // mix ACES and Linear here as a compromise to avoid over-darkening legacy content
-    color = mix(toneMapACES_Hill(color), color, aces_mix);
-#endif
-
-    return color;
-}
-
-//===============================================================
-
-//=================================
-// borrowed noise from:
-//  <https://www.shadertoy.com/view/4dS3Wd>
-//  By Morgan McGuire @morgan3d, http://graphicscodex.com
-//
-float hash(float n) { return fract(sin(n) * 1e4); }
-float hash(vec2 p) { return fract(1e4 * sin(17.0 * p.x + p.y * 0.1) * (0.1 + abs(sin(p.y * 13.0 + p.x)))); }
-
-float noise(float x) {
-    float i = floor(x);
-    float f = fract(x);
-    float u = f * f * (3.0 - 2.0 * f);
-    return mix(hash(i), hash(i + 1.0), u);
-}
-
-float noise(vec2 x) {
-    vec2 i = floor(x);
-    vec2 f = fract(x);
-
-    // Four corners in 2D of a tile
-    float a = hash(i);
-    float b = hash(i + vec2(1.0, 0.0));
-    float c = hash(i + vec2(0.0, 1.0));
-    float d = hash(i + vec2(1.0, 1.0));
-
-    // Simple 2D lerp using smoothstep envelope between the values.
-    // return vec3(mix(mix(a, b, smoothstep(0.0, 1.0, f.x)),
-    //          mix(c, d, smoothstep(0.0, 1.0, f.x)),
-    //          smoothstep(0.0, 1.0, f.y)));
-
-    // Same code, with the clamps in smoothstep and common subexpressions
-    // optimized away.
-    vec2 u = f * f * (3.0 - 2.0 * f);
-    return mix(a, b, u.x) + (c - a) * u.y * (1.0 - u.x) + (d - b) * u.x * u.y;
-}
-
-//=============================
-
-void debugExposure(inout vec3 color)
-{
-    float exp_scale = texture(exposureMap, vec2(0.5,0.5)).r;
-    exp_scale *= 0.5;
-    if (abs(vary_fragcoord.y-exp_scale) < 0.01 && vary_fragcoord.x < 0.1)
-    {
-        color = vec3(1,0,0);
-    }
-}
-
 vec3 legacyGamma(vec3 color)
 {
     vec3 c = 1. - clamp(color, vec3(0.), vec3(1.));
@@ -175,23 +47,12 @@ void main()
 {
     //this is the one of the rare spots where diffuseRect contains linear color values (not sRGB)
     vec4 diff = texture(diffuseRect, vary_fragcoord);
+    diff.rgb = linear_to_srgb(diff.rgb);
 
 #ifdef LEGACY_GAMMA
-    diff.rgb = linear_to_srgb(diff.rgb);
     diff.rgb = legacyGamma(diff.rgb);
-#else
-#ifndef NO_POST
-    diff.rgb = toneMap(diff.rgb);
 #endif
-    diff.rgb = linear_to_srgb(diff.rgb);
-#endif
-
-    vec2 tc = vary_fragcoord.xy*screen_res*4.0;
-    vec3 seed = (diff.rgb+vec3(1.0))*vec3(tc.xy, tc.x+tc.y);
-    vec3 nz = vec3(noise(seed.rg), noise(seed.gb), noise(seed.rb));
-    diff.rgb += nz*0.003;
 
-    //debugExposure(diff.rgb);
     frag_color = max(diff, vec4(0));
 }
 
diff --git a/indra/newview/app_settings/shaders/class1/deferred/postDeferredNoDoFF.glsl b/indra/newview/app_settings/shaders/class1/deferred/postDeferredNoDoFF.glsl
index 07384ebe9b..32b0a1ac8e 100644
--- a/indra/newview/app_settings/shaders/class1/deferred/postDeferredNoDoFF.glsl
+++ b/indra/newview/app_settings/shaders/class1/deferred/postDeferredNoDoFF.glsl
@@ -33,10 +33,57 @@ uniform sampler2D depthMap;
 uniform vec2 screen_res;
 in vec2 vary_fragcoord;
 
+//=================================
+// borrowed noise from:
+//  <https://www.shadertoy.com/view/4dS3Wd>
+//  By Morgan McGuire @morgan3d, http://graphicscodex.com
+//
+float hash(float n) { return fract(sin(n) * 1e4); }
+float hash(vec2 p) { return fract(1e4 * sin(17.0 * p.x + p.y * 0.1) * (0.1 + abs(sin(p.y * 13.0 + p.x)))); }
+
+float noise(float x) {
+    float i = floor(x);
+    float f = fract(x);
+    float u = f * f * (3.0 - 2.0 * f);
+    return mix(hash(i), hash(i + 1.0), u);
+}
+
+float noise(vec2 x) {
+    vec2 i = floor(x);
+    vec2 f = fract(x);
+
+    // Four corners in 2D of a tile
+    float a = hash(i);
+    float b = hash(i + vec2(1.0, 0.0));
+    float c = hash(i + vec2(0.0, 1.0));
+    float d = hash(i + vec2(1.0, 1.0));
+
+    // Simple 2D lerp using smoothstep envelope between the values.
+    // return vec3(mix(mix(a, b, smoothstep(0.0, 1.0, f.x)),
+    //          mix(c, d, smoothstep(0.0, 1.0, f.x)),
+    //          smoothstep(0.0, 1.0, f.y)));
+
+    // Same code, with the clamps in smoothstep and common subexpressions
+    // optimized away.
+    vec2 u = f * f * (3.0 - 2.0 * f);
+    return mix(a, b, u.x) + (c - a) * u.y * (1.0 - u.x) + (d - b) * u.x * u.y;
+}
+
+//=============================
+
+
+
 void main()
 {
     vec4 diff = texture(diffuseRect, vary_fragcoord.xy);
 
+#ifdef HAS_NOISE
+    vec2 tc = vary_fragcoord.xy*screen_res*4.0;
+    vec3 seed = (diff.rgb+vec3(1.0))*vec3(tc.xy, tc.x+tc.y);
+    vec3 nz = vec3(noise(seed.rg), noise(seed.gb), noise(seed.rb));
+    diff.rgb += nz*0.003;
+#endif
+
     frag_color = diff;
 
     gl_FragDepth = texture(depthMap, vary_fragcoord.xy).r;
diff --git a/indra/newview/app_settings/shaders/class1/deferred/postDeferredTonemap.glsl b/indra/newview/app_settings/shaders/class1/deferred/postDeferredTonemap.glsl
new file mode 100644
index 0000000000..fc6d4d7727
--- /dev/null
+++ b/indra/newview/app_settings/shaders/class1/deferred/postDeferredTonemap.glsl
@@ -0,0 +1,178 @@
+/**
+ * @file postDeferredTonemap.glsl
+ *
+ * $LicenseInfo:firstyear=2024&license=viewerlgpl$
+ * Second Life Viewer Source Code
+ * Copyright (C) 2024, Linden Research, Inc.
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation;
+ * version 2.1 of the License only.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
+ *
+ * Linden Research, Inc., 945 Battery Street, San Francisco, CA  94111  USA
+ * $/LicenseInfo$
+ */
+
+/*[EXTRA_CODE_HERE]*/
+
+out vec4 frag_color;
+
+uniform sampler2D diffuseRect;
+uniform sampler2D exposureMap;
+
+uniform vec2 screen_res;
+in vec2 vary_fragcoord;
+
+vec3 linear_to_srgb(vec3 cl);
+
+//===============================================================
+// tone mapping taken from Khronos sample implementation
+//===============================================================
+
+// sRGB => XYZ => D65_2_D60 => AP1 => RRT_SAT
+const mat3 ACESInputMat = mat3
+(
+    0.59719, 0.07600, 0.02840,
+    0.35458, 0.90834, 0.13383,
+    0.04823, 0.01566, 0.83777
+);
+
+
+// ODT_SAT => XYZ => D60_2_D65 => sRGB
+const mat3 ACESOutputMat = mat3
+(
+    1.60475, -0.10208, -0.00327,
+    -0.53108,  1.10813, -0.07276,
+    -0.07367, -0.00605,  1.07602
+);
+
+// ACES tone map (faster approximation)
+// see: https://knarkowicz.wordpress.com/2016/01/06/aces-filmic-tone-mapping-curve/
+vec3 toneMapACES_Narkowicz(vec3 color)
+{
+    const float A = 2.51;
+    const float B = 0.03;
+    const float C = 2.43;
+    const float D = 0.59;
+    const float E = 0.14;
+    return clamp((color * (A * color + B)) / (color * (C * color + D) + E), 0.0, 1.0);
+}
+
+
+// ACES filmic tone map approximation
+// see https://github.com/TheRealMJP/BakingLab/blob/master/BakingLab/ACES.hlsl
+vec3 RRTAndODTFit(vec3 color)
+{
+    vec3 a = color * (color + 0.0245786) - 0.000090537;
+    vec3 b = color * (0.983729 * color + 0.4329510) + 0.238081;
+    return a / b;
+}
+
+
+// tone mapping
+vec3 toneMapACES_Hill(vec3 color)
+{
+    color = ACESInputMat * color;
+
+    // Apply RRT and ODT
+    color = RRTAndODTFit(color);
+
+    color = ACESOutputMat * color;
+
+    // Clamp to [0, 1]
+    color = clamp(color, 0.0, 1.0);
+
+    return color;
+}
+
+// Khronos Neutral tonemapping
+// https://github.com/KhronosGroup/ToneMapping/tree/main
+// Input color is non-negative and resides in the Linear Rec. 709 color space.
+// Output color is also Linear Rec. 709, but in the [0, 1] range.
+vec3 PBRNeutralToneMapping( vec3 color )
+{
+  const float startCompression = 0.8 - 0.04;
+  const float desaturation = 0.15;
+
+  float x = min(color.r, min(color.g, color.b));
+  float offset = x < 0.08 ? x - 6.25 * x * x : 0.04;
+  color -= offset;
+
+  float peak = max(color.r, max(color.g, color.b));
+  if (peak < startCompression) return color;
+
+  const float d = 1. - startCompression;
+  float newPeak = 1. - d * d / (peak + d - startCompression);
+  color *= newPeak / peak;
+
+  float g = 1. - 1. / (desaturation * (peak - newPeak) + 1.);
+  return mix(color, newPeak * vec3(1, 1, 1), g);
+}
+
+uniform float exposure;
+uniform float tonemap_mix;
+uniform int tonemap_type;
+
+vec3 toneMap(vec3 color)
+{
+#ifndef NO_POST
+    float exp_scale = texture(exposureMap, vec2(0.5,0.5)).r;
+
+    color *= exposure * exp_scale;
+
+    vec3 clamped_color = clamp(color.rgb, vec3(0.0), vec3(1.0));
+
+    switch(tonemap_type)
+    {
+    case 0:
+        color = PBRNeutralToneMapping(color);
+        break;
+    case 1:
+        color = toneMapACES_Hill(color);
+        break;
+    }
+
+    // mix tonemapped and linear here to provide adjustment
+    color = mix(clamped_color, color, tonemap_mix);
+#endif
+
+    return color;
+}
+
+//===============================================================
+
+void debugExposure(inout vec3 color)
+{
+    float exp_scale = texture(exposureMap, vec2(0.5,0.5)).r;
+    exp_scale *= 0.5;
+    if (abs(vary_fragcoord.y-exp_scale) < 0.01 && vary_fragcoord.x < 0.1)
+    {
+        color = vec3(1,0,0);
+    }
+}
+
+void main()
+{
+    //this is the one of the rare spots where diffuseRect contains linear color values (not sRGB)
+    vec4 diff = texture(diffuseRect, vary_fragcoord);
+
+#ifndef NO_POST
+    diff.rgb = toneMap(diff.rgb);
+#else
+    diff.rgb = clamp(diff.rgb, vec3(0.0), vec3(1.0));
+#endif
+
+    //debugExposure(diff.rgb);
+    frag_color = max(diff, vec4(0));
+}
+
diff --git a/indra/newview/app_settings/shaders/class1/deferred/textureUtilV.glsl b/indra/newview/app_settings/shaders/class1/deferred/textureUtilV.glsl
index 767416d564..c75a0e0d5d 100644
--- a/indra/newview/app_settings/shaders/class1/deferred/textureUtilV.glsl
+++ b/indra/newview/app_settings/shaders/class1/deferred/textureUtilV.glsl
@@ -94,36 +94,48 @@ vec2 terrain_texture_transform(vec2 vertex_texcoord, vec4[2] khr_gltf_transform)
 // Take the rotation only from both transforms and apply to the tangent. This
 // accounts for the change of the topology of the normal texture when a texture
 // rotation is applied to it.
+// In practice, this applies the inverse of the texture transform to the tangent.
+// It is effectively an inverse of the rotation
 // *HACK: Assume the imported GLTF model did not have both normal texture
 // transforms and tangent vertices. The use of this function is inconsistent
 // with the GLTF sample viewer when that is the case. See getNormalInfo in
 // https://raw.githubusercontent.com/KhronosGroup/glTF-Sample-Viewer/47a191931461a6f2e14de48d6da0f0eb6ec2d147/source/Renderer/shaders/material_info.glsl
 // We may want to account for this case during GLTF model import.
 // -Cosmic,2023-06-06
-vec3 tangent_space_transform(vec4 vertex_tangent, vec3 vertex_normal, vec4[2] khr_gltf_transform, mat4 sl_animation_transform)
+vec4 tangent_space_transform(vec4 vertex_tangent, vec3 vertex_normal, vec4[2] khr_gltf_transform, mat4 sl_animation_transform)
 {
-    vec2 weights = vec2(0, 1);
-
-    // Apply texture animation first to avoid shearing and other artifacts (rotation only)
-    mat2 sl_rot_scale;
-    sl_rot_scale[0][0] = sl_animation_transform[0][0];
-    sl_rot_scale[0][1] = sl_animation_transform[0][1];
-    sl_rot_scale[1][0] = sl_animation_transform[1][0];
-    sl_rot_scale[1][1] = sl_animation_transform[1][1];
-    weights = sl_rot_scale * weights;
-    // Remove scale
-    weights = normalize(weights);
-
-    // Convert to left-handed coordinate system
-    weights.y = -weights.y;
+    // Immediately convert to left-handed coordinate system, but it has no
+    // effect here because y is 0 ((1,0) -> (1,0))
+    vec2 weights = vec2(1, 0);
 
-    // Apply KHR_texture_transform (rotation only)
-    float khr_rotation = khr_gltf_transform[0].z;
+    // Apply inverse KHR_texture_transform (rotation and scale sign only)
+    float khr_rotation = -khr_gltf_transform[0].z;
     mat2 khr_rotation_mat = mat2(
         cos(khr_rotation),-sin(khr_rotation),
         sin(khr_rotation), cos(khr_rotation)
     );
     weights = khr_rotation_mat * weights;
+    vec2 khr_scale_sign = sign(khr_gltf_transform[0].xy);
+    weights *= khr_scale_sign.xy;
+
+    // *NOTE: Delay conversion to right-handed coordinate system here, to
+    // remove the need for computing the inverse of the SL texture animation
+    // matrix.
+
+    // Apply texture animation last to avoid shearing and other artifacts (rotation only)
+    mat2 inv_sl_rot_scale;
+    inv_sl_rot_scale[0][0] = sl_animation_transform[0][0];
+    inv_sl_rot_scale[0][1] = sl_animation_transform[0][1];
+    inv_sl_rot_scale[1][0] = sl_animation_transform[1][0];
+    inv_sl_rot_scale[1][1] = sl_animation_transform[1][1];
+    weights = inv_sl_rot_scale * weights;
+    // *NOTE: Scale to be removed later
+
+    // Set weights to default if 0 for some reason
+    weights.x += 1.0 - abs(sign(sign(weights.x) + (0.5 * sign(weights.y))));
+
+    // Remove scale from SL texture animation transform
+    weights = normalize(weights);
 
     // Convert back to right-handed coordinate system
     weights.y = -weights.y;
@@ -132,27 +144,41 @@ vec3 tangent_space_transform(vec4 vertex_tangent, vec3 vertex_normal, vec4[2] kh
     // from the normal and tangent, as seen in the fragment shader
     vec3 vertex_binormal = vertex_tangent.w * cross(vertex_normal, vertex_tangent.xyz);
 
-    return (weights.x * vertex_binormal.xyz) + (weights.y * vertex_tangent.xyz);
+    // An additional sign flip prevents the binormal from being flipped as a
+    // result of a propagation of the tangent sign during the cross product.
+    float sign_flip = khr_scale_sign.x * khr_scale_sign.y;
+    return vec4((weights.x * vertex_tangent.xyz) + (weights.y * vertex_binormal.xyz), vertex_tangent.w * sign_flip);
 }
 
 // Similar to tangent_space_transform but no texture animation support.
-vec3 terrain_tangent_space_transform(vec4 vertex_tangent, vec3 vertex_normal, vec4[2] khr_gltf_transform)
+vec4 terrain_tangent_space_transform(vec4 vertex_tangent, vec3 vertex_normal, vec4[2] khr_gltf_transform)
 {
-    // Immediately convert to left-handed coordinate system ((0,1) -> (0, -1))
-    vec2 weights = vec2(0, -1);
+    // Immediately convert to left-handed coordinate system, but it has no
+    // effect here because y is 0 ((1,0) -> (1,0))
+    vec2 weights = vec2(1, 0);
 
-    // Apply KHR_texture_transform (rotation only)
-    float khr_rotation = khr_gltf_transform[0].z;
+    // Apply inverse KHR_texture_transform (rotation and scale sign only)
+    float khr_rotation = -khr_gltf_transform[0].z;
     mat2 khr_rotation_mat = mat2(
         cos(khr_rotation),-sin(khr_rotation),
         sin(khr_rotation), cos(khr_rotation)
     );
     weights = khr_rotation_mat * weights;
+    vec2 khr_scale_sign = sign(khr_gltf_transform[0].xy);
+    weights *= khr_scale_sign.xy;
+
+    // Set weights to default if 0 for some reason
+    weights.x += 1.0 - abs(sign(sign(weights.x) + (0.5 * sign(weights.y))));
 
     // Convert back to right-handed coordinate system
     weights.y = -weights.y;
 
+    // Similar to the MikkTSpace-compatible method of extracting the binormal
+    // from the normal and tangent, as seen in the fragment shader
     vec3 vertex_binormal = vertex_tangent.w * cross(vertex_normal, vertex_tangent.xyz);
 
-    return (weights.x * vertex_binormal.xyz) + (weights.y * vertex_tangent.xyz);
+    // An additional sign flip prevents the binormal from being flipped as a
+    // result of a propagation of the tangent sign during the cross product.
+    float sign_flip = khr_scale_sign.x * khr_scale_sign.y;
+    return vec4((weights.x * vertex_tangent.xyz) + (weights.y * vertex_binormal.xyz), vertex_tangent.w * sign_flip);
 }