only apply asm for shuffle for single precision

CNugteren · Nov 1, 2023 · ed32af0 · ed32af0
1 parent 27648f5
commit ed32af0
Show file tree

Hide file tree

Showing 2 changed files with 7 additions and 6 deletions.
diff --git a/src/kernels/level3/xgemm_part1.opencl b/src/kernels/level3/xgemm_part1.opencl
@@ -139,15 +139,15 @@ R"(
 #endif
 
 #if USE_SUBGROUP_SHUFFLING == 1 && SUBGROUP_SHUFFLING_GCN == 1
-  #define SUBGROUP_SIZE 32              // Assumes subgroup size is always 4 on AMD GCN GPUs
+  #define SUBGROUP_SIZE 32    // Assumes subgroup size is always 32 on AMD Navi GPUs
     #define NAVI_SHFL(s0, l)  \
     { \
 		__asm ( \
-		  "ds_bpermute_b32  %[dos0], %[ol0], %[os0]\n" \
+		  "ds_bpermute_b32  %[d], %[l], %[s]\n" \
        "s_waitcnt lgkmcnt(0)\n" \
-		  : [dos0] "=&v" (s0) \
-		  : [ol0] "v" (l), \
-            [os0] "0" (s0)); \
+		  : [d] "=&v" (s0) \
+		  : [l] "v" (l), \
+            [s] "0" (s0)); \
 	 }
    #define NAVI_LID() \
      if (get_work_dim() == 2) { \

diff --git a/src/utilities/compile.cpp b/src/utilities/compile.cpp
@@ -78,7 +78,8 @@ std::shared_ptr<Program> CompileFromSource(
     }
   }
 
-  if (device.IsGPU() && device.IsAMD() && device.Name().find("gfx1") != std::string::npos) {
+  if (device.IsGPU() && device.IsAMD() && device.Name().find("gfx1") != std::string::npos &&
+      precision == Precision::kSingle) { // only for Navi cards (gfx1XXX)
     header_string += "#define USE_SUBGROUP_SHUFFLING 1\n";
     header_string += "#define SUBGROUP_SHUFFLING_GCN 1\n";
   }