[Kernel][Triton][AMD] Use block size heuristic for avg 2.8x speedup f…

…or int8 models (#11698)

Signed-off-by: Randall Smith <[email protected]>

vllm/model_executor/layers/quantization/compressed_tensors/triton_scaled_mm.py

@@ -128,7 +128,8 @@ def triton_scaled_mm(input: torch.Tensor,
                      bias: Optional[torch.Tensor] = None,
                      block_size_m: int = 32,
                      block_size_n: int = 32,
-                     block_size_k: int = 32) -> torch.Tensor:
+                     block_size_k: int = 32,
+                     use_heuristic=True) -> torch.Tensor:
     M, K = input.shape
     N = weight.shape[1]
 
@@ -152,6 +153,20 @@ def triton_scaled_mm(input: torch.Tensor,
 
     has_scalar = lambda x: x.shape[0] == 1 and x.shape[1] == 1
 
+    if use_heuristic:
+        is_small_N = N < 8192
+        next_power_of_2_M = max(32, triton.next_power_of_2(M))
+        if next_power_of_2_M <= 32:
+            tile_shape = (64, 64, 256) if is_small_N else (64, 128, 256)
+        elif next_power_of_2_M <= 64:
+            tile_shape = (64, 64, 256)
+        elif next_power_of_2_M <= 128:
+            tile_shape = (64, 128, 128)
+        else:
+            tile_shape = (128, 128, 128)
+
+    block_size_m, block_size_n, block_size_k = tile_shape
+
     block_size_sa = 1 if has_scalar(scale_a) else block_size_m
     block_size_sb = 1 if has_scalar(scale_b) else block_size_n