From 9c361faab3a1a3ee8a06dffa24abe560261dedc8 Mon Sep 17 00:00:00 2001
From: Michael LIAO <michaelhongbin.liao@amd.com>
Date: Mon, 30 Mar 2020 09:10:16 -0400
Subject: [PATCH] [vdi] Fix calculation of MaxWaves

- Consider the case where `usedVGPRs` is zero.
- This fixes [SWDEV-228537](http://ontrack-internal.amd.com/browse/SWDEV-228537)

Change-Id: I8675311f5fe24fb59c5d45bada122afefb55b128


[ROCm/clr commit: 55d869df99e613673b518820213132f442af7501]
---
 projects/clr/hipamd/vdi/hip_platform.cpp | 11 +++++------
 1 file changed, 5 insertions(+), 6 deletions(-)

diff --git a/projects/clr/hipamd/vdi/hip_platform.cpp b/projects/clr/hipamd/vdi/hip_platform.cpp
index 688181870c..747f7457a0 100755
--- a/projects/clr/hipamd/vdi/hip_platform.cpp
+++ b/projects/clr/hipamd/vdi/hip_platform.cpp
@@ -700,17 +700,16 @@ hipError_t ihipOccupancyMaxActiveBlocksPerMultiprocessor(int* numBlocks, int* nu
   }
   // Find threads accupancy per CU => simd_per_cu * GPR usage
   constexpr size_t MaxWavesPerSimd = 8;  // Limited by SPI 32 per CU, hence 8 per SIMD
-  size_t VgprWaves = wrkGrpInfo->availableVGPRs_ / amd::alignUp(wrkGrpInfo->usedVGPRs_, 4);
-
-  size_t GprWaves;
+  size_t VgprWaves = MaxWavesPerSimd;
+  if (wrkGrpInfo->usedVGPRs_ > 0) {
+    VgprWaves = wrkGrpInfo->availableVGPRs_ / amd::alignUp(wrkGrpInfo->usedVGPRs_, 4);
+  }
+  size_t GprWaves = VgprWaves;
   if (wrkGrpInfo->usedSGPRs_ > 0) {
     const size_t maxSGPRs = (device->info().gfxipVersion_ < 800) ? 512 : 800;
     size_t SgprWaves = maxSGPRs / amd::alignUp(wrkGrpInfo->usedSGPRs_, 16);
     GprWaves = std::min(VgprWaves, SgprWaves);
   }
-  else {
-    GprWaves = VgprWaves;
-  }
 
   size_t alu_accupancy = device->info().simdPerCU_ * std::min(MaxWavesPerSimd, GprWaves);
   alu_accupancy *= wrkGrpInfo->wavefrontSize_;