<html> <head> <meta http-equiv="Content-Type" content="text/html; charset=us-ascii"> </head> <body style="word-wrap: break-word; -webkit-nbsp-mode: space; -webkit-line-break: after-white-space;" class=""> Reviewed-by: Tim Rowley <<a href="mailto:timothy.o.rowley@intel.com" class="">timothy.o.rowley@intel.com</a>> <div class=""> <div> <blockquote type="cite" class=""> <div class="">On Nov 7, 2016, at 6:18 PM, Ilia Mirkin <<a href="mailto:imirkin@alum.mit.edu" class="">imirkin@alum.mit.edu</a>> wrote:</div> <div class=""> <div class="">Most logic op usage is probably going to end up with normalized textures. Scale the floating point values and convert to integer before performing the logic operations. Signed-off-by: Ilia Mirkin <<a href="mailto:imirkin@alum.mit.edu" class="">imirkin@alum.mit.edu</a>> --- The gl-1.1-xor-copypixels test still fails. The image stays the same. I'm suspecting it's for reasons outside of this patch. I'm not too familiar with the whole swr infrastructure, perhaps there was an eaiser way to do all this. I looked for conversion helper functions but couldn't find anything that would fit nicely here. Feel free to point me in the right direction. .../drivers/swr/rasterizer/jitter/blend_jit.cpp | 81 +++++++++++++++++----- 1 file changed, 64 insertions(+), 17 deletions(-) diff --git a/src/gallium/drivers/swr/rasterizer/jitter/blend_jit.cpp b/src/gallium/drivers/swr/rasterizer/jitter/blend_jit.cpp index 1452d27..d69d503 100644 --- a/src/gallium/drivers/swr/rasterizer/jitter/blend_jit.cpp +++ b/src/gallium/drivers/swr/rasterizer/jitter/blend_jit.cpp @@ -649,29 +649,54 @@ struct BlendJit : public Builder if(state.blendState.logicOpEnable) { const SWR_FORMAT_INFO& info = GetFormatInfo(state.format); - SWR_ASSERT(info.type[0] == SWR_TYPE_UINT); Value* vMask[4]; + float scale[4]; + + if (!state.blendState.blendEnable) { + Clamp(state.format, src); + Clamp(state.format, dst); + } + for(uint32_t i = 0; i < 4; i++) { - switch(info.bpc[i]) + if (info.type[i] == SWR_TYPE_UNUSED) { - case 0: vMask[i] = VIMMED1(0x00000000); break; - case 2: vMask[i] = VIMMED1(0x00000003); break; - case 5: vMask[i] = VIMMED1(0x0000001F); break; - case 6: vMask[i] = VIMMED1(0x0000003F); break; - case 8: vMask[i] = VIMMED1(0x000000FF); break; - case 10: vMask[i] = VIMMED1(0x000003FF); break; - case 11: vMask[i] = VIMMED1(0x000007FF); break; - case 16: vMask[i] = VIMMED1(0x0000FFFF); break; - case 24: vMask[i] = VIMMED1(0x00FFFFFF); break; - case 32: vMask[i] = VIMMED1(0xFFFFFFFF); break; + continue; + } + + if (info.bpc[i] >= 32) { + vMask[i] = VIMMED1(0xFFFFFFFF); + scale[i] = 0xFFFFFFFF; + } else { + vMask[i] = VIMMED1((1 << info.bpc[i]) - 1); + if (info.type[i] == SWR_TYPE_SNORM) + scale[i] = (1 << (info.bpc[i] - 1)) - 1; + else + scale[i] = (1 << info.bpc[i]) - 1; + } + + switch (info.type[i]) { default: - vMask[i] = VIMMED1(0x0); - SWR_ASSERT(0, "Unsupported bpc for logic op\n"); + SWR_ASSERT(0, "Unsupported type for logic op\n"); + /* fallthrough */ + case SWR_TYPE_UINT: + case SWR_TYPE_SINT: + src[i] = BITCAST(src[i], mSimdInt32Ty); + dst[i] = BITCAST(dst[i], mSimdInt32Ty); + break; + case SWR_TYPE_SNORM: + src[i] = FADD(src[i], VIMMED1(0.5f)); + dst[i] = FADD(dst[i], VIMMED1(0.5f)); + /* fallthrough */ + case SWR_TYPE_UNORM: + src[i] = FP_TO_UI( + FMUL(src[i], VIMMED1(scale[i])), + mSimdInt32Ty); + dst[i] = FP_TO_UI( + FMUL(dst[i], VIMMED1(scale[i])), + mSimdInt32Ty); break; } - src[i] = BITCAST(src[i], mSimdInt32Ty);//, vMask[i]); - dst[i] = BITCAST(dst[i], mSimdInt32Ty); } LogicOpFunc(state.blendState.logicOpFunc, src, dst, result); @@ -679,10 +704,32 @@ struct BlendJit : public Builder // store results out for(uint32_t i = 0; i < 4; ++i) { + if (info.type[i] == SWR_TYPE_UNUSED) + { + continue; + } + // clear upper bits from PS output not in RT format after doing logic op result[i] = AND(result[i], vMask[i]); - STORE(BITCAST(result[i], mSimdFP32Ty), pResult, {i}); + switch (info.type[i]) { + default: + SWR_ASSERT(0, "Unsupported type for logic op\n"); + /* fallthrough */ + case SWR_TYPE_UINT: + case SWR_TYPE_SINT: + result[i] = BITCAST(result[i], mSimdFP32Ty); + break; + case SWR_TYPE_SNORM: + case SWR_TYPE_UNORM: + result[i] = FMUL(UI_TO_FP(result[i], mSimdFP32Ty), + VIMMED1(1.0f / scale[i])); + if (info.type[i] == SWR_TYPE_SNORM) + result[i] = FADD(result[i], VIMMED1(-0.5f)); + break; + } + + STORE(result[i], pResult, {i}); } } -- 2.7.3 </div> </div> </blockquote> </div> </div> </body> </html>