[AMDGPU] Fix MaxWorkGroupsPerCU for large workgroups

[rampitec]

Summary

This patch corrects the maximum workgroups per CU if we have big
workgroups (more than 128). This calculation contributes to the
occupancy calculation in respect to LDS size.

[kzhuravl]

Can you upload a full diff?

[kzhuravl]

LGTM.

[rampitec]

Upload a full diff. No changes.

lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp

 unsigned getEUsPerCU(const FeatureBitset &Features) {
   return 4;
 }
 
 unsigned getMaxWorkGroupsPerCU(const FeatureBitset &Features,
                                unsigned FlatWorkGroupSize) {
   if (!Features.test(FeatureGCN))
     return 8;
-  return getWavesPerWorkGroup(Features, FlatWorkGroupSize) == 1 ? 40 : 16;
+  unsigned N = getWavesPerWorkGroup(Features, FlatWorkGroupSize);
+  if (N == 1)
+    return 40;
+  N = 40 / N;
+  return std::min(N, 16u);
 }
 
 unsigned getMaxWavesPerCU(const FeatureBitset &Features) {
   return getMaxWavesPerEU(Features) * getEUsPerCU(Features);
 }
 
 unsigned getMaxWavesPerCU(const FeatureBitset &Features,
                           unsigned FlatWorkGroupSize) {

test/CodeGen/AMDGPU/large-work-group-promote-alloca.ll

entry:
   %arrayidx12 = getelementptr inbounds [5 x i32], [5 x i32]* %stack, i32 0, i32 1
   %3 = load i32, i32* %arrayidx12
   %arrayidx13 = getelementptr inbounds i32, i32 addrspace(1)* %out, i32 1
   store i32 %3, i32 addrspace(1)* %arrayidx13
   ret void
 }
 
 ; ALL-LABEL: @occupancy_0(
-; ALL: alloca [5 x i32]
+; CI-NOT: alloca [5 x i32]
+; SI: alloca [5 x i32]
 define void @occupancy_0(i32 addrspace(1)* nocapture %out, i32 addrspace(1)* nocapture %in) #3 {
 entry:
   %stack = alloca [5 x i32], align 4
   %0 = load i32, i32 addrspace(1)* %in, align 4
   %arrayidx1 = getelementptr inbounds [5 x i32], [5 x i32]* %stack, i32 0, i32 %0
   store i32 4, i32* %arrayidx1, align 4
   %arrayidx2 = getelementptr inbounds i32, i32 addrspace(1)* %in, i32 1
   %1 = load i32, i32 addrspace(1)* %arrayidx2, align 4
   %arrayidx3 = getelementptr inbounds [5 x i32], [5 x i32]* %stack, i32 0, i32 %1
   store i32 5, i32* %arrayidx3, align 4
   %arrayidx10 = getelementptr inbounds [5 x i32], [5 x i32]* %stack, i32 0, i32 0
   %2 = load i32, i32* %arrayidx10, align 4
   store i32 %2, i32 addrspace(1)* %out, align 4
   %arrayidx12 = getelementptr inbounds [5 x i32], [5 x i32]* %stack, i32 0, i32 1
   %3 = load i32, i32* %arrayidx12
   %arrayidx13 = getelementptr inbounds i32, i32 addrspace(1)* %out, i32 1
   store i32 %3, i32 addrspace(1)* %arrayidx13
   ret void
 }
 
 ; ALL-LABEL: @occupancy_max(
-; ALL: alloca [5 x i32]
+; CI-NOT: alloca [5 x i32]
+; SI: alloca [5 x i32]
 define void @occupancy_max(i32 addrspace(1)* nocapture %out, i32 addrspace(1)* nocapture %in) #4 {
 entry:
   %stack = alloca [5 x i32], align 4
   %0 = load i32, i32 addrspace(1)* %in, align 4
   %arrayidx1 = getelementptr inbounds [5 x i32], [5 x i32]* %stack, i32 0, i32 %0
   store i32 4, i32* %arrayidx1, align 4
   %arrayidx2 = getelementptr inbounds i32, i32 addrspace(1)* %in, i32 1
   %1 = load i32, i32 addrspace(1)* %arrayidx2, align 4