From 637dc1f5a98e1c6cc6eecfb476ab1c20dca0c6aa Mon Sep 17 00:00:00 2001 From: Nicky Date: Fri, 22 Apr 2016 12:59:41 +0200 Subject: Windows x64: 64 bit implies SSE2, handle this accordingly when detecting if SSE2 is enabled. (transplanted from 93492b84cb752dc79c74d9667f11edd76ace8f0b) --- indra/llmath/llsimdmath.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'indra/llmath') diff --git a/indra/llmath/llsimdmath.h b/indra/llmath/llsimdmath.h index cebd2ace7d..9f078ec1ef 100644 --- a/indra/llmath/llsimdmath.h +++ b/indra/llmath/llsimdmath.h @@ -31,7 +31,7 @@ #error "Please include llmath.h before this file." #endif -#if ( ( LL_DARWIN || LL_LINUX ) && !(__SSE2__) ) || ( LL_WINDOWS && ( _M_IX86_FP < 2 ) ) +#if ( ( LL_DARWIN || LL_LINUX ) && !(__SSE2__) ) || ( LL_WINDOWS && ( _M_IX86_FP < 2 && !_M_AMD64 ) ) #error SSE2 not enabled. LLVector4a and related class will not compile. #endif -- cgit v1.3 From 7640c2fb446846005a191abb62d3cce1a64d2a6e Mon Sep 17 00:00:00 2001 From: Nicky Date: Fri, 22 Apr 2016 12:59:59 +0200 Subject: Windows x64: Cannot use inline assembly. (transplanted from 0b621f8a1ee707527325eb70e59ef02c63e2bd10) --- indra/llmath/llmath.h | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'indra/llmath') diff --git a/indra/llmath/llmath.h b/indra/llmath/llmath.h index 93b9f22b25..b66a3c63d6 100644 --- a/indra/llmath/llmath.h +++ b/indra/llmath/llmath.h @@ -153,7 +153,7 @@ inline F64 llabs(const F64 a) inline S32 lltrunc( F32 f ) { -#if LL_WINDOWS && !defined( __INTEL_COMPILER ) +#if LL_WINDOWS && !defined( __INTEL_COMPILER ) && !defined( _M_AMD64 ) // Avoids changing the floating point control word. // Add or subtract 0.5 - epsilon and then round const static U32 zpfp[] = { 0xBEFFFFFF, 0x3EFFFFFF }; @@ -179,7 +179,7 @@ inline S32 lltrunc( F64 f ) inline S32 llfloor( F32 f ) { -#if LL_WINDOWS && !defined( __INTEL_COMPILER ) +#if LL_WINDOWS && !defined( __INTEL_COMPILER ) && !defined( _M_AMD64 ) // Avoids changing the floating point control word. // Accurate (unlike Stereopsis version) for all values between S32_MIN and S32_MAX and slightly faster than Stereopsis version. // Add -(0.5 - epsilon) and then round -- cgit v1.3 From e8aa2dd71fff7a39f2b03039b23afa8bdf804fcb Mon Sep 17 00:00:00 2001 From: Nicky Date: Fri, 22 Apr 2016 14:58:25 +0200 Subject: x64: Do not use a union of LLColor4U. Especially having the two pointer in there will blow up the struct to at least 8 byte, which will break VBO packing as this class needs to be 4 byte in size. (transplanted from 847df86d6b5daa69dcfc428df18876a9c1e8bef6) --- indra/llimage/llimage.cpp | 3 ++- indra/llmath/v4coloru.h | 42 ++++++++++++++++++++++++++++++++++-------- indra/newview/llface.cpp | 4 ++-- indra/newview/llnetmap.cpp | 6 +++--- indra/newview/llvosky.cpp | 2 +- indra/newview/llvosky.h | 4 ++-- 6 files changed, 44 insertions(+), 17 deletions(-) (limited to 'indra/llmath') diff --git a/indra/llimage/llimage.cpp b/indra/llimage/llimage.cpp index f71607096c..a6cbcc131e 100644 --- a/indra/llimage/llimage.cpp +++ b/indra/llimage/llimage.cpp @@ -1218,9 +1218,10 @@ void LLImageRaw::fill( const LLColor4U& color ) if( 4 == getComponents() ) { U32* data = (U32*) getData(); + U32 rgbaColor = color.asRGBA(); for( S32 i = 0; i < pixels; i++ ) { - data[i] = color.mAll; + data[ i ] = rgbaColor; } } else diff --git a/indra/llmath/v4coloru.h b/indra/llmath/v4coloru.h index fddad34978..31ae3e3c1a 100644 --- a/indra/llmath/v4coloru.h +++ b/indra/llmath/v4coloru.h @@ -47,14 +47,7 @@ class LLColor4U { public: - union - { - U8 mV[LENGTHOFCOLOR4U]; - U32 mAll; - LLColor4* mSources; - LLColor4U* mSourcesU; - }; - + U8 mV[LENGTHOFCOLOR4U]; LLColor4U(); // Initializes LLColor4U to (0, 0, 0, 1) LLColor4U(U8 r, U8 g, U8 b); // Initializes LLColor4U to (r, g, b, 1) @@ -132,6 +125,9 @@ public: return LLColor4(*this); } + U32 asRGBA() const; + void fromRGBA( U32 aVal ); + static LLColor4U white; static LLColor4U black; static LLColor4U red; @@ -565,6 +561,36 @@ void LLColor4U::setVecScaleClamp(const LLColor3& color) mV[3] = 255; } +inline U32 LLColor4U::asRGBA() const +{ + U32 nRet( 0 ); + + // Little endian: values are swapped in memory. The original code access the array like a U32, so we need to swap here + + nRet |= mV[ 3 ]; + nRet <<= 8; + nRet |= mV[ 2 ]; + nRet <<= 8; + nRet |= mV[ 1 ]; + nRet <<= 8; + nRet |= mV[ 0 ]; + + return nRet; +} + +inline void LLColor4U::fromRGBA( U32 aVal ) +{ + // Little endian: values are swapped in memory. The original code access the array like a U32, so we need to swap here + + mV[ 0 ] = aVal & 0xFF; + aVal >>= 8; + mV[ 1 ] = aVal & 0xFF; + aVal >>= 8; + mV[ 2 ] = aVal & 0xFF; + aVal >>= 8; + mV[ 3 ] = aVal & 0xFF; +} + #endif diff --git a/indra/newview/llface.cpp b/indra/newview/llface.cpp index de349a03d4..481c66aaf5 100644 --- a/indra/newview/llface.cpp +++ b/indra/newview/llface.cpp @@ -2132,7 +2132,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, LLVector4a src; U32 vec[4]; - vec[0] = vec[1] = vec[2] = vec[3] = color.mAll; + vec[0] = vec[1] = vec[2] = vec[3] = color.asRGBA(); src.loadua((F32*) vec); @@ -2168,7 +2168,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, LLColor4U glow4u = LLColor4U(0,0,0,glow); - U32 glow32 = glow4u.mAll; + U32 glow32 = glow4u.asRGBA(); U32 vec[4]; vec[0] = vec[1] = vec[2] = vec[3] = glow32; diff --git a/indra/newview/llnetmap.cpp b/indra/newview/llnetmap.cpp index 5fc73c67d1..72faa5a9e7 100644 --- a/indra/newview/llnetmap.cpp +++ b/indra/newview/llnetmap.cpp @@ -735,7 +735,7 @@ void LLNetMap::renderPoint(const LLVector3 &pos_local, const LLColor4U &color, continue; } S32 offset = px + py * image_width; - ((U32*)datap)[offset] = color.mAll; + ((U32*)datap)[offset] = color.asRGBA(); } // top line @@ -748,7 +748,7 @@ void LLNetMap::renderPoint(const LLVector3 &pos_local, const LLColor4U &color, continue; } S32 offset = px + py * image_width; - ((U32*)datap)[offset] = color.mAll; + ((U32*)datap)[offset] = color.asRGBA(); } } else @@ -770,7 +770,7 @@ void LLNetMap::renderPoint(const LLVector3 &pos_local, const LLColor4U &color, continue; } S32 offset = p_x + p_y * image_width; - ((U32*)datap)[offset] = color.mAll; + ((U32*)datap)[offset] = color.asRGBA(); } } } diff --git a/indra/newview/llvosky.cpp b/indra/newview/llvosky.cpp index 4dab213fa0..6b4a450e6f 100644 --- a/indra/newview/llvosky.cpp +++ b/indra/newview/llvosky.cpp @@ -283,7 +283,7 @@ void LLSkyTex::create(const F32 brightness) S32 offset = basic_offset * sComponents; U32* pix = (U32*)(data + offset); LLColor4U temp = LLColor4U(mSkyData[basic_offset]); - *pix = temp.mAll; + *pix = temp.asRGBA(); } } createGLImage(sCurrent); diff --git a/indra/newview/llvosky.h b/indra/newview/llvosky.h index ee8e91fb71..9cfb9773bd 100644 --- a/indra/newview/llvosky.h +++ b/indra/newview/llvosky.h @@ -171,7 +171,7 @@ protected: { S32 offset = (i * sResolution + j) * sComponents; U32* pix = (U32*) &(mImageRaw[sCurrent]->getData()[offset]); - *pix = col.mAll; + *pix = col.asRGBA(); } LLColor4U getPixel(const S32 i, const S32 j) @@ -179,7 +179,7 @@ protected: LLColor4U col; S32 offset = (i * sResolution + j) * sComponents; U32* pix = (U32*) &(mImageRaw[sCurrent]->getData()[offset]); - col.mAll = *pix; + col.fromRGBA( *pix ); return col; } -- cgit v1.3 From 944c497090b09e30377a67d60049c641b5e10e55 Mon Sep 17 00:00:00 2001 From: Nicky Date: Sun, 24 Apr 2016 12:55:50 +0200 Subject: Code stylistics. (transplanted from 570ea799407270069974021eca3a5056d6908f58) --- indra/llmath/v4coloru.h | 12 +----------- 1 file changed, 1 insertion(+), 11 deletions(-) (limited to 'indra/llmath') diff --git a/indra/llmath/v4coloru.h b/indra/llmath/v4coloru.h index 31ae3e3c1a..704ce852d9 100644 --- a/indra/llmath/v4coloru.h +++ b/indra/llmath/v4coloru.h @@ -563,19 +563,9 @@ void LLColor4U::setVecScaleClamp(const LLColor3& color) inline U32 LLColor4U::asRGBA() const { - U32 nRet( 0 ); - // Little endian: values are swapped in memory. The original code access the array like a U32, so we need to swap here - nRet |= mV[ 3 ]; - nRet <<= 8; - nRet |= mV[ 2 ]; - nRet <<= 8; - nRet |= mV[ 1 ]; - nRet <<= 8; - nRet |= mV[ 0 ]; - - return nRet; + return (mV[3] << 24) | (mV[2] << 16) | (mV[1] << 8) | mV[0]; } inline void LLColor4U::fromRGBA( U32 aVal ) -- cgit v1.3 From f5e983962703b5cb39278048b1c35e712b2b2263 Mon Sep 17 00:00:00 2001 From: Nat Goodspeed Date: Wed, 16 Nov 2016 15:39:00 -0500 Subject: DRTVWR-418: Replace preprocessor tests for Windows-specific _M_AMD64 with tests on ADDRESS_SIZE, which is now set on the compiler command line. --- indra/llcommon/llpreprocessor.h | 2 +- indra/llmath/llmath.h | 4 ++-- indra/llmath/llsimdmath.h | 2 +- 3 files changed, 4 insertions(+), 4 deletions(-) (limited to 'indra/llmath') diff --git a/indra/llcommon/llpreprocessor.h b/indra/llcommon/llpreprocessor.h index 7c277c2bed..3698d9db44 100644 --- a/indra/llcommon/llpreprocessor.h +++ b/indra/llcommon/llpreprocessor.h @@ -139,7 +139,7 @@ #pragma warning (disable : 4180) // qualifier applied to function type has no meaning; ignored //#pragma warning( disable : 4284 ) // silly MS warning deep inside their include file -#ifdef _M_AMD64 +#if ADDRESS_SIZE == 64 // That one is all over the place for x64 builds. #pragma warning( disable : 4267 ) // 'var' : conversion from 'size_t' to 'type', possible loss of data) #endif diff --git a/indra/llmath/llmath.h b/indra/llmath/llmath.h index b66a3c63d6..92a48df53c 100644 --- a/indra/llmath/llmath.h +++ b/indra/llmath/llmath.h @@ -153,7 +153,7 @@ inline F64 llabs(const F64 a) inline S32 lltrunc( F32 f ) { -#if LL_WINDOWS && !defined( __INTEL_COMPILER ) && !defined( _M_AMD64 ) +#if LL_WINDOWS && !defined( __INTEL_COMPILER ) && (ADDRESS_SIZE == 32) // Avoids changing the floating point control word. // Add or subtract 0.5 - epsilon and then round const static U32 zpfp[] = { 0xBEFFFFFF, 0x3EFFFFFF }; @@ -179,7 +179,7 @@ inline S32 lltrunc( F64 f ) inline S32 llfloor( F32 f ) { -#if LL_WINDOWS && !defined( __INTEL_COMPILER ) && !defined( _M_AMD64 ) +#if LL_WINDOWS && !defined( __INTEL_COMPILER ) && (ADDRESS_SIZE == 32) // Avoids changing the floating point control word. // Accurate (unlike Stereopsis version) for all values between S32_MIN and S32_MAX and slightly faster than Stereopsis version. // Add -(0.5 - epsilon) and then round diff --git a/indra/llmath/llsimdmath.h b/indra/llmath/llsimdmath.h index 9f078ec1ef..54a275633f 100644 --- a/indra/llmath/llsimdmath.h +++ b/indra/llmath/llsimdmath.h @@ -31,7 +31,7 @@ #error "Please include llmath.h before this file." #endif -#if ( ( LL_DARWIN || LL_LINUX ) && !(__SSE2__) ) || ( LL_WINDOWS && ( _M_IX86_FP < 2 && !_M_AMD64 ) ) +#if ( ( LL_DARWIN || LL_LINUX ) && !(__SSE2__) ) || ( LL_WINDOWS && ( _M_IX86_FP < 2 && ADDRESS_SIZE == 32 ) ) #error SSE2 not enabled. LLVector4a and related class will not compile. #endif -- cgit v1.3