From 2053abc3b13eabc5689cd8a20184ffa802a72278 Mon Sep 17 00:00:00 2001
From: Aidan Belton <abeltons@amd.com>
Date: Mon, 9 Dec 2024 20:59:48 +0530
Subject: [PATCH] SWDEV-475380 - fix perftests on cuda

Change-Id: Iae6fc6cfdc4c2e6cb07562a03ff4e055601ed463
---
 .../memory/hipPerfBufferCopySpeedAll2All.cc       | 15 +++++++++++++++
 1 file changed, 15 insertions(+)

diff --git a/catch/perftests/memory/hipPerfBufferCopySpeedAll2All.cc b/catch/perftests/memory/hipPerfBufferCopySpeedAll2All.cc
index b32c3265a0..9e6655a385 100644
--- a/catch/perftests/memory/hipPerfBufferCopySpeedAll2All.cc
+++ b/catch/perftests/memory/hipPerfBufferCopySpeedAll2All.cc
@@ -69,12 +69,22 @@ static void mallocDevBuf(void** pp, size_t size, DEV_MEM_TYPE memType) {
     HIP_CHECK(hipMalloc(pp, size));
     break;
   case FINE_GRAINED:
+#if HT_AMD
     HIP_CHECK(hipExtMallocWithFlags(pp, size, hipDeviceMallocFinegrained));
+#else
+    fprintf(stderr, "Unsupported memType for nvidia hardware: %d\n", memType);
+    REQUIRE(false);
+#endif
     break;
   case EXTENDED_FINE_GRAINED:
     // Extended - Scope Fine Grained Memory: read is cached, write is not
     // Perf gain compared with cacheable write
+#if HT_AMD
     HIP_CHECK(hipExtMallocWithFlags(pp, size, hipDeviceMallocUncached));
+#else
+    fprintf(stderr, "Unsupported memType for nvidia hardware: %d\n", memType);
+    REQUIRE(false);
+#endif
     break;
   default:
     fprintf(stderr, "Unknown memType = %d\n", memType);
@@ -232,12 +242,17 @@ static void testCopyPerf(bool toRemote, bool kernelCopy, bool onOneGpu,
 
 static void testCopyPerf(bool toRemote, bool kernelCopy, bool onOneGpu) {
   fprintf(stderr, "**********************************************************\n");
+#if HT_AMD
   for (int srcType = COARSE_GRAINED; srcType < UNKNOWN_MEM; srcType++) {
     for (int dstType = COARSE_GRAINED; dstType < UNKNOWN_MEM; dstType++) {
       testCopyPerf(toRemote, kernelCopy, onOneGpu,
         static_cast<DEV_MEM_TYPE>(srcType), static_cast<DEV_MEM_TYPE>(dstType));
     }
   }
+#else
+  // Only support coarse grained memory allocation on nvidia GPUs
+  testCopyPerf(toRemote, kernelCopy, onOneGpu, COARSE_GRAINED, COARSE_GRAINED);
+#endif
 }
 
 /**