leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 26 Packages Projects Releases Wiki Activity

v4.5 dev update. (#3153)

Browse Source
This commit is contained in:
Junkai-Wu
2026-04-07 12:16:05 -04:00
committed by GitHub
parent 418d38a5de
commit a221da7ccf
265 changed files with 4913 additions and 1478 deletions
Download Patch File Download Diff File Expand all files Collapse all files
examples/python/CuTeDSL/ampere/smem_allocator.py
+16 -3
View File
@@ -129,8 +129,8 @@ def kernel(
# ptr<i64, smem, align<128>>
# ptr<f32, smem, align<8>>
print(struct_in_smem.a.data_ptr())
print(struct_in_smem.b)
print(struct_in_smem.c.real)
print(struct_in_smem.b.ptr)
print(struct_in_smem.c.real.ptr)
# ptr<i8, smem, align<512>>
print(section_in_smem)
# ptr<i64, smem, align<64>>
@@ -138,6 +138,17 @@ def kernel(
# tensor<ptr<f16, smem, align<32>> o (16,4):(1,16)>
print(tensor_in_smem)
# assign struct member array element
cute.printf("struct_in_smem.a[0] = {}", struct_in_smem.a[0])
struct_in_smem.a[0] = 2
cute.printf("struct_in_smem.a[0] = {}", struct_in_smem.a[0])
# assign struct member scalar
cute.printf("struct_in_smem.b.ptr = {}", struct_in_smem.b.ptr)
cute.printf("struct_in_smem.b: value = {}", struct_in_smem.b.ptr.load())
struct_in_smem.b = 16
cute.printf("struct_in_smem.b: value = {}", struct_in_smem.b.ptr.load())
# fill MemRange tensor in struct and copy to dst
a_tensor = struct_in_smem.a.get_tensor(cute.make_layout((8, 4)))
a_tensor.fill(const_a)
@@ -169,7 +180,9 @@ def host(
):
# Note: Shared Memory size is automatically calculated now
kernel(const_a, dst_a, const_b, dst_b, const_c, dst_c).launch(
grid=(1, 1, 1), block=(1, 1, 1)
grid=(1, 1, 1),
block=(1, 1, 1),
# Automatically calculate the launch kernel shared memory usage when `smem=None`
)