-
Notifications
You must be signed in to change notification settings - Fork 11.6k
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
AMDGPU: Fix allocating GDS globals to LDS offsets
These don't seem to be very well used or tested, but try to make the behavior a bit more consistent with LDS globals. I'm not sure what the definition for amdgpu-gds-size is supposed to mean. For now I assumed it's allocating a static size at the beginning of the allocation, and any known globals are allocated after it.
- Loading branch information
Showing
6 changed files
with
172 additions
and
23 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,132 @@ | ||
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py | ||
; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s | ||
|
||
@gds0 = internal addrspace(2) global [4 x i32] undef, align 4 | ||
@lds0 = internal addrspace(3) global [4 x i32] undef, align 128 | ||
@lds1 = internal addrspace(3) global [4 x i32] undef, align 256 | ||
|
||
; These two objects should be allocated at the same constant offsets | ||
; from the base. | ||
define amdgpu_kernel void @alloc_lds_gds(i32 addrspace(1)* %out) #1 { | ||
; GCN-LABEL: alloc_lds_gds: | ||
; GCN: ; %bb.0: | ||
; GCN-NEXT: v_mov_b32_e32 v0, 5 | ||
; GCN-NEXT: v_mov_b32_e32 v1, 0 | ||
; GCN-NEXT: s_mov_b32 m0, 16 | ||
; GCN-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) | ||
; GCN-NEXT: ds_add_u32 v1, v0 offset:12 gds | ||
; GCN-NEXT: s_waitcnt lgkmcnt(0) | ||
; GCN-NEXT: buffer_wbinvl1 | ||
; GCN-NEXT: s_waitcnt lgkmcnt(0) | ||
; GCN-NEXT: ds_add_u32 v1, v0 offset:12 | ||
; GCN-NEXT: s_waitcnt lgkmcnt(0) | ||
; GCN-NEXT: s_endpgm | ||
%gep.gds = getelementptr [4 x i32], [4 x i32] addrspace(2)* @gds0, i32 0, i32 3 | ||
%val0 = atomicrmw add i32 addrspace(2)* %gep.gds, i32 5 acq_rel | ||
%gep.lds = getelementptr [4 x i32], [4 x i32] addrspace(3)* @lds0, i32 0, i32 3 | ||
%val1 = atomicrmw add i32 addrspace(3)* %gep.lds, i32 5 acq_rel | ||
ret void | ||
} | ||
|
||
; The LDS alignment shouldn't change offset of GDS. | ||
define amdgpu_kernel void @alloc_lds_gds_align(i32 addrspace(1)* %out) #1 { | ||
; GCN-LABEL: alloc_lds_gds_align: | ||
; GCN: ; %bb.0: | ||
; GCN-NEXT: v_mov_b32_e32 v0, 5 | ||
; GCN-NEXT: v_mov_b32_e32 v1, 0 | ||
; GCN-NEXT: s_mov_b32 m0, 16 | ||
; GCN-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) | ||
; GCN-NEXT: ds_add_u32 v1, v0 offset:12 gds | ||
; GCN-NEXT: s_waitcnt lgkmcnt(0) | ||
; GCN-NEXT: buffer_wbinvl1 | ||
; GCN-NEXT: s_waitcnt lgkmcnt(0) | ||
; GCN-NEXT: ds_add_u32 v1, v0 offset:140 | ||
; GCN-NEXT: s_waitcnt lgkmcnt(0) | ||
; GCN-NEXT: ds_add_u32 v1, v0 offset:12 | ||
; GCN-NEXT: s_waitcnt lgkmcnt(0) | ||
; GCN-NEXT: s_endpgm | ||
%gep.gds = getelementptr [4 x i32], [4 x i32] addrspace(2)* @gds0, i32 0, i32 3 | ||
%val0 = atomicrmw add i32 addrspace(2)* %gep.gds, i32 5 acq_rel | ||
|
||
%gep.lds0 = getelementptr [4 x i32], [4 x i32] addrspace(3)* @lds0, i32 0, i32 3 | ||
%val1 = atomicrmw add i32 addrspace(3)* %gep.lds0, i32 5 acq_rel | ||
|
||
%gep.lds1 = getelementptr [4 x i32], [4 x i32] addrspace(3)* @lds1, i32 0, i32 3 | ||
%val2 = atomicrmw add i32 addrspace(3)* %gep.lds1, i32 5 acq_rel | ||
ret void | ||
} | ||
|
||
@gds_align8 = internal addrspace(2) global [4 x i32] undef, align 8 | ||
@gds_align32 = internal addrspace(2) global [4 x i32] undef, align 32 | ||
|
||
define amdgpu_kernel void @gds_global_align(i32 addrspace(1)* %out) { | ||
; GCN-LABEL: gds_global_align: | ||
; GCN: ; %bb.0: | ||
; GCN-NEXT: v_mov_b32_e32 v0, 5 | ||
; GCN-NEXT: v_mov_b32_e32 v1, 0 | ||
; GCN-NEXT: s_mov_b32 m0, 32 | ||
; GCN-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) | ||
; GCN-NEXT: ds_add_u32 v1, v0 offset:28 gds | ||
; GCN-NEXT: s_waitcnt lgkmcnt(0) | ||
; GCN-NEXT: buffer_wbinvl1 | ||
; GCN-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) | ||
; GCN-NEXT: ds_add_u32 v1, v0 offset:12 gds | ||
; GCN-NEXT: s_waitcnt lgkmcnt(0) | ||
; GCN-NEXT: buffer_wbinvl1 | ||
; GCN-NEXT: s_endpgm | ||
%gep.gds0 = getelementptr [4 x i32], [4 x i32] addrspace(2)* @gds_align8, i32 0, i32 3 | ||
%val0 = atomicrmw add i32 addrspace(2)* %gep.gds0, i32 5 acq_rel | ||
%gep.gds1 = getelementptr [4 x i32], [4 x i32] addrspace(2)* @gds_align32, i32 0, i32 3 | ||
%val1 = atomicrmw add i32 addrspace(2)* %gep.gds1, i32 5 acq_rel | ||
ret void | ||
} | ||
|
||
define amdgpu_kernel void @gds_global_align_plus_attr(i32 addrspace(1)* %out) #0 { | ||
; GCN-LABEL: gds_global_align_plus_attr: | ||
; GCN: ; %bb.0: | ||
; GCN-NEXT: v_mov_b32_e32 v0, 5 | ||
; GCN-NEXT: v_mov_b32_e32 v1, 0 | ||
; GCN-NEXT: s_movk_i32 m0, 0x420 | ||
; GCN-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) | ||
; GCN-NEXT: ds_add_u32 v1, v0 offset:1052 gds | ||
; GCN-NEXT: s_waitcnt lgkmcnt(0) | ||
; GCN-NEXT: buffer_wbinvl1 | ||
; GCN-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) | ||
; GCN-NEXT: ds_add_u32 v1, v0 offset:1036 gds | ||
; GCN-NEXT: s_waitcnt lgkmcnt(0) | ||
; GCN-NEXT: buffer_wbinvl1 | ||
; GCN-NEXT: s_endpgm | ||
%gep.gds0 = getelementptr [4 x i32], [4 x i32] addrspace(2)* @gds_align8, i32 0, i32 3 | ||
%val0 = atomicrmw add i32 addrspace(2)* %gep.gds0, i32 5 acq_rel | ||
%gep.gds1 = getelementptr [4 x i32], [4 x i32] addrspace(2)* @gds_align32, i32 0, i32 3 | ||
%val1 = atomicrmw add i32 addrspace(2)* %gep.gds1, i32 5 acq_rel | ||
ret void | ||
} | ||
|
||
@small.gds = internal addrspace(2) global i8 undef, align 1 | ||
@gds.external = external unnamed_addr addrspace(3) global [0 x i32], align 4 | ||
|
||
define amdgpu_kernel void @gds_extern_align(i32 addrspace(1)* %out, [4 x i32] addrspace(2)* %gds.arg) #0 { | ||
; GCN-LABEL: gds_extern_align: | ||
; GCN: ; %bb.0: | ||
; GCN-NEXT: s_load_dword s0, s[0:1], 0x8 | ||
; GCN-NEXT: v_mov_b32_e32 v0, 5 | ||
; GCN-NEXT: s_movk_i32 m0, 0x401 | ||
; GCN-NEXT: s_movk_i32 s1, 0x400 | ||
; GCN-NEXT: ;;#ASMSTART | ||
; GCN-NEXT: ; use s1 | ||
; GCN-NEXT: ;;#ASMEND | ||
; GCN-NEXT: s_waitcnt lgkmcnt(0) | ||
; GCN-NEXT: v_mov_b32_e32 v1, s0 | ||
; GCN-NEXT: s_waitcnt vmcnt(0) lgkmcnt(0) | ||
; GCN-NEXT: ds_add_u32 v1, v0 offset:12 gds | ||
; GCN-NEXT: s_waitcnt lgkmcnt(0) | ||
; GCN-NEXT: buffer_wbinvl1 | ||
; GCN-NEXT: s_endpgm | ||
call void asm sideeffect "; use $0","s"(i8 addrspace(2)* @small.gds) | ||
%gep.gds0 = getelementptr [4 x i32], [4 x i32] addrspace(2)* %gds.arg, i32 0, i32 3 | ||
%val0 = atomicrmw add i32 addrspace(2)* %gep.gds0, i32 5 acq_rel | ||
ret void | ||
} | ||
|
||
attributes #0 = { "amdgpu-gds-size"="1024" } |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters