leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 25 Packages Projects Releases Wiki Activity

v3.9 (#2185)

Browse Source 
* v3.8 update x

* fix blackwell gg

* doc change

* doc change

* doc change

---------

Co-authored-by: yuzhai <yuzhai@nvidia.com>
Co-authored-by: Haicheng Wu <haichengw@nvidia.com>
Co-authored-by: Haicheng Wu <57973641+hwu36@users.noreply.github.com>
This commit is contained in:
Yujia Zhai
2025-03-21 01:52:23 -04:00
committed by GitHub
co-authored by yuzhai Haicheng Wu Haicheng Wu
parent 8c4d1dc47d
commit 62750a2b75
334 changed files with 91517 additions and 2656 deletions
Download Patch File Download Diff File Expand all files Collapse all files
include/cutlass/conv/collective/sm100_implicit_gemm_umma_warpspecialized.hpp
+21 -2
View File
@@ -538,7 +538,7 @@ public:
const auto & input_stride = problem_shape.stride_A;
implementable &= input_stride[ProblemShape::RankT - 1] == 1;
int input_shape_size = 1;
int64_t input_shape_size = 1;
for (int i = ProblemShape::RankT - 2; i >= 0; --i) {
input_shape_size *= input_shape[i + 1];
implementable &= input_stride[i] == input_shape_size;
@@ -548,7 +548,7 @@ public:
const auto & output_stride = problem_shape.stride_C;
implementable &= output_stride[ProblemShape::RankT - 1] == 1;
int output_shape_size = 1;
int64_t output_shape_size = 1;
for (int i = ProblemShape::RankT - 2; i >= 0; --i) {
output_shape_size *= output_shape[i + 1];
implementable &= output_stride[i] == output_shape_size;
@@ -625,6 +625,25 @@ public:
}
}
// The extents of linearized problem shape should be int32_t type(maximum is 2^31-1).
if constexpr (is_im2col_A || is_im2col_B) {
auto [M, N, K, L] = cutlass::conv::detail::get_transformed_problem_shape_MNKL(problem_shape);
auto to_64b = [](auto S) { return transform_leaf(S, [](auto s) { return static_cast<int64_t>(s); }); };
if constexpr (ConvOp == conv::Operator::kFprop || ConvOp == conv::Operator::kDgrad) {
implementable &= (cute::product(to_64b(M)) <= cutlass::platform::numeric_limits<int32_t>::max()) &
(cute::product(to_64b(L)) <= cutlass::platform::numeric_limits<int32_t>::max());
}
else if constexpr (ConvOp == conv::Operator::kWgrad) {
implementable &= (cute::product(to_64b(K)) <= cutlass::platform::numeric_limits<int32_t>::max());
}
if (!implementable) {
CUTLASS_TRACE_HOST(" CAN IMPLEMENT: the extents exceed the maximum number.\n");
return false;
}
}
return true;
}
include/cutlass/conv/collective/sm90_implicit_gemm_gmma_ss_warpspecialized.hpp
+20 -2
View File
@@ -414,7 +414,7 @@ public:
const auto & input_stride = problem_shape.stride_A;
implementable &= input_stride[ProblemShape::RankT - 1] == 1;
int input_shape_size = 1;
int64_t input_shape_size = 1;
for (int i = ProblemShape::RankT - 2; i >= 0; --i) {
input_shape_size *= input_shape[i + 1];
implementable &= input_stride[i] == input_shape_size;
@@ -443,7 +443,7 @@ public:
const auto & output_stride = problem_shape.stride_C;
implementable &= output_stride[ProblemShape::RankT - 1] == 1;
int output_shape_size = 1;
int64_t output_shape_size = 1;
for (int i = ProblemShape::RankT - 2; i >= 0; --i) {
output_shape_size *= output_shape[i + 1];
implementable &= output_stride[i] == output_shape_size;
@@ -482,6 +482,24 @@ public:
return false;
}
if constexpr (is_im2col_A || is_im2col_B) {
auto [M, N, K, L] = cutlass::conv::detail::get_transformed_problem_shape_MNKL(problem_shape);
auto to_64b = [](auto S) { return transform_leaf(S, [](auto s) { return static_cast<int64_t>(s); }); };
if constexpr (ConvOp == conv::Operator::kFprop || ConvOp == conv::Operator::kDgrad) {
implementable &= (cute::product(to_64b(M)) <= cutlass::platform::numeric_limits<int32_t>::max()) &
(cute::product(to_64b(L)) <= cutlass::platform::numeric_limits<int32_t>::max());
}
else if constexpr (ConvOp == conv::Operator::kWgrad) {
implementable &= (cute::product(to_64b(K)) <= cutlass::platform::numeric_limits<int32_t>::max());
}
if (!implementable) {
CUTLASS_TRACE_HOST(" CAN IMPLEMENT: the extents exceed the maximum number.\n");
return false;
}
}
return true;
}