blob: 94f01ee143ef182b8d321ffada093e1422989094 [file] [edit]
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 6
# RUN: llc -mtriple=amdgpu11.00 -passes=si-load-store-opt -o - %s | FileCheck -check-prefix=GFX11 %s
# The immediate offset is only part of the address of a TBUFFER access, so
# merging must respect the known alignment of the whole address.
--- |
define amdgpu_vs void @load_x_x_16bit_align4() { ret void }
define amdgpu_vs void @load_x_x_x_16bit_align4() { ret void }
define amdgpu_vs void @load_x_x_8bit_align1() { ret void }
define amdgpu_vs void @load_x_x_8bit_align2() { ret void }
define amdgpu_vs void @load_x_x_32bit_align1() { ret void }
define amdgpu_vs void @load_x_x_16bit_leading_align4_second() { ret void }
define amdgpu_vs void @load_x_x_16bit_leading_align2_second() { ret void }
define amdgpu_vs void @load_x_x_16bit_basealign4_offset2(ptr addrspace(8) %rsrc) { ret void }
define amdgpu_vs void @store_x_x_16bit_align2() { ret void }
!llvm.module.flags = !{!0}
!0 = !{i32 7, !"amdgpu.tbuffer.oob.mode", i32 1}
...
---
name: load_x_x_16bit_align4
body: |
bb.0:
liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0
; GFX11-LABEL: name: load_x_x_16bit_align4
; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: %voff:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: %rsrc:sgpr_128 = REG_SEQUENCE $sgpr0, %subreg.sub0, $sgpr1, %subreg.sub1, $sgpr2, %subreg.sub2, $sgpr3, %subreg.sub3
; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFEN %voff, %rsrc, 0, 0, 29, 0, 0, implicit $exec :: (dereferenceable load (s32), addrspace 8)
; GFX11-NEXT: %x0:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFEN]].sub0
; GFX11-NEXT: %x1:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFEN]].sub1
%voff:vgpr_32 = COPY $vgpr0
%rsrc:sgpr_128 = REG_SEQUENCE $sgpr0, %subreg.sub0, $sgpr1, %subreg.sub1, $sgpr2, %subreg.sub2, $sgpr3, %subreg.sub3
%x0:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN %voff, %rsrc, 0, 0, 13, 0, 0, implicit $exec :: (dereferenceable load (s16), align 4, addrspace 8)
%x1:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN %voff, %rsrc, 0, 2, 13, 0, 0, implicit $exec :: (dereferenceable load (s16), align 2, addrspace 8)
...
---
name: load_x_x_x_16bit_align4
body: |
bb.0:
liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0
; GFX11-LABEL: name: load_x_x_x_16bit_align4
; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: %voff:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: %rsrc:sgpr_128 = REG_SEQUENCE $sgpr0, %subreg.sub0, $sgpr1, %subreg.sub1, $sgpr2, %subreg.sub2, $sgpr3, %subreg.sub3
; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XYZ_OFFEN:%[0-9]+]]:vreg_96 = TBUFFER_LOAD_FORMAT_XYZ_OFFEN %voff, %rsrc, 0, 0, 57, 0, 0, implicit $exec :: (dereferenceable load (s48), align 4, addrspace 8)
; GFX11-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_OFFEN]].sub0_sub1
; GFX11-NEXT: %x2:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XYZ_OFFEN]].sub2
; GFX11-NEXT: %x0:vgpr_32 = COPY [[COPY]].sub0
; GFX11-NEXT: %x1:vgpr_32 = COPY [[COPY]].sub1
%voff:vgpr_32 = COPY $vgpr0
%rsrc:sgpr_128 = REG_SEQUENCE $sgpr0, %subreg.sub0, $sgpr1, %subreg.sub1, $sgpr2, %subreg.sub2, $sgpr3, %subreg.sub3
%x0:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN %voff, %rsrc, 0, 0, 13, 0, 0, implicit $exec :: (dereferenceable load (s16), align 4, addrspace 8)
%x1:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN %voff, %rsrc, 0, 2, 13, 0, 0, implicit $exec :: (dereferenceable load (s16), align 2, addrspace 8)
%x2:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN %voff, %rsrc, 0, 4, 13, 0, 0, implicit $exec :: (dereferenceable load (s16), align 4, addrspace 8)
...
---
name: load_x_x_8bit_align1
body: |
bb.0:
liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0
; GFX11-LABEL: name: load_x_x_8bit_align1
; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: %voff:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: %rsrc:sgpr_128 = REG_SEQUENCE $sgpr0, %subreg.sub0, $sgpr1, %subreg.sub1, $sgpr2, %subreg.sub2, $sgpr3, %subreg.sub3
; GFX11-NEXT: %x0:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN %voff, %rsrc, 0, 0, 5, 0, 0, implicit $exec :: (dereferenceable load (s8), addrspace 8)
; GFX11-NEXT: %x1:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN %voff, %rsrc, 0, 1, 5, 0, 0, implicit $exec :: (dereferenceable load (s8), addrspace 8)
%voff:vgpr_32 = COPY $vgpr0
%rsrc:sgpr_128 = REG_SEQUENCE $sgpr0, %subreg.sub0, $sgpr1, %subreg.sub1, $sgpr2, %subreg.sub2, $sgpr3, %subreg.sub3
%x0:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN %voff, %rsrc, 0, 0, 5, 0, 0, implicit $exec :: (dereferenceable load (s8), align 1, addrspace 8)
%x1:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN %voff, %rsrc, 0, 1, 5, 0, 0, implicit $exec :: (dereferenceable load (s8), align 1, addrspace 8)
...
---
name: load_x_x_8bit_align2
body: |
bb.0:
liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0
; GFX11-LABEL: name: load_x_x_8bit_align2
; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: %voff:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: %rsrc:sgpr_128 = REG_SEQUENCE $sgpr0, %subreg.sub0, $sgpr1, %subreg.sub1, $sgpr2, %subreg.sub2, $sgpr3, %subreg.sub3
; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFEN %voff, %rsrc, 0, 0, 18, 0, 0, implicit $exec :: (dereferenceable load (s16), addrspace 8)
; GFX11-NEXT: %x0:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFEN]].sub0
; GFX11-NEXT: %x1:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFEN]].sub1
%voff:vgpr_32 = COPY $vgpr0
%rsrc:sgpr_128 = REG_SEQUENCE $sgpr0, %subreg.sub0, $sgpr1, %subreg.sub1, $sgpr2, %subreg.sub2, $sgpr3, %subreg.sub3
%x0:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN %voff, %rsrc, 0, 0, 5, 0, 0, implicit $exec :: (dereferenceable load (s8), align 2, addrspace 8)
%x1:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN %voff, %rsrc, 0, 1, 5, 0, 0, implicit $exec :: (dereferenceable load (s8), align 1, addrspace 8)
...
---
# 32-bit components are assumed to be 4-byte aligned.
name: load_x_x_32bit_align1
body: |
bb.0:
liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0
; GFX11-LABEL: name: load_x_x_32bit_align1
; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: %voff:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: %rsrc:sgpr_128 = REG_SEQUENCE $sgpr0, %subreg.sub0, $sgpr1, %subreg.sub1, $sgpr2, %subreg.sub2, $sgpr3, %subreg.sub3
; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFEN %voff, %rsrc, 0, 0, 50, 0, 0, implicit $exec :: (dereferenceable load (s64), align 1, addrspace 8)
; GFX11-NEXT: %x0:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFEN]].sub0
; GFX11-NEXT: %x1:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFEN]].sub1
%voff:vgpr_32 = COPY $vgpr0
%rsrc:sgpr_128 = REG_SEQUENCE $sgpr0, %subreg.sub0, $sgpr1, %subreg.sub1, $sgpr2, %subreg.sub2, $sgpr3, %subreg.sub3
%x0:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN %voff, %rsrc, 0, 0, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 8)
%x1:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN %voff, %rsrc, 0, 4, 22, 0, 0, implicit $exec :: (dereferenceable load (s32), align 1, addrspace 8)
...
---
# The access with the lower offset comes second, and is aligned enough.
name: load_x_x_16bit_leading_align4_second
body: |
bb.0:
liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0
; GFX11-LABEL: name: load_x_x_16bit_leading_align4_second
; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: %voff:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: %rsrc:sgpr_128 = REG_SEQUENCE $sgpr0, %subreg.sub0, $sgpr1, %subreg.sub1, $sgpr2, %subreg.sub2, $sgpr3, %subreg.sub3
; GFX11-NEXT: [[TBUFFER_LOAD_FORMAT_XY_OFFEN:%[0-9]+]]:vreg_64 = TBUFFER_LOAD_FORMAT_XY_OFFEN %voff, %rsrc, 0, 0, 29, 0, 0, implicit $exec :: (dereferenceable load (s32), addrspace 8)
; GFX11-NEXT: %x1:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFEN]].sub1
; GFX11-NEXT: %x0:vgpr_32 = COPY [[TBUFFER_LOAD_FORMAT_XY_OFFEN]].sub0
%voff:vgpr_32 = COPY $vgpr0
%rsrc:sgpr_128 = REG_SEQUENCE $sgpr0, %subreg.sub0, $sgpr1, %subreg.sub1, $sgpr2, %subreg.sub2, $sgpr3, %subreg.sub3
%x1:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN %voff, %rsrc, 0, 2, 13, 0, 0, implicit $exec :: (dereferenceable load (s16), align 2, addrspace 8)
%x0:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN %voff, %rsrc, 0, 0, 13, 0, 0, implicit $exec :: (dereferenceable load (s16), align 4, addrspace 8)
...
---
# The access with the lower offset comes second, and isn't aligned enough.
name: load_x_x_16bit_leading_align2_second
body: |
bb.0:
liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0
; GFX11-LABEL: name: load_x_x_16bit_leading_align2_second
; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: %voff:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: %rsrc:sgpr_128 = REG_SEQUENCE $sgpr0, %subreg.sub0, $sgpr1, %subreg.sub1, $sgpr2, %subreg.sub2, $sgpr3, %subreg.sub3
; GFX11-NEXT: %x1:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN %voff, %rsrc, 0, 2, 13, 0, 0, implicit $exec :: (dereferenceable load (s16), align 4, addrspace 8)
; GFX11-NEXT: %x0:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN %voff, %rsrc, 0, 0, 13, 0, 0, implicit $exec :: (dereferenceable load (s16), addrspace 8)
%voff:vgpr_32 = COPY $vgpr0
%rsrc:sgpr_128 = REG_SEQUENCE $sgpr0, %subreg.sub0, $sgpr1, %subreg.sub1, $sgpr2, %subreg.sub2, $sgpr3, %subreg.sub3
%x1:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN %voff, %rsrc, 0, 2, 13, 0, 0, implicit $exec :: (dereferenceable load (s16), align 4, addrspace 8)
%x0:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN %voff, %rsrc, 0, 0, 13, 0, 0, implicit $exec :: (dereferenceable load (s16), align 2, addrspace 8)
...
---
# The leading access is only 2-byte aligned, even though its base is 4-byte
# aligned.
name: load_x_x_16bit_basealign4_offset2
body: |
bb.0:
liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0
; GFX11-LABEL: name: load_x_x_16bit_basealign4_offset2
; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: %voff:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: %rsrc:sgpr_128 = REG_SEQUENCE $sgpr0, %subreg.sub0, $sgpr1, %subreg.sub1, $sgpr2, %subreg.sub2, $sgpr3, %subreg.sub3
; GFX11-NEXT: %x0:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN %voff, %rsrc, 0, 0, 13, 0, 0, implicit $exec :: (dereferenceable load (s16) from %ir.rsrc + 2, basealign 4, addrspace 8)
; GFX11-NEXT: %x1:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN %voff, %rsrc, 0, 2, 13, 0, 0, implicit $exec :: (dereferenceable load (s16) from %ir.rsrc + 4, align 4, addrspace 8)
%voff:vgpr_32 = COPY $vgpr0
%rsrc:sgpr_128 = REG_SEQUENCE $sgpr0, %subreg.sub0, $sgpr1, %subreg.sub1, $sgpr2, %subreg.sub2, $sgpr3, %subreg.sub3
%x0:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN %voff, %rsrc, 0, 0, 13, 0, 0, implicit $exec :: (dereferenceable load (s16) from %ir.rsrc + 2, basealign 4, addrspace 8)
%x1:vgpr_32 = TBUFFER_LOAD_FORMAT_X_OFFEN %voff, %rsrc, 0, 2, 13, 0, 0, implicit $exec :: (dereferenceable load (s16) from %ir.rsrc + 4, align 4, addrspace 8)
...
---
name: store_x_x_16bit_align2
body: |
bb.0:
liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2
; GFX11-LABEL: name: store_x_x_16bit_align2
; GFX11: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2
; GFX11-NEXT: {{ $}}
; GFX11-NEXT: %voff:vgpr_32 = COPY $vgpr0
; GFX11-NEXT: %d0:vgpr_32 = COPY $vgpr1
; GFX11-NEXT: %d1:vgpr_32 = COPY $vgpr2
; GFX11-NEXT: %rsrc:sgpr_128 = REG_SEQUENCE $sgpr0, %subreg.sub0, $sgpr1, %subreg.sub1, $sgpr2, %subreg.sub2, $sgpr3, %subreg.sub3
; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFEN_exact %d0, %voff, %rsrc, 0, 0, 13, 0, 0, implicit $exec :: (store (s16), addrspace 8)
; GFX11-NEXT: TBUFFER_STORE_FORMAT_X_OFFEN_exact %d1, %voff, %rsrc, 0, 2, 13, 0, 0, implicit $exec :: (store (s16), addrspace 8)
%voff:vgpr_32 = COPY $vgpr0
%d0:vgpr_32 = COPY $vgpr1
%d1:vgpr_32 = COPY $vgpr2
%rsrc:sgpr_128 = REG_SEQUENCE $sgpr0, %subreg.sub0, $sgpr1, %subreg.sub1, $sgpr2, %subreg.sub2, $sgpr3, %subreg.sub3
TBUFFER_STORE_FORMAT_X_OFFEN_exact %d0, %voff, %rsrc, 0, 0, 13, 0, 0, implicit $exec :: (store (s16), align 2, addrspace 8)
TBUFFER_STORE_FORMAT_X_OFFEN_exact %d1, %voff, %rsrc, 0, 2, 13, 0, 0, implicit $exec :: (store (s16), align 2, addrspace 8)
...