4bdd116b80
This PR adds a amdgns_load_to_lds intrinsic that abstracts over loads to LDS from global (address space 1) pointers and buffer fat pointers (address space 7), since they use the same API and "gather from a pointer to LDS" is something of an abstract operation. This commit adds the intrinsic and its lowerings for addrspaces 1 and 7, and updates the MLIR wrappers to use it (loosening up the restrictions on loads to LDS along the way to match the ground truth from target features). It also plumbs the intrinsic through to clang.
26 lines
2.5 KiB
Common Lisp
26 lines
2.5 KiB
Common Lisp
// RUN: %clang_cc1 -cl-std=CL2.0 -O0 -triple amdgcn-unknown-unknown -target-cpu gfx900 -S -verify=gfx,expected -o - %s
|
|
// RUN: %clang_cc1 -cl-std=CL2.0 -O0 -triple amdgcn-unknown-unknown -target-cpu gfx942 -S -verify=gfx,expected -o - %s
|
|
// RUN: %clang_cc1 -cl-std=CL2.0 -O0 -triple amdgcn-unknown-unknown -target-cpu gfx1010 -S -verify=gfx,expected -o - %s
|
|
// RUN: %clang_cc1 -cl-std=CL2.0 -O0 -triple amdgcn-unknown-unknown -target-cpu gfx950 -S -verify=gfx950,expected -o - %s
|
|
// REQUIRES: amdgpu-registered-target
|
|
|
|
typedef unsigned int u32;
|
|
|
|
void test_load_to_lds_unsupported_size(global u32* src, local u32 *dst, u32 size, u32 offset, u32 aux) {
|
|
__builtin_amdgcn_load_to_lds(src, dst, size, /*offset=*/0, /*aux=*/0); // expected-error{{argument to '__builtin_amdgcn_load_to_lds' must be a constant integer}}
|
|
__builtin_amdgcn_load_to_lds(src, dst, /*size=*/4, offset, /*aux=*/0); // expected-error{{argument to '__builtin_amdgcn_load_to_lds' must be a constant integer}}
|
|
__builtin_amdgcn_load_to_lds(src, dst, /*size=*/4, /*offset=*/0, aux); // expected-error{{argument to '__builtin_amdgcn_load_to_lds' must be a constant integer}}
|
|
__builtin_amdgcn_load_to_lds(src, dst, /*size=*/5, /*offset=*/0, /*aux=*/0); // expected-error{{invalid size value}} gfx-note {{size must be 1, 2, or 4}} gfx950-note {{size must be 1, 2, 4, 12 or 16}}
|
|
__builtin_amdgcn_load_to_lds(src, dst, /*size=*/0, /*offset=*/0, /*aux=*/0); // expected-error{{invalid size value}} gfx-note {{size must be 1, 2, or 4}} gfx950-note {{size must be 1, 2, 4, 12 or 16}}
|
|
__builtin_amdgcn_load_to_lds(src, dst, /*size=*/3, /*offset=*/0, /*aux=*/0); // expected-error{{invalid size value}} gfx-note {{size must be 1, 2, or 4}} gfx950-note {{size must be 1, 2, 4, 12 or 16}}
|
|
__builtin_amdgcn_load_to_lds(src, dst, /*size=*/12, /*offset=*/0, /*aux=*/0); // gfx-error{{invalid size value}} gfx-note {{size must be 1, 2, or 4}}
|
|
__builtin_amdgcn_load_to_lds(src, dst, /*size=*/16, /*offset=*/0, /*aux=*/0); // gfx-error{{invalid size value}} gfx-note {{size must be 1, 2, or 4}}
|
|
__builtin_amdgcn_load_to_lds(src, dst, /*size=*/-1, /*offset=*/0, /*aux=*/0); // expected-error{{invalid size value}} gfx-note {{size must be 1, 2, or 4}} gfx950-note {{size must be 1, 2, 4, 12 or 16}}
|
|
}
|
|
|
|
__attribute__((target("gfx950-insts")))
|
|
void test_load_to_lds_via_target_feature(global u32* src, local u32 *dst, u32 size, u32 offset, u32 aux) {
|
|
__builtin_amdgcn_load_to_lds(src, dst, /*size=*/12, /*offset=*/0, /*aux=*/0);
|
|
__builtin_amdgcn_load_to_lds(src, dst, /*size=*/16, /*offset=*/0, /*aux=*/0);
|
|
}
|