Revert "32x320 blockscale fp8 kernel"

This reverts commit 0eb4e16.

Author: JohnNikolay84

aiter/fused_moe.py

@@ -457,7 +457,7 @@ def FinalFunc():
             in fused_moe_1stage_dict[get_gfx()]
         ):
             if q_type == QuantType.per_1x128:
-                run_1stage = True and (inter_dim % 256 == 0 or inter_dim % 320 == 0) and (token > 31)
+                run_1stage = True and (inter_dim % 256 == 0) and (token > 31)
             elif q_type == QuantType.per_Token and q_dtype_w in [dtypes.i8, dtypes.fp8]:
                 run_1stage = token > 32
             else:
@@ -483,8 +483,6 @@ def FinalFunc():
         f"[fused_moe] using {'1stage' if run_1stage else '2stage'} {'default' if cfg is None else tag} for {keys} "
     )
 
-    print('MyTag: ',tag)
-
     if (
         "ck" in kernelName1
         or q_dtype_w
@@ -497,8 +495,7 @@ def FinalFunc():
         or (q_dtype_w == dtypes.fp8 and q_type == QuantType.per_1x128)
         or (q_type == QuantType.per_1x128 and block_m == 16)
     ):
-        if inter_dim % 320 != 0:
-          return MOEMetadata(
+        return MOEMetadata(
             functools.partial(
                 aiter.ck_moe_stage1_fwd,
                 kernelName=kernelName1,
@@ -514,15 +511,14 @@ def FinalFunc():
             block_m,
             ksplit,
             run_1stage,
-          )
+        )
 
     # TODO: remove when stage2 support more size
     tmpList = [32, 64, 128]
     if block_m not in tmpList:
         tag = ""
         block_m = ([el for el in tmpList if block_m < el] + [128])[0]
 
-    print(run_1stage)
     return MOEMetadata(
         functools.partial(
             asm_stage1,

aiter/ops/moe_op.py

@@ -419,7 +419,6 @@ def ck_moe_stage2_fwd(
         mul_routed_weight_stage,
     )
 
-    print(kernelName)
     ck_moe_stage2(
         inter_states,
         w1,

csrc/py_itfs_cu/asm_fmoe.cu

@@ -822,7 +822,7 @@ void fmoe_fp8_blockscale_g1u1(torch::Tensor& out,               // [token_cnt, d
     int sub_X_cnt            = sorted_expert_ids.size(0);
     const char* enable_vskip = std::getenv("AITER_ENABLE_VSKIP");
 
-    if(out.dtype() == at::ScalarType::BFloat16 && (inter_dim % 256 == 0 || inter_dim % 320 == 0) && fc_scale_blkn == 128 &&
+    if(out.dtype() == at::ScalarType::BFloat16 && inter_dim % 256 == 0 && fc_scale_blkn == 128 &&
        fc_scale_blkk == 128)
     {
         if(activation == ActivationType::Silu)

hsa/gfx942/fmoe/gelu/fmoe_bf16_blockscaleFp8_g1u1_gelu.csv

@@ -3,4 +3,3 @@ _ZN5aiter47fmoe_bf16_blockscaleFp8_g1u1_vs_gelu_1tg_32x256E,fmoe_bf16_blockscale
 _ZN5aiter50fmoe_bf16_blockscaleFp8_g1u1_vs_gelu_1tg_ps_32x256E,fmoe_bf16_blockscaleFp8_g1u1_vs_gelu_1tg_ps_32x256.co,0,1,0,1,1,32,256
 _ZN5aiter49fmoe_bf16_blockscaleFp8_g1u1_novs_gelu_1tg_32x256E,fmoe_bf16_blockscaleFp8_g1u1_novs_gelu_1tg_32x256.co,0,0,0,1,0,32,256
 _ZN5aiter52fmoe_bf16_blockscaleFp8_g1u1_novs_gelu_1tg_ps_32x256E,fmoe_bf16_blockscaleFp8_g1u1_novs_gelu_1tg_ps_32x256.co,0,0,0,1,1,32,256
-_ZN5aiter40fmoe_bf16_blockscaleFp8_g1u1_gelu_32x320E,fmoe_bf16_blockscaleFp8_g1u1_gelu_32x320.co,0,0,0,1,0,32,320

hsa/gfx942/fmoe/gelu/fmoe_bf16_blockscaleFp8_g1u1_gelu_32x320.co

@@ -3,4 +3,3 @@ _ZN5aiter47fmoe_bf16_blockscaleFp8_g1u1_vs_silu_1tg_32x256E,fmoe_bf16_blockscale
 _ZN5aiter52fmoe_bf16_blockscaleFp8_g1u1_novs_silu_1tg_ps_32x256E,fmoe_bf16_blockscaleFp8_g1u1_novs_silu_1tg_ps_32x256.co,0,0,0,1,1,32,256
 _ZN5aiter50fmoe_bf16_blockscaleFp8_g1u1_vs_silu_1tg_ps_32x256E,fmoe_bf16_blockscaleFp8_g1u1_vs_silu_1tg_ps_32x256.co,0,1,0,1,1,32,256
 _ZN5aiter49fmoe_bf16_blockscaleFp8_g1u1_novs_silu_1tg_32x256E,fmoe_bf16_blockscaleFp8_g1u1_novs_silu_1tg_32x256.co,0,0,0,1,0,32,256
-_ZN5aiter40fmoe_bf16_blockscaleFp8_g1u1_silu_32x320E,fmoe_bf16_blockscaleFp8_g1u1_silu_32x320.co,0,0,0,1,0,32,320