[A8W8 GEMM] Optimized weight-preshuffled implementation & add quantization datatype for CK TILE rms_norm

Jenkinsfile

@@ -722,9 +722,6 @@ CRON_SETTINGS = BRANCH_NAME == "develop" ? '''0 23 * * * % RUN_FULL_QA=true;ROCM
 
 pipeline {
     agent none
-    triggers {
-        parameterizedCron(CRON_SETTINGS)
-    }
     options {
         parallelsAlwaysFailFast()
     }

example/65_gemm_multiply_multiply/CMakeLists.txt

@@ -1,4 +1,7 @@
 add_example_executable(example_gemm_multiply_multiply_xdl_fp8 gemm_multiply_multiply_xdl_fp8.cpp)
+# target_compile_options(example_gemm_multiply_multiply_xdl_fp8 PRIVATE -mllvm -greedy-reverse-local-assignment=1 -save-temps=$PWD -Wno-gnu-line-marker)
 add_example_executable(example_gemm_multiply_multiply_xdl_fp8_ab_scale gemm_multiply_multiply_xdl_fp8_ab_scale.cpp)
+add_example_executable(example_gemm_multiply_multiply_xdl_fp8_bpreshuffle gemm_multiply_multiply_xdl_fp8_bpreshuffle.cpp)
+# target_compile_options(example_gemm_multiply_multiply_xdl_fp8_bpreshuffle PRIVATE -mllvm -greedy-reverse-local-assignment=1 -save-temps=$PWD -Wno-gnu-line-marker)
 add_example_executable(example_gemm_add_add_xdl_fp16 gemm_add_add_xdl_fp16.cpp)
-add_example_executable(example_gemm_multiply_multiply_xdl_int8 gemm_multiply_multiply_xdl_int8.cpp)
+add_example_executable(example_gemm_multiply_multiply_xdl_int8 gemm_multiply_multiply_xdl_int8.cpp)

example/65_gemm_multiply_multiply/gemm_multiply_multiply_xdl_fp8.cpp

@@ -69,18 +69,21 @@ using AElementOp   = PassThrough;
 using BElementOp   = PassThrough;
 using CDEElementOp = MultiplyMultiply;
 
-static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::MNPadding;
+static constexpr auto GemmSpec = ck::tensor_operation::device::GemmSpecialization::Default;
 
 using DeviceOpInstance = ck::tensor_operation::device::DeviceGemmMultiD_Xdl_CShuffle_V3
     // clang-format off
-///######|  ALayout|  BLayout| DsLayout| ELayout|      AData|      BData|     DsData|     EData|     AccData|         CShuffle|           A|           B|          CDE|           GEMM| Block|  MPer|  NPer|  KPer| AK1| BK1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds|    CShuffle|    CShuffle| CBlockTransferClusterLengths|  CBlockTransfer|
-///######|         |         |         |        |       Type|       Type|       Type|      Type|        Type|         DataType| Elementwise| Elementwise|  Elementwise| Spacialization|  Size| Block| Block| Block|    |    |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| MXdlPerWave| NXdlPerWave|         _MBlock_MWaveMPerXdl| ScalarPerVector|
-///######|         |         |         |        |           |           |           |          |            |                 |   Operation|   Operation|    Operation|               |      |      |      |      |    |    |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |  PerShuffle|  PerShuffle|         _NBlock_NWaveNPerXdl|   _NWaveNPerXdl|
-///######|         |         |         |        |           |           |           |          |            |                 |            |            |             |               |      |      |      |      |    |    |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |            |            |                             |    S<C, D0, D1>|
-///###### RRR
-      ///<      Row,      Row, DsLayout, ELayout, A0DataType, B0DataType, DsDataType, EDataType, AccDataType, CShuffleDataType,  AElementOp,  BElementOp, CDEElementOp,       GemmSpec,   256,   256,   128,    64,  16,   4,  32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,    S<16, 16, 1>,    S<0, 2, 1>,     S<0, 2, 1>,             1,               8,              4,          0,          1,           1,               S<1, 32, 1, 8>,      S<8, 8, 1>,  ck::BlockGemmPipelineScheduler::Interwave, ck::BlockGemmPipelineVersion::v1, FP8>;
-///###### RCR
-         <      Row,      Col, DsLayout, ELayout, A0DataType, B0DataType, DsDataType, EDataType, AccDataType, CShuffleDataType,  AElementOp,  BElementOp, CDEElementOp,       GemmSpec,   256,   256,   128,    64,  16,  16,  32,   32,    4,    2,     S<4, 64, 1>,     S<1, 0, 2>,    S<1, 0, 2>,               2,             16,             16,          0,     S<4, 64, 1>,    S<1, 0, 2>,     S<1, 0, 2>,             2,              16,             16,          0,          1,           1,               S<1, 32, 1, 8>,      S<8, 8, 1>,  ck::BlockGemmPipelineScheduler::Interwave, ck::BlockGemmPipelineVersion::v1, FP8>;
+         <Row, Col, DsLayout, ELayout,
+          A0DataType, B0DataType, DsDataType, EDataType, AccDataType, CShuffleDataType,
+          AElementOp,  BElementOp, CDEElementOp,       GemmSpec,   256,
+          128,   128,   128,
+          16,   16,
+          16,   16,
+          4,    4,
+          S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0,
+          S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0,
+          1, 2, S<1, 32, 1, 8>, S<8, 8, 1>,
+          ck::BlockGemmPipelineScheduler::Intrawave, ck::BlockGemmPipelineVersion::v3, FP8>;
 // clang-format on
 
 int main(int argc, char* argv[])
@@ -229,7 +232,7 @@ int main(int argc, char* argv[])
             "not support this GEMM problem");
     }
 
-    float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel, 20, 50});
+    float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel, 0, 20, 50, true, 50});
 
     std::size_t flop = std::size_t(2) * M * N * K;
     std::size_t num_btype =