Merge branch 'andriy/lwpck-2243' into andriy/lwpck-2388

CMakePresets.json

@@ -11,7 +11,6 @@
             "cacheVariables": {
                 "CMAKE_BUILD_TYPE": "Debug",
                 "CMAKE_EXPORT_COMPILE_COMMANDS": "ON",
-                "GPU_TARGETS": "gfx950",
                 "BUILD_DEV": "ON",
                 "CMAKE_CXX_COMPILER": "/opt/rocm/llvm/bin/clang++",
                 "CMAKE_PREFIX_PATH": "/opt/rocm"
@@ -32,6 +31,7 @@
                 "NONE": ""
             },
             "cacheVariables": {
+                "GPU_TARGETS": "gfx950",
                 "CMAKE_BUILD_TYPE": "Debug",
                 "CMAKE_CXX_FLAGS": "-O0 -ggdb"
             }
@@ -39,8 +39,29 @@
         {
             "name": "MI355-release",
             "displayName": "MI355 Release",
-            "inherits": "MI355-debug",
+            "inherits": "linux-debug",
+            "cacheVariables": {
+                "GPU_TARGETS": "gfx950",
+                "CMAKE_BUILD_TYPE": "Release",
+                "CMAKE_CXX_FLAGS": "-O3"
+            }
+        },
+        {
+            "name": "MI300X-release",
+            "displayName": "MI300X Release",
+            "inherits": "linux-debug",
+            "cacheVariables": {
+                "GPU_TARGETS": "gfx942",
+                "CMAKE_BUILD_TYPE": "Release",
+                "CMAKE_CXX_FLAGS": "-O3"
+            }
+        },
+        {
+            "name": "MI250-release",
+            "displayName": "MI250 Release",
+            "inherits": "linux-debug",
             "cacheVariables": {
+                "GPU_TARGETS": "gfx90a",
                 "CMAKE_BUILD_TYPE": "Release",
                 "CMAKE_CXX_FLAGS": "-O3"
             }
@@ -76,6 +97,26 @@
                 "Release"
             ],
             "jobs": 128
+        },
+        {
+            "name": "MI300X-release",
+            "displayName": "MI300X",
+            "configurePreset": "MI300X-release",
+            "description": "Build Environment for MI300X Release.",
+            "inherits": [
+                "Release"
+            ],
+            "jobs": 128
+        },
+        {
+            "name": "MI250-release",
+            "displayName": "MI250",
+            "configurePreset": "MI250-release",
+            "description": "Build Environment for MI250 Release.",
+            "inherits": [
+                "Release"
+            ],
+            "jobs": 128
         }
     ]
 }

include/ck/utility/amd_ck_fp8.hpp

@@ -276,7 +276,7 @@ struct f8_ocp_t
     __host__ explicit operator float() const
 #endif
     {
-#if CK_FP8_CVT_FAST_PATH
+#if defined(__gfx950__) || defined(__gfx1200__) || defined(__gfx1201__)
         return fp8_impl::cast_to_f32_from_f8<default_interpret>(this->data);
 #else
         return fp8_impl::cast_from_f8<float, wm, we, false>(
@@ -290,7 +290,7 @@ struct f8_ocp_t
     __host__ explicit operator _Float16() const
 #endif
     {
-#if CK_FP8_CVT_FAST_PATH
+#if defined(__gfx950__) || defined(__gfx1200__) || defined(__gfx1201__)
         return static_cast<_Float16>(fp8_impl::cast_to_f32_from_f8<default_interpret>(this->data));
 #else
         return fp8_impl::cast_from_f8<_Float16, wm, we, false>(
@@ -322,7 +322,7 @@ struct bf8_ocp_t
     __host__ explicit operator float() const
 #endif
     {
-#if CK_FP8_CVT_FAST_PATH
+#if defined(__gfx950__) || defined(__gfx1200__) || defined(__gfx1201__)
         return fp8_impl::cast_to_f32_from_f8<default_interpret>(this->data);
 #else
         return fp8_impl::cast_from_f8<float, wm, we, false>(
@@ -336,7 +336,7 @@ struct bf8_ocp_t
     __host__ explicit operator _Float16() const
 #endif
     {
-#if CK_FP8_CVT_FAST_PATH
+#if defined(__gfx950__) || defined(__gfx1200__) || defined(__gfx1201__)
         return static_cast<_Float16>(fp8_impl::cast_to_f32_from_f8<default_interpret>(this->data));
 #else
         return fp8_impl::cast_from_f8<_Float16, wm, we, false>(