P4 to Git Change 1135675 by atimofee@atimofee-hsa on 2015/03/30 10:03:20

ECR #333753 - HSA HLC: add query fast_fma32 chip capability

Affected files ...

... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/common/compiler_stage.hpp#12 edit
... //depot/stg/opencl/drivers/opencl/compiler/lib/backends/common/linker.cpp#124 edit
... //depot/stg/opencl/drivers/opencl/compiler/llvm/include/llvm/AMDLLVMContextHook.h#25 edit
... //depot/stg/opencl/drivers/opencl/compiler/llvm/include/llvm/AMDResolveLinker.h#4 edit
... //depot/stg/opencl/drivers/opencl/compiler/llvm/lib/Linker/AMDResolveLinker.cpp#8 edit
... //depot/stg/opencl/drivers/opencl/compiler/llvm/tools/llvm-link/llvm-link.cpp#52 edit
... //depot/stg/opencl/drivers/opencl/compiler/llvm32/include/llvm/AMDLLVMContextHook.h#3 edit
... //depot/stg/opencl/drivers/opencl/compiler/llvm32/include/llvm/AMDResolveLinker.h#3 edit
... //depot/stg/opencl/drivers/opencl/compiler/llvm32/lib/Linker/AMDResolveLinker.cpp#3 edit
... //depot/stg/opencl/drivers/opencl/compiler/llvm32/tools/llvm-link/llvm-link.cpp#3 edit
... //depot/stg/opencl/drivers/opencl/tests/hsa/src/ocl/features/have_fast_fma32/have_fast_fma32.cl#1 add
... //depot/stg/opencl/drivers/opencl/tests/hsa/src/ocl/features/have_fast_fma32/have_fast_fma32.lua#1 add
... //depot/stg/opencl/drivers/opencl/tests/hsa/tlst/ocl_features.tlst#17 edit

rocclr/compiler/lib/backends/common/compiler_stage.hpp

@@ -96,6 +96,7 @@ namespace amdcl
       void setUseNative(const char * Val) { if(Val) hookup_.amdoptions.OptUseNative = Val; }
       void setDenormsAreZero(bool Val) { hookup_.amdoptions.DenormsAreZero = Val; }
       void setUniformWorkGroupSize(bool Val) { hookup_.amdoptions.UniformWorkGroupSize = Val; }
+      void setHaveFastFMA32(bool Val) { hookup_.amdoptions.HaveFastFMA32 = Val; }
 
       /*! Returns the llvm binary */
       llvm::Module* LLVMBinary() const { return llvmbinary_; }

rocclr/compiler/lib/backends/common/linker.cpp

@@ -24,7 +24,6 @@
 #include "llvm/AMDFixupKernelModule.h"
 #include "llvm/AMDResolveLinker.h"
 #include "llvm/AMDPrelinkOpt.h"
-#include "llvm/AMDUtils.h"
 #include "llvm/ADT/Triple.h"
 #include "llvm/ADT/StringMap.h"
 #include "llvm/Analysis/AMDLocalArrayUsage.h"
@@ -559,6 +558,7 @@ amdcl::OCLLinker::link(llvm::Module* input, std::vector<llvm::Module*> &libs)
   if (Options()->oVariables->EnableBuildTiming) {
     time_prelinkopt = amd::Os::timeNanos();
   }
+  llvm::StringRef chip(aclGetChip(Elf()->target));
   setGPU(IsGPUTarget);
   setFiniteMathOnly(Options()->oVariables->FiniteMathOnly);
   setNoSignedZeros(Options()->oVariables->NoSignedZeros);
@@ -571,6 +571,11 @@ amdcl::OCLLinker::link(llvm::Module* input, std::vector<llvm::Module*> &libs)
   setUseNative(Options()->oVariables->OptUseNative);
   setDenormsAreZero(Options()->oVariables->DenormsAreZero);
   setUniformWorkGroupSize(Options()->oVariables->UniformWorkGroupSize);
+  setHaveFastFMA32(chip == "Cypress"
+                || chip == "Cayman"
+                || chip == "Tahiti"
+                || chip == "Hawaii"
+                || chip == "Carrizo");
   LLVMBinary()->getContext().setAMDLLVMContextHook(&hookup_);
 
   std::string clp_errmsg;
@@ -602,9 +607,9 @@ amdcl::OCLLinker::link(llvm::Module* input, std::vector<llvm::Module*> &libs)
   }
 
   std::string ErrorMessage;
-
+#ifdef LEGACY_COMPLIB
   createASICIDFunctions(LLVMBinary());
-
+#endif // LEGACY_COMPLIB
   // Link libraries to get every functions that are referenced.
   std::string ErrorMsg;
   if (resolveLink(LLVMBinary(), LibMs, &ErrorMsg)) {