commit 7b5a54e369035a5754be2e836b7b583b3db8c884
author Stanislav Mekhanoshin <Stanislav.Mekhanoshin@amd.com> Fri Jul 19 21:29:51 2019 +0000
committer Stanislav Mekhanoshin <Stanislav.Mekhanoshin@amd.com> Fri Jul 19 21:29:51 2019 +0000
tree 121773266809914bc1622ee35de65da09cccdec9
parent 34da8dfba089e0f5523c158c219bb20ac98b1df1

[AMDGPU] Fixed occupancy calculation for gfx10

Differential Revision: https://reviews.llvm.org/D65010

llvm-svn: 366616

llvm/lib/Target/AMDGPU/AMDGPUSubtarget.cpp

diff --git a/llvm/lib/Target/AMDGPU/AMDGPUSubtarget.cpp b/llvm/lib/Target/AMDGPU/AMDGPUSubtarget.cpp
index 1eb9b83..716c387 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUSubtarget.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUSubtarget.cpp
@@ -591,25 +591,12 @@
 }
 
 unsigned GCNSubtarget::getOccupancyWithNumVGPRs(unsigned VGPRs) const {
-  if (VGPRs <= 24)
-    return 10;
-  if (VGPRs <= 28)
-    return 9;
-  if (VGPRs <= 32)
-    return 8;
-  if (VGPRs <= 36)
-    return 7;
-  if (VGPRs <= 40)
-    return 6;
-  if (VGPRs <= 48)
-    return 5;
-  if (VGPRs <= 64)
-    return 4;
-  if (VGPRs <= 84)
-    return 3;
-  if (VGPRs <= 128)
-    return 2;
-  return 1;
+  unsigned MaxWaves = getMaxWavesPerEU();
+  unsigned Granule = getVGPRAllocGranule();
+  if (VGPRs < Granule)
+    return MaxWaves;
+  unsigned RoundedRegs = ((VGPRs + Granule - 1) / Granule) * Granule;
+  return std::min(getTotalNumVGPRs() / RoundedRegs, MaxWaves);
 }
 
 unsigned GCNSubtarget::getReservedNumSGPRs(const MachineFunction &MF) const {

llvm/lib/Target/AMDGPU/AMDGPUSubtarget.h

diff --git a/llvm/lib/Target/AMDGPU/AMDGPUSubtarget.h b/llvm/lib/Target/AMDGPU/AMDGPUSubtarget.h
index 78c3b82..1f30d76 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUSubtarget.h
+++ b/llvm/lib/Target/AMDGPU/AMDGPUSubtarget.h
@@ -884,7 +884,7 @@
   /// \returns Maximum number of waves per execution unit supported by the
   /// subtarget without any kind of limitation.
   unsigned getMaxWavesPerEU() const {
-    return AMDGPU::IsaInfo::getMaxWavesPerEU();
+    return AMDGPU::IsaInfo::getMaxWavesPerEU(this);
   }
 
   /// \returns Number of waves per work group supported by the subtarget and

llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp

diff --git a/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp b/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp
index e90f40e..ef50d37 100644
--- a/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp
@@ -241,7 +241,7 @@
 }
 
 unsigned getMaxWavesPerCU(const MCSubtargetInfo *STI) {
-  return getMaxWavesPerEU() * getEUsPerCU(STI);
+  return getMaxWavesPerEU(STI) * getEUsPerCU(STI);
 }
 
 unsigned getMaxWavesPerCU(const MCSubtargetInfo *STI,
@@ -253,9 +253,11 @@
   return 1;
 }
 
-unsigned getMaxWavesPerEU() {
+unsigned getMaxWavesPerEU(const MCSubtargetInfo *STI) {
   // FIXME: Need to take scratch memory into account.
-  return 10;
+  if (!isGFX10(*STI))
+    return 10;
+  return 20;
 }
 
 unsigned getMaxWavesPerEU(const MCSubtargetInfo *STI,
@@ -317,7 +319,7 @@
   if (Version.Major >= 10)
     return 0;
 
-  if (WavesPerEU >= getMaxWavesPerEU())
+  if (WavesPerEU >= getMaxWavesPerEU(STI))
     return 0;
 
   unsigned MinNumSGPRs = getTotalNumSGPRs(STI) / (WavesPerEU + 1);
@@ -394,17 +396,19 @@
 }
 
 unsigned getTotalNumVGPRs(const MCSubtargetInfo *STI) {
-  return 256;
+  if (!isGFX10(*STI))
+    return 256;
+  return STI->getFeatureBits().test(FeatureWavefrontSize32) ? 1024 : 512;
 }
 
 unsigned getAddressableNumVGPRs(const MCSubtargetInfo *STI) {
-  return getTotalNumVGPRs(STI);
+  return 256;
 }
 
 unsigned getMinNumVGPRs(const MCSubtargetInfo *STI, unsigned WavesPerEU) {
   assert(WavesPerEU != 0);
 
-  if (WavesPerEU >= getMaxWavesPerEU())
+  if (WavesPerEU >= getMaxWavesPerEU(STI))
     return 0;
   unsigned MinNumVGPRs =
       alignDown(getTotalNumVGPRs(STI) / (WavesPerEU + 1),

llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h

diff --git a/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h b/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h
index 209ef7e..590df20 100644
--- a/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h
+++ b/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h
@@ -94,7 +94,7 @@
 
 /// \returns Maximum number of waves per execution unit for given subtarget \p
 /// STI without any kind of limitation.
-unsigned getMaxWavesPerEU();
+unsigned getMaxWavesPerEU(const MCSubtargetInfo *STI);
 
 /// \returns Maximum number of waves per execution unit for given subtarget \p
 /// STI and limited by given \p FlatWorkGroupSize.