From 5dff283f4eb60ecc1739b346e4c04d0d25bc3752 Mon Sep 17 00:00:00 2001
From: "Jayaprakash, Karthik" <Karthik.Jayaprakash@amd.com>
Date: Mon, 12 May 2025 11:09:03 -0400
Subject: [PATCH] SWDEV-526855 - Modify the SIMDPerCU calculation for
 gfx1250/1. (#275)

[ROCm/clr commit: acb1f7e8d5e16ed9eca5b56dafc42e3c10811b52]
---
 projects/clr/hipamd/src/hip_platform.cpp | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)

diff --git a/projects/clr/hipamd/src/hip_platform.cpp b/projects/clr/hipamd/src/hip_platform.cpp
index ea095adf81..ae50e4ce09 100644
--- a/projects/clr/hipamd/src/hip_platform.cpp
+++ b/projects/clr/hipamd/src/hip_platform.cpp
@@ -404,8 +404,14 @@ hipError_t ihipOccupancyMaxActiveBlocksPerMultiprocessor(
     const size_t SgprWaves = maxSGPRs / amd::alignUp(wrkGrpInfo->usedSGPRs_, 16);
     GprWaves = std::min(VgprWaves, SgprWaves);
   }
-  uint32_t simdPerCU = (device.isa().versionMajor() <= 9) ? device.info().simdPerCU_
-                                                          : (wrkGrpInfo->isWGPMode_ ? 4 : 2);
+
+  // The table contains SIMD per CU, not per WGP, so when WGP mode is set on kernel metadata,
+  // multiply the number of SIMDs by 2, to account for 2CUs in 1 WGP.
+  uint32_t simdPerCU = device.isa().simdPerCU();
+  if (wrkGrpInfo->isWGPMode_) {
+    simdPerCU  *= 2;
+  }
+
   const size_t alu_occupancy = simdPerCU * std::min(MaxWavesPerSimd, GprWaves);
   const int alu_limited_threads = alu_occupancy * wrkGrpInfo->wavefrontSize_;