Revert "Add SwapAB Optimization for triton fused_moe_kernel on SM90." (#16676)

python/sglang/srt/layers/moe/fused_moe_triton/fused_moe_triton_kernels.py

@@ -1,6 +1,5 @@
 from __future__ import annotations
 
-import functools
 import os
 from typing import Any, Dict, List, Optional
 
@@ -21,11 +20,9 @@ from sglang.srt.layers.quantization.int8_kernel import (
 from sglang.srt.utils import (
     cpu_has_amx_support,
     get_bool_env_var,
-    get_device_name,
     is_cpu,
     is_cuda,
     is_hip,
-    is_sm90_supported,
 )
 
 try:
@@ -55,24 +52,6 @@ def support_tensor_descriptor():
     return _support_tensor_descriptor
 
 
-# In theory, swap_ab should benefit all SM90 GPUs.
-# However, since it has only been verified on H20 (not H100/H200),
-# it is currently enabled only on H20.
-@functools.lru_cache(maxsize=8)
-def should_enable_swap_ab(
-    BLOCK_SIZE_M: int,
-    BLOCK_SIZE_N: int,
-) -> bool:
-    device_name = get_device_name()
-    is_h20_device = device_name and "H20" in device_name and "H200" not in device_name
-    return (
-        is_h20_device
-        and is_sm90_supported()
-        and BLOCK_SIZE_M < 64
-        and BLOCK_SIZE_N >= 64
-    )
-
-
 @triton.jit
 def write_zeros_to_output(
     c_ptr,
@@ -381,7 +360,6 @@ def fused_moe_kernel(
     even_Ks: tl.constexpr,
     c_sorted: tl.constexpr,
     filter_expert: tl.constexpr,
-    swap_ab: tl.constexpr,
 ):
     """
     Implements the fused computation for a Mixture of Experts (MOE) using
@@ -520,9 +498,6 @@ def fused_moe_kernel(
     # We accumulate into a `[BLOCK_SIZE_M, BLOCK_SIZE_N]` block
     # of fp32 values for higher accuracy.
     # `accumulator` will be converted back to fp16 after the loop.
-    if swap_ab:
-        accumulator = tl.zeros((BLOCK_SIZE_N, BLOCK_SIZE_M), dtype=tl.float32)
-    else:
     accumulator = tl.zeros((BLOCK_SIZE_M, BLOCK_SIZE_N), dtype=tl.float32)
 
     for k_start in range(0, K, BLOCK_SIZE_K):
@@ -564,17 +539,12 @@ def fused_moe_kernel(
                     a_scale_ptrs + offs_ks * stride_ask, mask=token_mask, other=0.0
                 )
                 b_scale = tl.load(b_scale_ptrs + offs_ks * stride_bsk)
-                if swap_ab:
-                    a, b = tl.trans(b, (1, 0)), tl.trans(a, (1, 0))
-                    a_scale, b_scale = b_scale, a_scale
                 if BLOCK_SIZE_N > group_n:
                     accumulator += tl.dot(a, b) * a_scale[:, None] * b_scale[None, :]
                 else:
                     accumulator += tl.dot(a, b) * (a_scale[:, None] * b_scale)
             else:
                 if use_fp8_w8a8:
-                    if swap_ab:
-                        a, b = tl.trans(b, (1, 0)), tl.trans(a, (1, 0))
                     accumulator = tl.dot(a, b, acc=accumulator)
                 else:
                     accumulator += tl.dot(a, b)
@@ -586,9 +556,6 @@ def fused_moe_kernel(
         if b_desc is None:
             b_ptrs += BLOCK_SIZE_K * stride_bk
 
-    if swap_ab:
-        accumulator = tl.trans(accumulator, (1, 0))
-
     if use_int8_w8a16:
         accumulator *= b_scale
     elif use_fp8_w8a8 or use_int8_w8a8:
@@ -648,11 +615,6 @@ def invoke_fused_moe_kernel(
     assert topk_weights.stride(1) == 1
     assert sorted_token_ids.stride(0) == 1
 
-    if use_fp8_w8a8:
-        swap_ab = should_enable_swap_ab(config["BLOCK_SIZE_M"], config["BLOCK_SIZE_N"])
-    else:
-        swap_ab = False
-
     padded_size = 0
     if use_fp8_w8a8:
         assert B_scale is not None
@@ -824,7 +786,6 @@ def invoke_fused_moe_kernel(
             even_Ks=even_Ks,
             c_sorted=c_sorted,
             filter_expert=filter_expert,
-            swap_ab=swap_ab,
             **config,
         )