remove atten kernel workarounds as we move over to rocm 5.3 (#496)

include/ck/ck.hpp

@@ -150,26 +150,10 @@
 // workaround: compiler gnerating inefficient ds_write instructions
 #define CK_WORKAROUND_SWDEV_XXXXXX_INT8_DS_WRITE_ISSUE 1
 
-// (gfx908 only) workaround: compiler crash in fused kernels on mainline #9110; #10738 seems ok
-// error message was "fatal error: error in backend: Error while trying to spill VGPR0 from class
-// VGPR_32: Cannot scavenge register without an emergency spill slot!"
-// this fall back to less ideal way of handle NPadding in fused attention kernel
-#ifdef __gfx908__
-#define CK_WORKAROUND_SWDEV_XXXXXX_ATTN_KERNEL_CLANG_CANNOT_SCAVENGE_REGISTER 1
-#else
-// for __gfx90a__, ...
-#define CK_WORKAROUND_SWDEV_XXXXXX_ATTN_KERNEL_CLANG_CANNOT_SCAVENGE_REGISTER 0
-#endif // __gfx908__
-
 // workaround: verifaction failure, due to compiler regression, for conv bwd-data fp16 using some
 // tuning parameter
 #define CK_WORKAROUND_SWDEV_325164 0
 
-// workaround: disable broken fused attention kernel instance that does not pass validation
-// issue found on mi100/#10738 combo when irregular KPerBlock attention kernel has acc0 scaling
-// enabled
-#define CK_WORKAROUND_DISABLE_BROKEN_ATTN_KERNEL_INSTANCE 1
-
 namespace ck {
 
 enum struct InMemoryDataOperationEnum

include/ck/tensor_operation/gpu/grid/gridwise_batched_gemm_softmax_gemm_xdl_cshuffle_v1.hpp

@@ -694,12 +694,7 @@ struct GridwiseBatchedGemmSoftmaxGemm_Xdl_CShuffle
                                                   FloatGemmAcc,
                                                   decltype(threadid_to_m_n_thread_cluster_adaptor),
                                                   decltype(thread_cluster_desc_m_n),
-                                                  decltype(thread_slice_desc_m_n)
-#if CK_WORKAROUND_SWDEV_XXXXXX_ATTN_KERNEL_CLANG_CANNOT_SCAVENGE_REGISTER
-                                                      ,
-                                                  true
-#endif
-                                                  >{};
+                                                  decltype(thread_slice_desc_m_n)>{};
 
         const index_t num_gemm1_k_block_outer_loop =
             b_grid_desc_bk0_n_bk1.GetLength(I1) / NPerBlock;

library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm/device_batched_gemm_softmax_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp

@@ -66,9 +66,7 @@ using device_batched_gemm_softmax_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_
         //#######################################|        |         |         |        |      |       |       |      |        |         |            |            |            |            |            |               |         |      |       |      |      |      |      |    |    |     |     |     |  Wave|  Wave|  Wave|                |               |               |               |               |               |          |                 |                |                |                |                |                |           |                 |                |                |                |                |                |           |            |            |                             |                |           |
         DeviceBatchedGemmSoftmaxGemm_Xdl_CShuffle<     Row,      Col,      Row,     Row,   F16,    F16,    F16,   F16,     F32,      F16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,     GemmPadded,        1,   256,    256,   128,    40,    64,    32,   4,   4,    2,   32,   32,     2,     4,     2,     S<2,128, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,     false,      S<2,128, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               4,               4,      false,     S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8,   Masking>,
         DeviceBatchedGemmSoftmaxGemm_Xdl_CShuffle<     Row,      Col,      Row,     Row,   F16,    F16,    F16,   F16,     F32,      F16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,     GemmPadded,        1,   256,    256,   128,    40,   128,    32,   4,   4,    2,   32,   32,     2,     4,     4,     S<2,128, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,     false,      S<2,128, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               4,               4,      false,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8,   Masking>,
-#if CK_WORKAROUND_DISABLE_BROKEN_ATTN_KERNEL_INSTANCE == 0
         DeviceBatchedGemmSoftmaxGemm_Xdl_CShuffle<     Row,      Col,      Row,     Row,   F16,    F16,    F16,   F16,     F32,      F16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,     GemmPadded,        1,   256,    128,   256,    40,    64,    32,   4,   4,    2,   32,   32,     1,     8,     2,     S<2,128, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,     false,      S<2,128, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               4,               4,      false,     S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8,   Masking>,
-#endif
         DeviceBatchedGemmSoftmaxGemm_Xdl_CShuffle<     Row,      Col,      Row,     Row,   F16,    F16,    F16,   F16,     F32,      F16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,     GemmPadded,        1,   256,    128,   256,    40,   128,    32,   4,   4,    2,   32,   32,     1,     8,     4,     S<2,128, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,     false,      S<2,128, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               4,               4,      false,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8,   Masking>,
         DeviceBatchedGemmSoftmaxGemm_Xdl_CShuffle<     Row,      Col,      Row,     Row,   F16,    F16,    F16,   F16,     F32,      F16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,     GemmPadded,        1,   256,    128,   128,    40,    64,    32,   4,   4,    2,   32,   32,     1,     4,     2,     S<2,128, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,     false,      S<2,128, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               4,               4,      false,     S<16, 16, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8,   Masking>,
         DeviceBatchedGemmSoftmaxGemm_Xdl_CShuffle<     Row,      Col,      Row,     Row,   F16,    F16,    F16,   F16,     F32,      F16, PassThrough, PassThrough,       Scale, PassThrough, PassThrough,     GemmPadded,        1,   256,    128,   128,    40,   128,    32,   4,   4,    2,   32,   32,     1,     4,     4,     S<2,128, 1>,     S<1, 0, 2>,     S<1, 0, 2>,              2,              4,              4,     false,      S<2,128, 1>,      S<1, 0, 2>,      S<1, 0, 2>,               2,               4,               4,      false,     S< 8, 32, 1>,      S<0, 2, 1>,      S<0, 2, 1>,               1,               4,               2,      false,           1,           2,               S<1, 32, 1, 8>,               8,   Masking>