fixed vector load siz for fp4

This commit is contained in:
Sami Remes
2026-01-16 12:04:34 -05:00
parent 16ca5cb532
commit f09e10936d
7 changed files with 66 additions and 50 deletions

View File

@@ -673,8 +673,8 @@ struct UniversalGemmKernel
using AiLayout = remove_cvref_t<std::tuple_element_t<i.value, AsLayout>>;
using AiDataType = remove_cvref_t<std::tuple_element_t<i.value, AsDataType>>;
static_assert(GemmPipeline::GetVectorSizeA() == GemmPipeline::GetVectorSizeB(), "Vector size of A and B must be the same!");
static_assert(GemmPipeline::GetVectorSizeA() == 16, "Vector size of A must be 16!");
static_assert(GemmPipeline::GetVectorSizeB() == 16, "Vector size of B must be 16!");
static_assert(GemmPipeline::GetVectorSizeA() == 32, "Vector size of A must be 16!");
static_assert(GemmPipeline::GetVectorSizeB() == 32, "Vector size of B must be 16!");
if constexpr(std::is_same_v<AiLayout, tensor_layout::gemm::RowMajor>)
{
return make_naive_tensor_view<address_space_enum::global>(