Index: lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp =================================================================== --- lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp +++ lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp @@ -560,13 +560,13 @@ // SGPRBlocks is actual number of SGPR blocks minus 1. ProgInfo.SGPRBlocks = alignTo(ProgInfo.NumSGPRsForWavesPerEU, - STM.getSGPRAllocGranule()); - ProgInfo.SGPRBlocks = ProgInfo.SGPRBlocks / STM.getSGPRAllocGranule() - 1; + STM.getSGPREncodingGranule()); + ProgInfo.SGPRBlocks = ProgInfo.SGPRBlocks / STM.getSGPREncodingGranule() - 1; // VGPRBlocks is actual number of VGPR blocks minus 1. ProgInfo.VGPRBlocks = alignTo(ProgInfo.NumVGPRsForWavesPerEU, - STM.getVGPRAllocGranule()); - ProgInfo.VGPRBlocks = ProgInfo.VGPRBlocks / STM.getVGPRAllocGranule() - 1; + STM.getVGPREncodingGranule()); + ProgInfo.VGPRBlocks = ProgInfo.VGPRBlocks / STM.getVGPREncodingGranule() - 1; // Set the value to initialize FP_ROUND and FP_DENORM parts of the mode // register. Index: lib/Target/AMDGPU/AMDGPUSubtarget.h =================================================================== --- lib/Target/AMDGPU/AMDGPUSubtarget.h +++ lib/Target/AMDGPU/AMDGPUSubtarget.h @@ -644,6 +644,13 @@ /// \returns SGPR allocation granularity supported by the subtarget. unsigned getSGPRAllocGranule() const { + if (getGeneration() >= AMDGPUSubtarget::VOLCANIC_ISLANDS) + return 16; + return 8; + } + + /// \returns SGPR encoding granularity supported by the subtarget. + unsigned getSGPREncodingGranule() const { return 8; } @@ -689,6 +696,11 @@ return 4; } + /// \returns VGPR encoding granularity supported by the subtarget. + unsigned getVGPREncodingGranule() const { + return getVGPRAllocGranule(); + } + /// \returns Total number of VGPRs supported by the subtarget. unsigned getTotalNumVGPRs() const { return 256;