enabled dl_gemm

CMakeLists.txt

@@ -113,12 +113,12 @@ message("checking which targets are supported")
 #Setting GPU_TARGETS on command line will override this list
 if(NOT PROFILER_ONLY)
     rocm_check_target_ids(DEFAULT_GPU_TARGETS
-        TARGETS "gfx908;gfx90a;gfx940;gfx941;gfx942;gfx1030;gfx1100;gfx1101;gfx1102")
+        TARGETS "gfx908;gfx90a;gfx940;gfx941;gfx942;gfx1030;gfx1100;gfx1101;gfx1102;gfx1200")
 else()
     add_definitions(-DPROFILER_ONLY)
     set(GPU_TARGETS "" CACHE STRING "" FORCE)
     if(GPU_TARGETS)
-        message(FATAL_ERROR "For PROFILE_ONLY build, please do not set GPU_TARGETS, use GPU_ARCH = gfx90, gfx94, gfx10, or gfx11")
+        message(FATAL_ERROR "For PROFILE_ONLY build, please do not set GPU_TARGETS, use GPU_ARCH = gfx90, gfx94, gfx10, gfx11 or gfx12")
     endif()
     if(GPU_ARCH MATCHES "gfx90")
         rocm_check_target_ids(DEFAULT_GPU_TARGETS TARGETS "gfx908;gfx90a")
@@ -128,8 +128,10 @@ else()
         rocm_check_target_ids(DEFAULT_GPU_TARGETS TARGETS "gfx1030")
     elseif(GPU_ARCH MATCHES "gfx11")
         rocm_check_target_ids(DEFAULT_GPU_TARGETS TARGETS "gfx1100;gfx1101;gfx1102")
+    elseif(GPU_ARCH MATCHES "gfx12")
+        rocm_check_target_ids(DEFAULT_GPU_TARGETS TARGETS "gfx1200")
     else()
-        message(FATAL_ERROR "For PROFILE_ONLY build, please specify GPU_ARCH as gfx90, gfx94, gfx10, or gfx11")
+        message(FATAL_ERROR "For PROFILE_ONLY build, please specify GPU_ARCH as gfx90, gfx94, gfx10, gfx11 or gfx12")
     endif()
     set(GPU_TARGETS "${DEFAULT_GPU_TARGETS}" CACHE STRING " " FORCE)
 endif()

include/ck/ck.hpp

@@ -58,6 +58,9 @@
 #if defined(__gfx1100__) || defined(__gfx1101__) || defined(__gfx1102__) || defined(__gfx1103__)
 #define __gfx11__
 #endif
+#if defined(__gfx1200__)
+#define __gfx12__
+#endif
 
 // buffer resource
 #ifndef __HIP_DEVICE_COMPILE__ // for host code
@@ -67,7 +70,7 @@
 #define CK_BUFFER_RESOURCE_3RD_DWORD 0x00020000
 #elif defined(__gfx103__)
 #define CK_BUFFER_RESOURCE_3RD_DWORD 0x31014000
-#elif defined(__gfx11__)
+#elif defined(__gfx11__) || defined(__gfx12__)
 #define CK_BUFFER_RESOURCE_3RD_DWORD 0x31004000
 #endif
 
@@ -80,7 +83,7 @@
 #define CK_USE_AMD_V_FMAC_F32
 #define CK_USE_AMD_V_DOT2_F32_F16
 #define CK_USE_AMD_V_DOT4_I32_I8
-#elif defined(__gfx11__)
+#elif defined(__gfx11__) || defined(__gfx12__)
 #define CK_USE_AMD_V_FMAC_F32
 #define CK_USE_AMD_V_DOT2_F32_F16
 #define CK_USE_AMD_V_DOT4_I32_I8_GFX11
@@ -152,7 +155,7 @@
 #define CK_USE_SR_F8_CONVERSION 1
 
 // block synchronization only s_wait lgkmcnt(0), not vmcnt(0)
-#define CK_EXPERIMENTAL_BLOCK_SYNC_LDS_WITHOUT_SYNC_VMEM 1
+#define CK_EXPERIMENTAL_BLOCK_SYNC_LDS_WITHOUT_SYNC_VMEM 0
 
 // experimental feature: multi index implemented as array
 #define CK_EXPERIMENTAL_USE_DYNAMICALLY_INDEXED_MULTI_INDEX 0

include/ck/host_utility/device_prop.hpp

@@ -85,4 +85,9 @@ inline bool is_navi3_supported()
            ck::get_device_name() == "gfx1102" || ck::get_device_name() == "gfx1103";
 }
 
+inline bool is_navi4_supported()
+{
+	return ck::get_device_name() == "gfx1200";
+}
+
 } // namespace ck

include/ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp

@@ -537,7 +537,7 @@ struct DeviceGemmDl : public DeviceGemm<ALayout,
         }
 
         if(ck::get_device_name() == "gfx906" || ck::is_navi2_supported() ||
-           ck::is_navi3_supported())
+           ck::is_navi3_supported() || ck::is_navi4_supported())
         {
             return GridwiseGemm::CheckValidity(
                 arg.a_grid_desc_k0_m_k1_, arg.b_grid_desc_k0_n_k1_, arg.c_grid_desc_m_n_);