aboutsummaryrefslogtreecommitdiff
path: root/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
diff options
context:
space:
mode:
Diffstat (limited to 'llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp')
-rw-r--r--llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp13
1 files changed, 13 insertions, 0 deletions
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
index 31a80e0..c865082 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
@@ -25,6 +25,7 @@
#include "AMDGPUMacroFusion.h"
#include "AMDGPUPerfHintAnalysis.h"
#include "AMDGPUPreloadKernArgProlog.h"
+#include "AMDGPUPrepareAGPRAlloc.h"
#include "AMDGPURemoveIncompatibleFunctions.h"
#include "AMDGPUReserveWWMRegs.h"
#include "AMDGPUResourceUsageAnalysis.h"
@@ -499,6 +500,7 @@ extern "C" LLVM_ABI LLVM_EXTERNAL_VISIBILITY void LLVMInitializeAMDGPUTarget() {
initializeGlobalISel(*PR);
initializeAMDGPUAsmPrinterPass(*PR);
initializeAMDGPUDAGToDAGISelLegacyPass(*PR);
+ initializeAMDGPUPrepareAGPRAllocLegacyPass(*PR);
initializeGCNDPPCombineLegacyPass(*PR);
initializeSILowerI1CopiesLegacyPass(*PR);
initializeAMDGPUGlobalISelDivergenceLoweringPass(*PR);
@@ -1196,6 +1198,7 @@ public:
bool addRegBankSelect() override;
void addPreGlobalInstructionSelect() override;
bool addGlobalInstructionSelect() override;
+ void addPreRegAlloc() override;
void addFastRegAlloc() override;
void addOptimizedRegAlloc() override;
@@ -1539,6 +1542,11 @@ void GCNPassConfig::addFastRegAlloc() {
TargetPassConfig::addFastRegAlloc();
}
+void GCNPassConfig::addPreRegAlloc() {
+ if (getOptLevel() != CodeGenOptLevel::None)
+ addPass(&AMDGPUPrepareAGPRAllocLegacyID);
+}
+
void GCNPassConfig::addOptimizedRegAlloc() {
if (EnableDCEInRA)
insertPass(&DetectDeadLanesID, &DeadMachineInstructionElimID);
@@ -2235,6 +2243,11 @@ void AMDGPUCodeGenPassBuilder::addOptimizedRegAlloc(
Base::addOptimizedRegAlloc(addPass);
}
+void AMDGPUCodeGenPassBuilder::addPreRegAlloc(AddMachinePass &addPass) const {
+ if (getOptLevel() != CodeGenOptLevel::None)
+ addPass(AMDGPUPrepareAGPRAllocPass());
+}
+
Error AMDGPUCodeGenPassBuilder::addRegAssignmentOptimized(
AddMachinePass &addPass) const {
// TODO: Check --regalloc-npm option