CMF v0 FP8 triton tune (#4873)

Summary:
X-link: facebookresearch/FBGEMM#1895

Pull Request resolved: #4873

Boost Triton FP8 perf for CMF v0 model, shape (2048, 5120, 5376)

Reviewed By: RandySheriff

Differential Revision: D82254779

fbshipit-source-id: e13bd8a68195904db37db8cc8748226a77564db7

Author: JChunX

fbgemm_gpu/experimental/gemm/triton_gemm/fp8_gemm.py

@@ -3868,6 +3868,20 @@ def get_full_non_persistent_tuning_space():
         num_warps=8,
         num_stages=2,
     ),
+    triton.Config(
+        {
+            "BLOCK_M": 256,
+            "BLOCK_N": 256,
+            "BLOCK_K": 128,
+            "GROUP_M": 2,
+            "SPLIT_K": 1,
+            "waves_per_eu": 0,
+            "matrix_instr_nonkdim": 16,
+            "kpack": 1,
+        },
+        num_warps=8,
+        num_stages=2,
+    ),
     triton.Config(
         {
             "BLOCK_M": 256,
@@ -3876,12 +3890,26 @@ def get_full_non_persistent_tuning_space():
             "GROUP_M": 2,
             "SPLIT_K": 1,
             "waves_per_eu": 2,
-            "matrix_instr_nonkdim": 32,
-            "kpack": 2,
+            "matrix_instr_nonkdim": 16,
+            "kpack": 1,
         },
         num_warps=8,
         num_stages=2,
     ),
+    triton.Config(
+        {
+            "BLOCK_M": 128,
+            "BLOCK_N": 256,
+            "BLOCK_K": 64,
+            "GROUP_M": 2,
+            "SPLIT_K": 1,
+            "waves_per_eu": 2,
+            "matrix_instr_nonkdim": 16,
+            "kpack": 1,
+        },
+        num_warps=4,
+        num_stages=2,
+    ),
     triton.Config(
         {
             "BLOCK_M": 256,