leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 29 Packages Projects Releases Wiki Activity

CUTLASS 3.3.0 (#1167)

Browse Source 
* Release 3.3.0

Adds support for mixed precision GEMMs On Hopper and Ampere
Adds support for < 16B aligned GEMMs on Hopper
Enhancements to EVT
Enhancements to Python interface
Enhancements to Sub-byte type handling in CuTe
Several other bug-fixes and performance improvements.

* minor doc update
This commit is contained in:
Pradeep Ramani
2023-11-02 11:09:05 -04:00
committed by GitHub
parent 922fb5108b
commit c008b4aea8
263 changed files with 16214 additions and 5008 deletions
Download Patch File Download Diff File Expand all files Collapse all files
test/unit/gemm/warp/gemm_mixed_input_sm80.cu
+11 -5
View File
@@ -55,6 +55,7 @@
////////////////////////////////////////////////////////////////////////////////
/// F32 <= F16 * I8 + F32 (Upcast on Operand B)
////////////////////////////////////////////////////////////////////////////////
TEST(SM80_warp_gemm_mixed_input_tensor_op_crosswise_f16_i8, 128x128x64_64x64x64_16x8x16) {
using Shape = cutlass::gemm::GemmShape<64, 64, 64>;
using InstructionShape = cutlass::gemm::GemmShape<16, 8, 16>;
@@ -98,6 +99,7 @@ TEST(SM80_warp_gemm_mixed_input_tensor_op_crosswise_f16_i8, 64x64x64_64x64x64_16
////////////////////////////////////////////////////////////////////////////////
/// F32 <= I8 * F16 + F32 (Upcast on Operand A)
////////////////////////////////////////////////////////////////////////////////
TEST(SM80_warp_gemm_mixed_input_tensor_op_crosswise_i8_f16, 128x128x64_64x64x64_16x8x16) {
using Shape = cutlass::gemm::GemmShape<64, 64, 64>;
using InstructionShape = cutlass::gemm::GemmShape<16, 8, 16>;
@@ -118,7 +120,6 @@ TEST(SM80_warp_gemm_mixed_input_tensor_op_crosswise_i8_f16, 128x128x64_64x64x64_
.run();
}
TEST(SM80_warp_gemm_mixed_input_tensor_op_crosswise_i8_f16, 64x64x64_64x64x64_16x8x16) {
using Shape = cutlass::gemm::GemmShape<64, 64, 64>;
using InstructionShape = cutlass::gemm::GemmShape<16, 8, 16>;
@@ -142,6 +143,7 @@ TEST(SM80_warp_gemm_mixed_input_tensor_op_crosswise_i8_f16, 64x64x64_64x64x64_16
////////////////////////////////////////////////////////////////////////////////
/// F32 <= F16 * U8 + F32 (Upcast on Operand B)
////////////////////////////////////////////////////////////////////////////////
TEST(SM80_warp_gemm_mixed_input_tensor_op_crosswise_f16_u8, 64x64x64_64x64x64_16x8x16) {
using Shape = cutlass::gemm::GemmShape<64, 64, 64>;
using InstructionShape = cutlass::gemm::GemmShape<16, 8, 16>;
@@ -185,6 +187,7 @@ TEST(SM80_warp_gemm_mixed_input_tensor_op_crosswise_f16_u8, 128x128x64_64x64x64_
////////////////////////////////////////////////////////////////////////////////
/// F32 <= U8 * F16 + F32 (Upcast on Operand A)
////////////////////////////////////////////////////////////////////////////////
TEST(SM80_warp_gemm_mixed_input_tensor_op_crosswise_u8_f16, 64x64x64_64x64x64_16x8x16) {
using Shape = cutlass::gemm::GemmShape<64, 64, 64>;
using InstructionShape = cutlass::gemm::GemmShape<16, 8, 16>;
@@ -225,10 +228,10 @@ TEST(SM80_warp_gemm_mixed_input_tensor_op_crosswise_u8_f16, 128x128x64_64x64x64_
.run();
}
////////////////////////////////////////////////////////////////////////////////
/// F32 <= B16 * U8 + F32 (Upcast on Operand B)
////////////////////////////////////////////////////////////////////////////////
TEST(SM80_warp_gemm_mixed_input_tensor_op_crosswise_bf16_u8, 64x64x64_64x64x64_16x8x16) {
using Shape = cutlass::gemm::GemmShape<64, 64, 64>;
using InstructionShape = cutlass::gemm::GemmShape<16, 8, 16>;
@@ -252,6 +255,7 @@ TEST(SM80_warp_gemm_mixed_input_tensor_op_crosswise_bf16_u8, 64x64x64_64x64x64_1
////////////////////////////////////////////////////////////////////////////////
/// F32 <= U8 * BF16 + F32 (Upcast on Operand A)
////////////////////////////////////////////////////////////////////////////////
TEST(SM80_warp_gemm_mixed_input_tensor_op_crosswise_u8_bf16, 64x64x64_64x64x64_16x8x16) {
using Shape = cutlass::gemm::GemmShape<64, 64, 64>;
using InstructionShape = cutlass::gemm::GemmShape<16, 8, 16>;
@@ -273,8 +277,9 @@ TEST(SM80_warp_gemm_mixed_input_tensor_op_crosswise_u8_bf16, 64x64x64_64x64x64_1
}
////////////////////////////////////////////////////////////////////////////////
/// F32 <= B16 * I8 + F32 (Upcast on Operand B)
/// F32 <= I8 * BF16 + F32 (Upcast on Operand A)
////////////////////////////////////////////////////////////////////////////////
TEST(SM80_warp_gemm_mixed_input_tensor_op_crosswise_bf16_i8, 64x64x64_64x64x64_16x8x16) {
using Shape = cutlass::gemm::GemmShape<64, 64, 64>;
using InstructionShape = cutlass::gemm::GemmShape<16, 8, 16>;
@@ -296,8 +301,9 @@ TEST(SM80_warp_gemm_mixed_input_tensor_op_crosswise_bf16_i8, 64x64x64_64x64x64_1
}
////////////////////////////////////////////////////////////////////////////////
/// F32 <= I8 * BF16 + F32 (Upcast on Operand A)
/// F32 <= B16 * I8 + F32 (Upcast on Operand B)
////////////////////////////////////////////////////////////////////////////////
TEST(SM80_warp_gemm_mixed_input_tensor_op_crosswise_i8_bf16, 64x64x64_64x64x64_16x8x16) {
using Shape = cutlass::gemm::GemmShape<64, 64, 64>;
using InstructionShape = cutlass::gemm::GemmShape<16, 8, 16>;
@@ -318,4 +324,4 @@ TEST(SM80_warp_gemm_mixed_input_tensor_op_crosswise_i8_bf16, 64x64x64_64x64x64_1
.run();
}
#endif // if defined(CUTLASS_ARCH_MMA_SM80_SUPPORTED)
#endif // if defined(CUTLASS_ARCH_MMA_SM80_SUPPORTED)