From: uvos <redacted>
Date: Tue, 29 Jul 2025 15:44:30 +0000 (+0200)
Subject: HIP: add GGML_HIP_MMQ_MFMA option to allow disableing the MFMA path. (llama/14930)
X-Git-Tag: upstream/0.0.2404~16
X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=2393b2c963aa30f1fd0a31b8822e8ece13acd57e;p=pkg%2Fggml%2Fsources%2Fggml

HIP: add GGML_HIP_MMQ_MFMA option to allow disableing the MFMA path. (llama/14930)

This is useful for testing for regressions on GCN with CDNA hardware.

With GGML_HIP_MMQ_MFMA=Off and GGML_CUDA_FORCE_MMQ=On we can conveniently test the GCN code path on CDNA. As CDNA is just GCN renamed with MFMA added and limited use ACC registers, this provides a good alternative for regression testing when GCN hardware is not available.
---

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 20467c54..231250ef 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -174,6 +174,7 @@ option(GGML_HIP_GRAPHS                      "ggml: use HIP graph, experimental,
 option(GGML_HIP_NO_VMM                      "ggml: do not try to use HIP VMM"                 ON)
 option(GGML_HIP_ROCWMMA_FATTN               "ggml: enable rocWMMA for FlashAttention"         OFF)
 option(GGML_HIP_FORCE_ROCWMMA_FATTN_GFX12   "ggml: enable rocWMMA FlashAttention on GFX12"    OFF)
+option(GGML_HIP_MMQ_MFMA                    "ggml: enable MFMA MMA for CDNA in MMQ"           ON)
 option(GGML_MUSA_GRAPHS                     "ggml: use MUSA graph, experimental, unstable"    OFF)
 option(GGML_MUSA_MUDNN_COPY                 "ggml: enable muDNN for accelerated copy"         OFF)
 option(GGML_VULKAN                          "ggml: use Vulkan"                                OFF)
diff --git a/src/ggml-cuda/common.cuh b/src/ggml-cuda/common.cuh
index cdc3bb5a..19fcc598 100644
--- a/src/ggml-cuda/common.cuh
+++ b/src/ggml-cuda/common.cuh
@@ -227,7 +227,7 @@ typedef float2 dfloat2;
 #define FP16_MMA_AVAILABLE
 #endif // defined(GGML_HIP_ROCWMMA_FATTN) && (defined(CDNA) || defined(RDNA3) || (defined(GGML_HIP_ROCWMMA_FATTN_GFX12) && defined(RDNA4)))
 
-#if defined(GGML_USE_HIP) && defined(__HIP_PLATFORM_AMD__) && defined(CDNA3)
+#if defined(GGML_USE_HIP) && defined(__HIP_PLATFORM_AMD__) && defined(CDNA3) && !defined(GGML_HIP_NO_MMQ_MFMA)
 #define AMD_MFMA_AVAILABLE
 #endif // defined(GGML_USE_HIP) && defined(__HIP_PLATFORM_AMD__) && defined(CDNA3)
 
@@ -295,7 +295,11 @@ static bool fp32_mma_hardware_available(const int cc) {
 
 // AMD CDNA3 matrix cores.. Will add support for other CDNA generations later.
 static bool amd_mfma_available(const int cc) {
-    return cc >= GGML_CUDA_CC_OFFSET_AMD && GGML_CUDA_CC_IS_CDNA3(cc);
+#if !defined(GGML_HIP_NO_MMQ_MFMA)
+    return GGML_CUDA_CC_IS_CDNA3(cc);
+#else
+    return false;
+#endif //!defined(GGML_HIP_NO_MMQ_MFMA)
 }
 
 // Volta technically had FP16 tensor cores but they work very differently compared to Turing and later.
diff --git a/src/ggml-hip/CMakeLists.txt b/src/ggml-hip/CMakeLists.txt
index e29df985..e92ec7fa 100644
--- a/src/ggml-hip/CMakeLists.txt
+++ b/src/ggml-hip/CMakeLists.txt
@@ -113,6 +113,10 @@ if (GGML_HIP_ROCWMMA_FATTN)
     add_compile_definitions(GGML_HIP_ROCWMMA_FATTN)
 endif()
 
+if (NOT GGML_HIP_MMQ_MFMA)
+    add_compile_definitions(GGML_HIP_NO_MMQ_MFMA)
+endif()
+
 if (GGML_HIP_FORCE_ROCWMMA_FATTN_GFX12 OR ${hip_VERSION} VERSION_GREATER_EQUAL 7.0)
     add_compile_definitions(GGML_HIP_ROCWMMA_FATTN_GFX12)
 endif()