diff options
Diffstat (limited to 'indra/llimage')
-rw-r--r-- | indra/llimage/llimage.h | 42 | ||||
-rw-r--r-- | indra/llimage/llimagefilter.cpp | 114 | ||||
-rw-r--r-- | indra/llimage/llimagej2c.cpp | 28 |
3 files changed, 89 insertions, 95 deletions
diff --git a/indra/llimage/llimage.h b/indra/llimage/llimage.h index 42eecbb97c..8b966b8ea3 100644 --- a/indra/llimage/llimage.h +++ b/indra/llimage/llimage.h @@ -32,37 +32,37 @@ #include "llpointer.h" #include "lltrace.h" -const S32 MIN_IMAGE_MIP = 2; // 4x4, only used for expand/contract power of 2 -const S32 MAX_IMAGE_MIP = 12; // 4096x4096 +constexpr S32 MIN_IMAGE_MIP = 2; // 4x4, only used for expand/contract power of 2 +constexpr S32 MAX_IMAGE_MIP = 12; // 4096x4096 // *TODO : Use MAX_IMAGE_MIP as max discard level and modify j2c management so that the number // of levels is read from the header's file, not inferred from its size. -const S32 MAX_DISCARD_LEVEL = 5; +constexpr S32 MAX_DISCARD_LEVEL = 5; // JPEG2000 size constraints // Those are declared here as they are germane to other image constraints used in the viewer // and declared right here. Some come from the JPEG2000 spec, some conventions specific to SL. -const S32 MAX_DECOMPOSITION_LEVELS = 32; // Number of decomposition levels cannot exceed 32 according to jpeg2000 spec -const S32 MIN_DECOMPOSITION_LEVELS = 5; // the SL viewer will *crash* trying to decode images with fewer than 5 decomposition levels (unless image is small that is) -const S32 MAX_PRECINCT_SIZE = 4096; // No reason to be bigger than MAX_IMAGE_SIZE -const S32 MIN_PRECINCT_SIZE = 4; // Can't be smaller than MIN_BLOCK_SIZE -const S32 MAX_BLOCK_SIZE = 64; // Max total block size is 4096, hence 64x64 when using square blocks -const S32 MIN_BLOCK_SIZE = 4; // Min block dim is 4 according to jpeg2000 spec -const S32 MIN_LAYER_SIZE = 2000; // Size of the first quality layer (after header). Must be > to FIRST_PACKET_SIZE!! -const S32 MAX_NB_LAYERS = 64; // Max number of layers we'll entertain in SL (practical limit) - -const S32 MIN_IMAGE_SIZE = (1<<MIN_IMAGE_MIP); // 4, only used for expand/contract power of 2 -const S32 MAX_IMAGE_SIZE = (1<<MAX_IMAGE_MIP); // 4096 -const S32 MIN_IMAGE_AREA = MIN_IMAGE_SIZE * MIN_IMAGE_SIZE; -const S32 MAX_IMAGE_AREA = MAX_IMAGE_SIZE * MAX_IMAGE_SIZE; -const S32 MAX_IMAGE_COMPONENTS = 8; -const S32 MAX_IMAGE_DATA_SIZE = MAX_IMAGE_AREA * MAX_IMAGE_COMPONENTS; //4096 * 4096 * 8 = 128 MB +constexpr S32 MAX_DECOMPOSITION_LEVELS = 32; // Number of decomposition levels cannot exceed 32 according to jpeg2000 spec +constexpr S32 MIN_DECOMPOSITION_LEVELS = 5; // the SL viewer will *crash* trying to decode images with fewer than 5 decomposition levels (unless image is small that is) +constexpr S32 MAX_PRECINCT_SIZE = 4096; // No reason to be bigger than MAX_IMAGE_SIZE +constexpr S32 MIN_PRECINCT_SIZE = 4; // Can't be smaller than MIN_BLOCK_SIZE +constexpr S32 MAX_BLOCK_SIZE = 64; // Max total block size is 4096, hence 64x64 when using square blocks +constexpr S32 MIN_BLOCK_SIZE = 4; // Min block dim is 4 according to jpeg2000 spec +constexpr S32 MIN_LAYER_SIZE = 2000; // Size of the first quality layer (after header). Must be > to FIRST_PACKET_SIZE!! +constexpr S32 MAX_NB_LAYERS = 64; // Max number of layers we'll entertain in SL (practical limit) + +constexpr S32 MIN_IMAGE_SIZE = (1<<MIN_IMAGE_MIP); // 4, only used for expand/contract power of 2 +constexpr S32 MAX_IMAGE_SIZE = (1<<MAX_IMAGE_MIP); // 4096 +constexpr S32 MIN_IMAGE_AREA = MIN_IMAGE_SIZE * MIN_IMAGE_SIZE; +constexpr S32 MAX_IMAGE_AREA = MAX_IMAGE_SIZE * MAX_IMAGE_SIZE; +constexpr S32 MAX_IMAGE_COMPONENTS = 8; +constexpr S32 MAX_IMAGE_DATA_SIZE = MAX_IMAGE_AREA * MAX_IMAGE_COMPONENTS; //4096 * 4096 * 8 = 128 MB // Note! These CANNOT be changed without modifying simulator code // *TODO: change both to 1024 when SIM texture fetching is deprecated -const S32 FIRST_PACKET_SIZE = 600; -const S32 MAX_IMG_PACKET_SIZE = 1000; -const S32 HTTP_PACKET_SIZE = 1496; +constexpr S32 FIRST_PACKET_SIZE = 600; +constexpr S32 MAX_IMG_PACKET_SIZE = 1000; +constexpr S32 HTTP_PACKET_SIZE = 1496; // Base classes for images. // There are two major parts for the image: diff --git a/indra/llimage/llimagefilter.cpp b/indra/llimage/llimagefilter.cpp index 0d15906afd..bfcb1f76de 100644 --- a/indra/llimage/llimagefilter.cpp +++ b/indra/llimage/llimagefilter.cpp @@ -253,7 +253,7 @@ void LLImageFilter::executeFilter(LLPointer<LLImageRaw> raw_image) bool abs_value = (mFilterData[i][index++].asReal() > 0.0); for (S32 k = 0; k < NUM_VALUES_IN_MAT3; k++) for (S32 j = 0; j < NUM_VALUES_IN_MAT3; j++) - kernel.mMatrix[k][j] = mFilterData[i][index++].asReal(); + kernel.mMatrix[k][j] = (F32)mFilterData[i][index++].asReal(); convolve(kernel,normalize,abs_value); } else if (filter_name == "colortransform") @@ -262,7 +262,7 @@ void LLImageFilter::executeFilter(LLPointer<LLImageRaw> raw_image) S32 index = 1; for (S32 k = 0; k < NUM_VALUES_IN_MAT3; k++) for (S32 j = 0; j < NUM_VALUES_IN_MAT3; j++) - transform.mMatrix[k][j] = mFilterData[i][index++].asReal(); + transform.mMatrix[k][j] = (F32)mFilterData[i][index++].asReal(); transform.transpose(); colorTransform(transform); } @@ -279,32 +279,32 @@ void LLImageFilter::executeFilter(LLPointer<LLImageRaw> raw_image) void LLImageFilter::blendStencil(F32 alpha, U8* pixel, U8 red, U8 green, U8 blue) { - F32 inv_alpha = 1.0 - alpha; + F32 inv_alpha = 1.0f - alpha; switch (mStencilBlendMode) { case STENCIL_BLEND_MODE_BLEND: // Classic blend of incoming color with the background image - pixel[VRED] = inv_alpha * pixel[VRED] + alpha * red; - pixel[VGREEN] = inv_alpha * pixel[VGREEN] + alpha * green; - pixel[VBLUE] = inv_alpha * pixel[VBLUE] + alpha * blue; + pixel[VRED] = (U8)(inv_alpha * pixel[VRED] + alpha * red); + pixel[VGREEN] = (U8)(inv_alpha * pixel[VGREEN] + alpha * green); + pixel[VBLUE] = (U8)(inv_alpha * pixel[VBLUE] + alpha * blue); break; case STENCIL_BLEND_MODE_ADD: // Add incoming color to the background image - pixel[VRED] = llclampb(pixel[VRED] + alpha * red); - pixel[VGREEN] = llclampb(pixel[VGREEN] + alpha * green); - pixel[VBLUE] = llclampb(pixel[VBLUE] + alpha * blue); + pixel[VRED] = (U8)llclampb(pixel[VRED] + alpha * red); + pixel[VGREEN] = (U8)llclampb(pixel[VGREEN] + alpha * green); + pixel[VBLUE] = (U8)llclampb(pixel[VBLUE] + alpha * blue); break; case STENCIL_BLEND_MODE_ABACK: // Add back background image to the incoming color - pixel[VRED] = llclampb(inv_alpha * pixel[VRED] + red); - pixel[VGREEN] = llclampb(inv_alpha * pixel[VGREEN] + green); - pixel[VBLUE] = llclampb(inv_alpha * pixel[VBLUE] + blue); + pixel[VRED] = (U8)llclampb(inv_alpha * pixel[VRED] + red); + pixel[VGREEN] = (U8)llclampb(inv_alpha * pixel[VGREEN] + green); + pixel[VBLUE] = (U8)llclampb(inv_alpha * pixel[VBLUE] + blue); break; case STENCIL_BLEND_MODE_FADE: // Fade incoming color to black - pixel[VRED] = alpha * red; - pixel[VGREEN] = alpha * green; - pixel[VBLUE] = alpha * blue; + pixel[VRED] = (U8)(alpha * red); + pixel[VGREEN] = (U8)(alpha * green); + pixel[VBLUE] = (U8)(alpha * blue); break; } } @@ -348,7 +348,7 @@ void LLImageFilter::colorTransform(const LLMatrix3 &transform) dst.clamp(0.0f,255.0f); // Blend result - blendStencil(getStencilAlpha(i,j), dst_data, dst.mV[VRED], dst.mV[VGREEN], dst.mV[VBLUE]); + blendStencil(getStencilAlpha(i,j), dst_data, (U8)dst.mV[VRED], (U8)dst.mV[VGREEN], (U8)dst.mV[VBLUE]); dst_data += components; } } @@ -463,7 +463,7 @@ void LLImageFilter::convolve(const LLMatrix3 &kernel, bool normalize, bool abs_v dst.clamp(0.0f,255.0f); // Blend result - blendStencil(getStencilAlpha(i,j), dst_data, dst.mV[VRED], dst.mV[VGREEN], dst.mV[VBLUE]); + blendStencil(getStencilAlpha(i,j), dst_data, (U8)dst.mV[VRED], (U8)dst.mV[VGREEN], (U8)dst.mV[VBLUE]); // Next pixel dst_data += components; @@ -499,7 +499,7 @@ void LLImageFilter::filterScreen(EScreenMode mode, const F32 wave_length, const S32 width = mImage->getWidth(); S32 height = mImage->getHeight(); - F32 wave_length_pixels = wave_length * (F32)(height) / 2.0; + F32 wave_length_pixels = wave_length * (F32)(height) / 2.0f; F32 sin = sinf(angle*DEG_TO_RAD); F32 cos = cosf(angle*DEG_TO_RAD); @@ -507,7 +507,7 @@ void LLImageFilter::filterScreen(EScreenMode mode, const F32 wave_length, const U8 gamma[256]; for (S32 i = 0; i < 256; i++) { - F32 gamma_i = llclampf((float)(powf((float)(i)/255.0,1.0/4.0))); + F32 gamma_i = llclampf((float)(powf((float)(i)/255.0f,1.0f/4.0f))); gamma[i] = (U8)(255.0 * gamma_i); } @@ -525,11 +525,11 @@ void LLImageFilter::filterScreen(EScreenMode mode, const F32 wave_length, const case SCREEN_MODE_2DSINE: di = cos*i + sin*j; dj = -sin*i + cos*j; - value = (sinf(2*F_PI*di/wave_length_pixels)*sinf(2*F_PI*dj/wave_length_pixels)+1.0)*255.0/2.0; + value = (sinf(2*F_PI*di/wave_length_pixels)*sinf(2*F_PI*dj/wave_length_pixels)+1.0f)*255.0f/2.0f; break; case SCREEN_MODE_LINE: dj = sin*i - cos*j; - value = (sinf(2*F_PI*dj/wave_length_pixels)+1.0)*255.0/2.0; + value = (sinf(2*F_PI*dj/wave_length_pixels)+1.0f)*255.0f/2.0f; break; } U8 dst_value = (dst_data[VRED] >= (U8)(value) ? gamma[dst_data[VRED] - (U8)(value)] : 0); @@ -556,16 +556,16 @@ void LLImageFilter::setStencil(EStencilShape shape, EStencilBlendMode mode, F32 mStencilCenterX = (S32)(mImage->getWidth() + params[0] * (F32)(mImage->getHeight()))/2; mStencilCenterY = (S32)(mImage->getHeight() + params[1] * (F32)(mImage->getHeight()))/2; mStencilWidth = (S32)(params[2] * (F32)(mImage->getHeight()))/2; - mStencilGamma = (params[3] <= 0.0 ? 1.0 : params[3]); + mStencilGamma = (params[3] <= 0.0f ? 1.0f : params[3]); - mStencilWavelength = (params[0] <= 0.0 ? 10.0 : params[0] * (F32)(mImage->getHeight()) / 2.0); + mStencilWavelength = (params[0] <= 0.0f ? 10.0f : params[0] * (F32)(mImage->getHeight()) / 2.0f); mStencilSine = sinf(params[1]*DEG_TO_RAD); mStencilCosine = cosf(params[1]*DEG_TO_RAD); - mStencilStartX = ((F32)(mImage->getWidth()) + params[0] * (F32)(mImage->getHeight()))/2.0; - mStencilStartY = ((F32)(mImage->getHeight()) + params[1] * (F32)(mImage->getHeight()))/2.0; - F32 end_x = ((F32)(mImage->getWidth()) + params[2] * (F32)(mImage->getHeight()))/2.0; - F32 end_y = ((F32)(mImage->getHeight()) + params[3] * (F32)(mImage->getHeight()))/2.0; + mStencilStartX = ((F32)(mImage->getWidth()) + params[0] * (F32)(mImage->getHeight()))/2.0f; + mStencilStartY = ((F32)(mImage->getHeight()) + params[1] * (F32)(mImage->getHeight()))/2.0f; + F32 end_x = ((F32)(mImage->getWidth()) + params[2] * (F32)(mImage->getHeight()))/2.0f; + F32 end_y = ((F32)(mImage->getHeight()) + params[3] * (F32)(mImage->getHeight()))/2.0f; mStencilGradX = end_x - mStencilStartX; mStencilGradY = end_y - mStencilStartY; mStencilGradN = mStencilGradX*mStencilGradX + mStencilGradY*mStencilGradY; @@ -578,14 +578,14 @@ F32 LLImageFilter::getStencilAlpha(S32 i, S32 j) { // alpha is a modified gaussian value, with a center and fading in a circular pattern toward the edges // The gamma parameter controls the intensity of the drop down from alpha 1.0 (center) to 0.0 - F32 d_center_square = (i - mStencilCenterX)*(i - mStencilCenterX) + (j - mStencilCenterY)*(j - mStencilCenterY); + F32 d_center_square = (F32)((i - mStencilCenterX)*(i - mStencilCenterX) + (j - mStencilCenterY)*(j - mStencilCenterY)); alpha = powf(F_E, -(powf((d_center_square/(mStencilWidth*mStencilWidth)),mStencilGamma)/2.0f)); } else if (mStencilShape == STENCIL_SHAPE_SCAN_LINES) { // alpha varies according to a squared sine function. F32 d = mStencilSine*i - mStencilCosine*j; - alpha = (sinf(2*F_PI*d/mStencilWavelength) > 0.0 ? 1.0 : 0.0); + alpha = (sinf(2*F_PI*d/mStencilWavelength) > 0.0f ? 1.0f : 0.0f); } else if (mStencilShape == STENCIL_SHAPE_GRADIENT) { @@ -756,11 +756,11 @@ void LLImageFilter::filterGamma(F32 gamma, const LLColor3& alpha) for (S32 i = 0; i < 256; i++) { - F32 gamma_i = llclampf((float)(powf((float)(i)/255.0,1.0/gamma))); + F32 gamma_i = llclampf((float)(powf((float)(i)/255.0f,1.0f/gamma))); // Blend in with alpha values - gamma_red_lut[i] = (U8)((1.0 - alpha.mV[0]) * (float)(i) + alpha.mV[0] * 255.0 * gamma_i); - gamma_green_lut[i] = (U8)((1.0 - alpha.mV[1]) * (float)(i) + alpha.mV[1] * 255.0 * gamma_i); - gamma_blue_lut[i] = (U8)((1.0 - alpha.mV[2]) * (float)(i) + alpha.mV[2] * 255.0 * gamma_i); + gamma_red_lut[i] = (U8)((1.0f - alpha.mV[0]) * (float)(i) + alpha.mV[0] * 255.0f * gamma_i); + gamma_green_lut[i] = (U8)((1.0f - alpha.mV[1]) * (float)(i) + alpha.mV[1] * 255.0f * gamma_i); + gamma_blue_lut[i] = (U8)((1.0f - alpha.mV[2]) * (float)(i) + alpha.mV[2] * 255.0f * gamma_i); } colorCorrect(gamma_red_lut,gamma_green_lut,gamma_blue_lut); @@ -808,23 +808,23 @@ void LLImageFilter::filterLinearize(F32 tail, const LLColor3& alpha) { U8 value_i = (i < min_v ? 0 : 255); // Blend in with alpha values - linear_red_lut[i] = (U8)((1.0 - alpha.mV[0]) * (float)(i) + alpha.mV[0] * value_i); - linear_green_lut[i] = (U8)((1.0 - alpha.mV[1]) * (float)(i) + alpha.mV[1] * value_i); - linear_blue_lut[i] = (U8)((1.0 - alpha.mV[2]) * (float)(i) + alpha.mV[2] * value_i); + linear_red_lut[i] = (U8)((1.0f - alpha.mV[0]) * (float)(i) + alpha.mV[0] * value_i); + linear_green_lut[i] = (U8)((1.0f - alpha.mV[1]) * (float)(i) + alpha.mV[1] * value_i); + linear_blue_lut[i] = (U8)((1.0f - alpha.mV[2]) * (float)(i) + alpha.mV[2] * value_i); } } else { // Linearize between min and max - F32 slope = 255.0 / (F32)(max_v - min_v); + F32 slope = 255.0f / (F32)(max_v - min_v); F32 translate = -min_v * slope; for (S32 i = 0; i < 256; i++) { U8 value_i = (U8)(llclampb((S32)(slope*i + translate))); // Blend in with alpha values - linear_red_lut[i] = (U8)((1.0 - alpha.mV[0]) * (float)(i) + alpha.mV[0] * value_i); - linear_green_lut[i] = (U8)((1.0 - alpha.mV[1]) * (float)(i) + alpha.mV[1] * value_i); - linear_blue_lut[i] = (U8)((1.0 - alpha.mV[2]) * (float)(i) + alpha.mV[2] * value_i); + linear_red_lut[i] = (U8)((1.0f - alpha.mV[0]) * (float)(i) + alpha.mV[0] * value_i); + linear_green_lut[i] = (U8)((1.0f - alpha.mV[1]) * (float)(i) + alpha.mV[1] * value_i); + linear_blue_lut[i] = (U8)((1.0f - alpha.mV[2]) * (float)(i) + alpha.mV[2] * value_i); } } @@ -863,9 +863,9 @@ void LLImageFilter::filterEqualize(S32 nb_classes, const LLColor3& alpha) for (S32 i = 0; i < 256; i++) { // Blend in current_value with alpha values - equalize_red_lut[i] = (U8)((1.0 - alpha.mV[0]) * (float)(i) + alpha.mV[0] * current_value); - equalize_green_lut[i] = (U8)((1.0 - alpha.mV[1]) * (float)(i) + alpha.mV[1] * current_value); - equalize_blue_lut[i] = (U8)((1.0 - alpha.mV[2]) * (float)(i) + alpha.mV[2] * current_value); + equalize_red_lut[i] = (U8)((1.0f - alpha.mV[0]) * (float)(i) + alpha.mV[0] * current_value); + equalize_green_lut[i] = (U8)((1.0f - alpha.mV[1]) * (float)(i) + alpha.mV[1] * current_value); + equalize_blue_lut[i] = (U8)((1.0f - alpha.mV[2]) * (float)(i) + alpha.mV[2] * current_value); if (cumulated_histo[i] >= current_count) { current_count += delta_count; @@ -884,15 +884,15 @@ void LLImageFilter::filterColorize(const LLColor3& color, const LLColor3& alpha) U8 green_lut[256]; U8 blue_lut[256]; - F32 red_composite = 255.0 * alpha.mV[0] * color.mV[0]; - F32 green_composite = 255.0 * alpha.mV[1] * color.mV[1]; - F32 blue_composite = 255.0 * alpha.mV[2] * color.mV[2]; + F32 red_composite = 255.0f * alpha.mV[0] * color.mV[0]; + F32 green_composite = 255.0f * alpha.mV[1] * color.mV[1]; + F32 blue_composite = 255.0f * alpha.mV[2] * color.mV[2]; for (S32 i = 0; i < 256; i++) { - red_lut[i] = (U8)(llclampb((S32)((1.0 - alpha.mV[0]) * (F32)(i) + red_composite))); - green_lut[i] = (U8)(llclampb((S32)((1.0 - alpha.mV[1]) * (F32)(i) + green_composite))); - blue_lut[i] = (U8)(llclampb((S32)((1.0 - alpha.mV[2]) * (F32)(i) + blue_composite))); + red_lut[i] = (U8)(llclampb((S32)((1.0f - alpha.mV[0]) * (F32)(i) + red_composite))); + green_lut[i] = (U8)(llclampb((S32)((1.0f - alpha.mV[1]) * (F32)(i) + green_composite))); + blue_lut[i] = (U8)(llclampb((S32)((1.0f - alpha.mV[2]) * (F32)(i) + blue_composite))); } colorCorrect(red_lut,green_lut,blue_lut); @@ -904,15 +904,15 @@ void LLImageFilter::filterContrast(F32 slope, const LLColor3& alpha) U8 contrast_green_lut[256]; U8 contrast_blue_lut[256]; - F32 translate = 128.0 * (1.0 - slope); + F32 translate = 128.0f * (1.0f - slope); for (S32 i = 0; i < 256; i++) { U8 value_i = (U8)(llclampb((S32)(slope*i + translate))); // Blend in with alpha values - contrast_red_lut[i] = (U8)((1.0 - alpha.mV[0]) * (float)(i) + alpha.mV[0] * value_i); - contrast_green_lut[i] = (U8)((1.0 - alpha.mV[1]) * (float)(i) + alpha.mV[1] * value_i); - contrast_blue_lut[i] = (U8)((1.0 - alpha.mV[2]) * (float)(i) + alpha.mV[2] * value_i); + contrast_red_lut[i] = (U8)((1.0f - alpha.mV[0]) * (float)(i) + alpha.mV[0] * value_i); + contrast_green_lut[i] = (U8)((1.0f - alpha.mV[1]) * (float)(i) + alpha.mV[1] * value_i); + contrast_blue_lut[i] = (U8)((1.0f - alpha.mV[2]) * (float)(i) + alpha.mV[2] * value_i); } colorCorrect(contrast_red_lut,contrast_green_lut,contrast_blue_lut); @@ -924,15 +924,15 @@ void LLImageFilter::filterBrightness(F32 add, const LLColor3& alpha) U8 brightness_green_lut[256]; U8 brightness_blue_lut[256]; - S32 add_value = (S32)(add * 255.0); + S32 add_value = (S32)(add * 255.0f); for (S32 i = 0; i < 256; i++) { U8 value_i = (U8)(llclampb(i + add_value)); // Blend in with alpha values - brightness_red_lut[i] = (U8)((1.0 - alpha.mV[0]) * (float)(i) + alpha.mV[0] * value_i); - brightness_green_lut[i] = (U8)((1.0 - alpha.mV[1]) * (float)(i) + alpha.mV[1] * value_i); - brightness_blue_lut[i] = (U8)((1.0 - alpha.mV[2]) * (float)(i) + alpha.mV[2] * value_i); + brightness_red_lut[i] = (U8)((1.0f - alpha.mV[0]) * (float)(i) + alpha.mV[0] * value_i); + brightness_green_lut[i] = (U8)((1.0f - alpha.mV[1]) * (float)(i) + alpha.mV[1] * value_i); + brightness_blue_lut[i] = (U8)((1.0f - alpha.mV[2]) * (float)(i) + alpha.mV[2] * value_i); } colorCorrect(brightness_red_lut,brightness_green_lut,brightness_blue_lut); diff --git a/indra/llimage/llimagej2c.cpp b/indra/llimage/llimagej2c.cpp index 0058b91b0f..42f3e92257 100644 --- a/indra/llimage/llimagej2c.cpp +++ b/indra/llimage/llimagej2c.cpp @@ -275,30 +275,24 @@ S32 LLImageJ2C::calcDataSizeJ2C(S32 w, S32 h, S32 comp, S32 discard_level, F32 r // For details about the equation used here, see https://wiki.lindenlab.com/wiki/THX1138_KDU_Improvements#Byte_Range_Study // Estimate the number of layers. This is consistent with what's done for j2c encoding in LLImageJ2CKDU::encodeImpl(). + constexpr S32 precision = 8; // assumed bitrate per component channel, might change in future for HDR support + constexpr S32 max_components = 4; // assumed the file has four components; three color and alpha S32 nb_layers = 1; - S32 surface = w*h; + const S32 surface = w*h; S32 s = 64*64; + S32 totalbytes = (S32)(s * max_components * precision * rate); // first level computed before loop while (surface > s) { + if (nb_layers <= (5 - discard_level)) + totalbytes += (S32)(s * max_components * precision * rate); nb_layers++; s *= 4; } - F32 layer_factor = 3.0f * (7 - llclamp(nb_layers,1,6)); - - // Compute w/pow(2,discard_level) and h/pow(2,discard_level) - w >>= discard_level; - h >>= discard_level; - w = llmax(w, 1); - h = llmax(h, 1); - - // Temporary: compute both new and old range and pick one according to the settings TextureNewByteRange - // *TODO: Take the old code out once we have enough tests done - S32 bytes; - S32 new_bytes = (S32) (sqrt((F32)(w*h))*(F32)(comp)*rate*1000.f/layer_factor); - S32 old_bytes = (S32)((F32)(w*h*comp)*rate); - bytes = (LLImage::useNewByteRange() && (new_bytes < old_bytes) ? new_bytes : old_bytes); - bytes = llmax(bytes, calcHeaderSizeJ2C()); - return bytes; + + totalbytes /= 8; // to bytes + totalbytes += calcHeaderSizeJ2C(); // header + + return totalbytes; } S32 LLImageJ2C::calcHeaderSize() |