[HGEMM] Refactor HGEMM WMMA 161616 kernels (#96)

* update hgemm benchmark option

* update hgemm benchmark option

* update hgemm benchmark option

* remove un-need hgemm codes

Author: DefTruth

README.md

@@ -137,18 +137,16 @@
 | ✔️ [hgemm_t_8x8_sliced_k_f16x4](./hgemm/hgemm.cu)|f16|f16|[link](./hgemm/)|⭐️⭐️⭐️|  
 | ✔️ [hgemm_t_8x8_sliced_k_f16x4_pack](./hgemm/hgemm.cu)|f16|f16|[link](./hgemm/)|⭐️⭐️⭐️|  
 | ✔️ [hgemm_t_8x8_sliced_k_f16x8_pack](./hgemm/hgemm.cu)|f16|f16|[link](./hgemm/)|⭐️⭐️⭐️|  
-| ✔️ [hgemm_t_8x8_sliced_k...bcf](./hgemm/hgemm.cu)|f16|f16|[link](./hgemm/)|⭐️⭐️⭐️|  
 | ✔️ [hgemm_t_8x8_sliced_k...dbuf](./hgemm/hgemm.cu)|f16|f16|[link](./hgemm/)|⭐️⭐️⭐️|  
 | ✔️ [hgemm_t_8/16x8...k16/32...dbuf](./hgemm/hgemm_async.cu)|f16|f16|[link](./hgemm/)|⭐️⭐️⭐️|  
 | ✔️ [hgemm_t_8/16x8...k16/32...async](./hgemm/hgemm_async.cu)|f16|f16|[link](./hgemm/)|⭐️⭐️⭐️|  
 | ✔️ [hgemm_wmma_m16n16k16...naive*](./hgemm/hgemm_wmma.cu)|f16|f16|[link](./hgemm/)|⭐️⭐️⭐️|  
 | ✔️ [hgemm_wmma_m16n16k16...mma4x2*](./hgemm/hgemm_wmma.cu)|f16|f16|[link](./hgemm/)|⭐️⭐️⭐️|  
-| ✔️ [hgemm_wmma_m16n16k16...warp2x4*](./hgemm/hgemm_wmma.cu)|f16|f16|[link](./hgemm/)|⭐️⭐️⭐️|  
-| ✔️ [hgemm_wmma_m16n16k16...async*](./hgemm/hgemm_wmma.cu)|f16|f16|[link](./hgemm/)|⭐️⭐️⭐️|  
+| ✔️ [hgemm_wmma_m16n16k16...mma4x4*](./hgemm/hgemm_wmma.cu)|f16|f16|[link](./hgemm/)|⭐️⭐️⭐️|  
 | ✔️ [hgemm_wmma_m16n16k16...dbuf*](./hgemm/hgemm_wmma.cu)|f16|f16|[link](./hgemm/)|⭐️⭐️⭐️|  
+| ✔️ [hgemm_wmma_m32n8k16....dbuf*](./hgemm/hgemm_wmma.cu)|f16|f16|[link](./hgemm/)|⭐️⭐️⭐️|  
 | ✔️ [hgemm_wmma_m16n16k16...stages*](./hgemm/hgemm_wmma_stage.cu)|f16|f16|[link](./hgemm/)|⭐️⭐️⭐️|  
 | ✔️ [hgemm_wmma_m16n16k16...swizzle*](./hgemm/hgemm_wmma_stage.cu)|f16|f16|[link](./hgemm/)|⭐️⭐️⭐️|  
-| ✔️ [hgemm_wmma_m32n8k16...dbuf*](./hgemm/hgemm_wmma.cu)|f16|f16|[link](./hgemm/)|⭐️⭐️⭐️|  
 | ✔️ [sgemv_k32_f32](./sgemv/sgemv.cu)|f32|f32|[link](./sgemv/)|⭐️⭐️⭐️|  
 | ✔️ [sgemv_k128_f32x4](./sgemv/sgemv.cu)|f32|f32|[link](./sgemv/)|⭐️⭐️⭐️|  
 | ✔️ [sgemv_k16_f32](./sgemv/sgemv.cu)|f32|f32|[link](./sgemv/)|⭐️⭐️⭐️|  

hgemm/README.md

@@ -9,7 +9,6 @@
 - [X] hgemm_t_8x8_sliced_k_f16x4_pack_kernel(pack 16x4)
 - [X] hgemm_t_8x8_sliced_k_f16x4_bcf_kernel(bank conflicts reduce)
 - [X] hgemm_t_8x8_sliced_k_f16x4_pack_bcf_kernel(bank conflicts reduce, pack)
-- [X] hgemm_t_4x4_sliced_k_f16x4_pack_bcf_kernel(bank conflicts reduce, pack)
 - [X] hgemm_t_8x8_sliced_k_f16x8_pack_bcf_kernel(bank conflicts reduce, pack)
 - [X] hgemm_t_8x8_sliced_k_f16x8_pack_bcf_dbuf_kernel(bank conflicts reduce, pack, double buffers)
 - [X] hgemm_t_8x8_sliced_k16/32_f16x8_pack_bcf_dbuf_kernel(pack, double buffers)
@@ -20,11 +19,10 @@
 - [X] hgemm_wmma_m16n16k16_mma4x2_warp2x4_async(WMMA, Tile MMA/Warp, Copy Async) 
 - [X] hgemm_wmma_m16n16k16_mma4x2_warp2x4_async_offset(WMMA, Tile MMA/Warp, Copy Async, Pad)
 - [X] hgemm_wmma_m16n16k16_mma4x2_warp2x4_dbuf_async(WMMA, Tile MMA/Warp, Copy Async, Double Buffers, Pad)  
-- [X] hgemm_wmma_m16n16k16_mma4x4_warp2x2x2_dbuf_async(WMMA, Tile MMA/Warp, Copy Async, Double Buffers, Pad)  
-- [X] hgemm_wmma_m32n8k16_mma2x4_warp2x4_dbuf_async(WMMA, Tile MMA/Warp, Copy Async, Double Buffers, Pad)
-- [X] hgemm_wmma_m16n16k16_mma4x2_warp2x4x2_rbuf_async(WMMA, Tile MMA/Warp, Copy Async, Double/Reg Buffers, Pad)
 - [X] hgemm_wmma_m16n16k16_mma4x2_warp2x4_stages(WMMA, Tile MMA/Warp, Copy Async, Stages, Pad, Block swizzle) 
+- [X] hgemm_wmma_m16n16k16_mma4x2_warp4x4_stages(WMMA, Tile MMA/Warp, Copy Async, Stages, Pad, Block swizzle)
 - [X] hgemm_wmma_m16n16k16_mma4x4_warp4x4_stages(WMMA, Tile MMA/Warp, Copy Async, Stages, Pad, Block swizzle) 
+- [X] hgemm_wmma_m32n8k16_mma2x4_warp2x4_dbuf_async(WMMA, Tile MMA/Warp, Copy Async, Double Buffers, Pad)
 - [X] PyTorch bindings
 
 ## 目前性能