derek.holloway
7502bf64a4
CodeQL / Analyze (actions) (push) Canceled after 0s
CodeQL / Analyze (python) (push) Canceled after 0s
pre-commit / pre-commit (push) Canceled after 0s
Multi-Arch CI / setup (push) Canceled after 0s
Publish build_manylinux_x86_64 images / publish_build_manylinux_x86_64 (push) Canceled after 0s
Setup (Multi-Arch) / setup (push) Canceled after 0s
Multi-Arch CI / Linux::${{ fromJSON(needs.setup.outputs.linux_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubi10 image / publish_no_rocm_image_ubi10 (push) Canceled after 0s
Multi-Arch CI / Windows::${{ fromJSON(needs.setup.outputs.windows_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Multi-Arch CI / Manifest Diff (push) Canceled after 0s
TheRock Manifest Diff Report / Generate Manifest Diff Report (push) Canceled after 0s
Multi-Arch CI / CI Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_ocl_rt images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_rocgdb images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Unit Tests / Unit Tests :: ubuntu-24.04 (push) Canceled after 0s
Unit Tests / Unit Tests :: windows-2022 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_media (push) Canceled after 0s
Unit Tests / Unit Tests Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_openmpi (push) Canceled after 0s
401 lines
28 KiB
LLVM
401 lines
28 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --filter-out-after "debug-location" --version 6
|
|
; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-amd-amdpal -mcpu=gfx950 -stop-after=instruction-select < %s | FileCheck -check-prefix=GFX-950 %s
|
|
|
|
;; BRC trunc tests - exercises UniBRC/DivBRC wildcard rules.
|
|
;; Each pair covers a unique subregister extraction sequence.
|
|
|
|
; --- sub0: 32-bit from 64-bit ---
|
|
|
|
define amdgpu_ps void @s_trunc_i64_to_i32(ptr addrspace(1) inreg %src, ptr addrspace(1) inreg %dst) {
|
|
; GFX-950-LABEL: name: s_trunc_i64_to_i32
|
|
; GFX-950: bb.1 (%ir-block.0):
|
|
; GFX-950-NEXT: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3
|
|
; GFX-950-NEXT: {{ $}}
|
|
; GFX-950-NEXT: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr0
|
|
; GFX-950-NEXT: [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr1
|
|
; GFX-950-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
|
|
; GFX-950-NEXT: [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr2
|
|
; GFX-950-NEXT: [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr3
|
|
; GFX-950-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64_xexec_xnull = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY3]], %subreg.sub1
|
|
; GFX-950-NEXT: [[S_LOAD_DWORDX2_IMM:%[0-9]+]]:sreg_64_xexec = S_LOAD_DWORDX2_IMM [[REG_SEQUENCE]], 0, 0 :: ("amdgpu-noclobber" load (s64) from %ir.src, addrspace 1)
|
|
; GFX-950-NEXT: [[COPY4:%[0-9]+]]:sreg_32 = COPY [[S_LOAD_DWORDX2_IMM]].sub0, debug-location !4
|
|
%val = load i64, ptr addrspace(1) %src
|
|
%trunc = trunc i64 %val to i32, !dbg !4
|
|
store i32 %trunc, ptr addrspace(1) %dst
|
|
ret void
|
|
}
|
|
|
|
define void @v_trunc_i64_to_i32(ptr addrspace(1) %src, ptr addrspace(1) %dst) {
|
|
; GFX-950-LABEL: name: v_trunc_i64_to_i32
|
|
; GFX-950: bb.1 (%ir-block.0):
|
|
; GFX-950-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
|
|
; GFX-950-NEXT: {{ $}}
|
|
; GFX-950-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
|
|
; GFX-950-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
|
|
; GFX-950-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
|
|
; GFX-950-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
|
|
; GFX-950-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
|
|
; GFX-950-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY3]], %subreg.sub1
|
|
; GFX-950-NEXT: [[GLOBAL_LOAD_DWORDX2_:%[0-9]+]]:vreg_64_align2 = GLOBAL_LOAD_DWORDX2 [[REG_SEQUENCE]], 0, 0, implicit $exec :: (load (s64) from %ir.src, addrspace 1)
|
|
; GFX-950-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[GLOBAL_LOAD_DWORDX2_]].sub0, debug-location !4
|
|
%val = load i64, ptr addrspace(1) %src
|
|
%trunc = trunc i64 %val to i32, !dbg !4
|
|
store i32 %trunc, ptr addrspace(1) %dst
|
|
ret void
|
|
}
|
|
|
|
; --- sub0_sub1: 64-bit from 96-bit ---
|
|
|
|
define amdgpu_ps void @s_trunc_i96_to_i64(ptr addrspace(1) inreg %src, ptr addrspace(1) inreg %dst) {
|
|
; GFX-950-LABEL: name: s_trunc_i96_to_i64
|
|
; GFX-950: bb.1 (%ir-block.0):
|
|
; GFX-950-NEXT: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3
|
|
; GFX-950-NEXT: {{ $}}
|
|
; GFX-950-NEXT: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr0
|
|
; GFX-950-NEXT: [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr1
|
|
; GFX-950-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
|
|
; GFX-950-NEXT: [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr2
|
|
; GFX-950-NEXT: [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr3
|
|
; GFX-950-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64_xexec_xnull = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY3]], %subreg.sub1
|
|
; GFX-950-NEXT: [[S_LOAD_DWORDX2_IMM:%[0-9]+]]:sreg_64_xexec = S_LOAD_DWORDX2_IMM [[REG_SEQUENCE]], 0, 0 :: ("amdgpu-noclobber" load (<2 x s32>) from %ir.src, addrspace 1)
|
|
; GFX-950-NEXT: [[S_LOAD_DWORD_IMM:%[0-9]+]]:sreg_32_xm0_xexec = S_LOAD_DWORD_IMM [[REG_SEQUENCE]], 8, 0 :: ("amdgpu-noclobber" load (s32) from %ir.src + 8, align 8, addrspace 1)
|
|
; GFX-950-NEXT: [[COPY4:%[0-9]+]]:sreg_32 = COPY [[S_LOAD_DWORDX2_IMM]].sub0
|
|
; GFX-950-NEXT: [[COPY5:%[0-9]+]]:sreg_32 = COPY [[S_LOAD_DWORDX2_IMM]].sub1
|
|
; GFX-950-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_96 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[S_LOAD_DWORD_IMM]], %subreg.sub2
|
|
; GFX-950-NEXT: [[COPY6:%[0-9]+]]:sreg_64 = COPY [[REG_SEQUENCE2]].sub0_sub1, debug-location !4
|
|
%val = load i96, ptr addrspace(1) %src
|
|
%trunc = trunc i96 %val to i64, !dbg !4
|
|
store i64 %trunc, ptr addrspace(1) %dst
|
|
ret void
|
|
}
|
|
|
|
define void @v_trunc_i96_to_i64(ptr addrspace(1) %src, ptr addrspace(1) %dst) {
|
|
; GFX-950-LABEL: name: v_trunc_i96_to_i64
|
|
; GFX-950: bb.1 (%ir-block.0):
|
|
; GFX-950-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
|
|
; GFX-950-NEXT: {{ $}}
|
|
; GFX-950-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
|
|
; GFX-950-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
|
|
; GFX-950-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
|
|
; GFX-950-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
|
|
; GFX-950-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
|
|
; GFX-950-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY3]], %subreg.sub1
|
|
; GFX-950-NEXT: [[GLOBAL_LOAD_DWORDX3_:%[0-9]+]]:vreg_96_align2 = GLOBAL_LOAD_DWORDX3 [[REG_SEQUENCE]], 0, 0, implicit $exec :: (load (<3 x s32>) from %ir.src, align 8, addrspace 1)
|
|
; GFX-950-NEXT: [[COPY4:%[0-9]+]]:vreg_64_align2 = COPY [[GLOBAL_LOAD_DWORDX3_]].sub0_sub1, debug-location !4
|
|
%val = load i96, ptr addrspace(1) %src
|
|
%trunc = trunc i96 %val to i64, !dbg !4
|
|
store i64 %trunc, ptr addrspace(1) %dst
|
|
ret void
|
|
}
|
|
|
|
; --- sub0_sub1_sub2: 96-bit from 128-bit ---
|
|
|
|
define amdgpu_ps void @s_trunc_i128_to_i96(ptr addrspace(1) inreg %src, ptr addrspace(1) inreg %dst) {
|
|
; GFX-950-LABEL: name: s_trunc_i128_to_i96
|
|
; GFX-950: bb.1 (%ir-block.0):
|
|
; GFX-950-NEXT: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3
|
|
; GFX-950-NEXT: {{ $}}
|
|
; GFX-950-NEXT: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr0
|
|
; GFX-950-NEXT: [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr1
|
|
; GFX-950-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
|
|
; GFX-950-NEXT: [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr2
|
|
; GFX-950-NEXT: [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr3
|
|
; GFX-950-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64_xexec_xnull = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY3]], %subreg.sub1
|
|
; GFX-950-NEXT: early-clobber %9:sgpr_128 = S_LOAD_DWORDX4_IMM_ec [[REG_SEQUENCE]], 0, 0 :: ("amdgpu-noclobber" load (<4 x s32>) from %ir.src, align 8, addrspace 1)
|
|
; GFX-950-NEXT: [[COPY4:%[0-9]+]]:sgpr_96 = COPY %9.sub0_sub1_sub2, debug-location !4
|
|
%val = load i128, ptr addrspace(1) %src
|
|
%trunc = trunc i128 %val to i96, !dbg !4
|
|
store i96 %trunc, ptr addrspace(1) %dst
|
|
ret void
|
|
}
|
|
|
|
define void @v_trunc_i128_to_i96(ptr addrspace(1) %src, ptr addrspace(1) %dst) {
|
|
; GFX-950-LABEL: name: v_trunc_i128_to_i96
|
|
; GFX-950: bb.1 (%ir-block.0):
|
|
; GFX-950-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
|
|
; GFX-950-NEXT: {{ $}}
|
|
; GFX-950-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
|
|
; GFX-950-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
|
|
; GFX-950-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
|
|
; GFX-950-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
|
|
; GFX-950-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
|
|
; GFX-950-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY3]], %subreg.sub1
|
|
; GFX-950-NEXT: [[GLOBAL_LOAD_DWORDX4_:%[0-9]+]]:vreg_128_align2 = GLOBAL_LOAD_DWORDX4 [[REG_SEQUENCE]], 0, 0, implicit $exec :: (load (<4 x s32>) from %ir.src, align 8, addrspace 1)
|
|
; GFX-950-NEXT: [[COPY4:%[0-9]+]]:vreg_96_align2 = COPY [[GLOBAL_LOAD_DWORDX4_]].sub0_sub1_sub2, debug-location !4
|
|
%val = load i128, ptr addrspace(1) %src
|
|
%trunc = trunc i128 %val to i96, !dbg !4
|
|
store i96 %trunc, ptr addrspace(1) %dst
|
|
ret void
|
|
}
|
|
|
|
; --- sub0_sub1_sub2_sub3: 128-bit from 160-bit ---
|
|
|
|
define amdgpu_ps void @s_trunc_i160_to_i128(ptr addrspace(1) inreg %src, ptr addrspace(1) inreg %dst) {
|
|
; GFX-950-LABEL: name: s_trunc_i160_to_i128
|
|
; GFX-950: bb.1 (%ir-block.0):
|
|
; GFX-950-NEXT: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3
|
|
; GFX-950-NEXT: {{ $}}
|
|
; GFX-950-NEXT: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr0
|
|
; GFX-950-NEXT: [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr1
|
|
; GFX-950-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
|
|
; GFX-950-NEXT: [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr2
|
|
; GFX-950-NEXT: [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr3
|
|
; GFX-950-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64_xexec_xnull = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY3]], %subreg.sub1
|
|
; GFX-950-NEXT: early-clobber %10:sgpr_128 = S_LOAD_DWORDX4_IMM_ec [[REG_SEQUENCE]], 0, 0 :: ("amdgpu-noclobber" load (<4 x s32>) from %ir.src, align 8, addrspace 1)
|
|
; GFX-950-NEXT: [[S_LOAD_DWORD_IMM:%[0-9]+]]:sreg_32_xm0_xexec = S_LOAD_DWORD_IMM [[REG_SEQUENCE]], 16, 0 :: ("amdgpu-noclobber" load (s32) from %ir.src + 16, align 8, addrspace 1)
|
|
; GFX-950-NEXT: [[COPY4:%[0-9]+]]:sreg_32 = COPY %10.sub0
|
|
; GFX-950-NEXT: [[COPY5:%[0-9]+]]:sreg_32 = COPY %10.sub1
|
|
; GFX-950-NEXT: [[COPY6:%[0-9]+]]:sreg_32 = COPY %10.sub2
|
|
; GFX-950-NEXT: [[COPY7:%[0-9]+]]:sreg_32 = COPY %10.sub3
|
|
; GFX-950-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_160 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY6]], %subreg.sub2, [[COPY7]], %subreg.sub3, [[S_LOAD_DWORD_IMM]], %subreg.sub4
|
|
; GFX-950-NEXT: [[COPY8:%[0-9]+]]:sgpr_128 = COPY [[REG_SEQUENCE2]].sub0_sub1_sub2_sub3, debug-location !4
|
|
%val = load i160, ptr addrspace(1) %src
|
|
%trunc = trunc i160 %val to i128, !dbg !4
|
|
store i128 %trunc, ptr addrspace(1) %dst
|
|
ret void
|
|
}
|
|
|
|
define void @v_trunc_i160_to_i128(ptr addrspace(1) %src, ptr addrspace(1) %dst) {
|
|
; GFX-950-LABEL: name: v_trunc_i160_to_i128
|
|
; GFX-950: bb.1 (%ir-block.0):
|
|
; GFX-950-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
|
|
; GFX-950-NEXT: {{ $}}
|
|
; GFX-950-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
|
|
; GFX-950-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
|
|
; GFX-950-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
|
|
; GFX-950-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
|
|
; GFX-950-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
|
|
; GFX-950-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY3]], %subreg.sub1
|
|
; GFX-950-NEXT: [[GLOBAL_LOAD_DWORDX4_:%[0-9]+]]:vreg_128_align2 = GLOBAL_LOAD_DWORDX4 [[REG_SEQUENCE]], 0, 0, implicit $exec :: (load (<4 x s32>) from %ir.src, align 8, addrspace 1)
|
|
; GFX-950-NEXT: [[GLOBAL_LOAD_DWORD:%[0-9]+]]:vgpr_32 = GLOBAL_LOAD_DWORD [[REG_SEQUENCE]], 16, 0, implicit $exec :: (load (s32) from %ir.src + 16, align 8, addrspace 1)
|
|
; GFX-950-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[GLOBAL_LOAD_DWORDX4_]].sub0
|
|
; GFX-950-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[GLOBAL_LOAD_DWORDX4_]].sub1
|
|
; GFX-950-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[GLOBAL_LOAD_DWORDX4_]].sub2
|
|
; GFX-950-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[GLOBAL_LOAD_DWORDX4_]].sub3
|
|
; GFX-950-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_160_align2 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY6]], %subreg.sub2, [[COPY7]], %subreg.sub3, [[GLOBAL_LOAD_DWORD]], %subreg.sub4
|
|
; GFX-950-NEXT: [[COPY8:%[0-9]+]]:vreg_128_align2 = COPY [[REG_SEQUENCE2]].sub0_sub1_sub2_sub3, debug-location !4
|
|
%val = load i160, ptr addrspace(1) %src
|
|
%trunc = trunc i160 %val to i128, !dbg !4
|
|
store i128 %trunc, ptr addrspace(1) %dst
|
|
ret void
|
|
}
|
|
|
|
; --- sub0_sub1_sub2_sub3_sub4: 160-bit from 192-bit ---
|
|
|
|
define amdgpu_ps void @s_trunc_i192_to_i160(ptr addrspace(1) inreg %src, ptr addrspace(1) inreg %dst) {
|
|
; GFX-950-LABEL: name: s_trunc_i192_to_i160
|
|
; GFX-950: bb.1 (%ir-block.0):
|
|
; GFX-950-NEXT: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3
|
|
; GFX-950-NEXT: {{ $}}
|
|
; GFX-950-NEXT: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr0
|
|
; GFX-950-NEXT: [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr1
|
|
; GFX-950-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
|
|
; GFX-950-NEXT: [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr2
|
|
; GFX-950-NEXT: [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr3
|
|
; GFX-950-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64_xexec_xnull = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY3]], %subreg.sub1
|
|
; GFX-950-NEXT: early-clobber %18:sgpr_128 = S_LOAD_DWORDX4_IMM_ec [[REG_SEQUENCE]], 0, 0 :: ("amdgpu-noclobber" load (<4 x s32>) from %ir.src, align 8, addrspace 1)
|
|
; GFX-950-NEXT: [[S_LOAD_DWORDX2_IMM:%[0-9]+]]:sreg_64_xexec = S_LOAD_DWORDX2_IMM [[REG_SEQUENCE]], 16, 0 :: ("amdgpu-noclobber" load (<2 x s32>) from %ir.src + 16, addrspace 1)
|
|
; GFX-950-NEXT: [[COPY4:%[0-9]+]]:sreg_32 = COPY %18.sub0
|
|
; GFX-950-NEXT: [[COPY5:%[0-9]+]]:sreg_32 = COPY %18.sub1
|
|
; GFX-950-NEXT: [[COPY6:%[0-9]+]]:sreg_32 = COPY %18.sub2
|
|
; GFX-950-NEXT: [[COPY7:%[0-9]+]]:sreg_32 = COPY %18.sub3
|
|
; GFX-950-NEXT: [[COPY8:%[0-9]+]]:sreg_32 = COPY [[S_LOAD_DWORDX2_IMM]].sub0
|
|
; GFX-950-NEXT: [[COPY9:%[0-9]+]]:sreg_32 = COPY [[S_LOAD_DWORDX2_IMM]].sub1
|
|
; GFX-950-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_192 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY6]], %subreg.sub2, [[COPY7]], %subreg.sub3, [[COPY8]], %subreg.sub4, [[COPY9]], %subreg.sub5
|
|
; GFX-950-NEXT: [[COPY10:%[0-9]+]]:sgpr_160 = COPY [[REG_SEQUENCE2]].sub0_sub1_sub2_sub3_sub4, debug-location !4
|
|
%val = load i192, ptr addrspace(1) %src
|
|
%trunc = trunc i192 %val to i160, !dbg !4
|
|
store i160 %trunc, ptr addrspace(1) %dst
|
|
ret void
|
|
}
|
|
|
|
define void @v_trunc_i192_to_i160(ptr addrspace(1) %src, ptr addrspace(1) %dst) {
|
|
; GFX-950-LABEL: name: v_trunc_i192_to_i160
|
|
; GFX-950: bb.1 (%ir-block.0):
|
|
; GFX-950-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
|
|
; GFX-950-NEXT: {{ $}}
|
|
; GFX-950-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
|
|
; GFX-950-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
|
|
; GFX-950-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
|
|
; GFX-950-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
|
|
; GFX-950-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
|
|
; GFX-950-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY3]], %subreg.sub1
|
|
; GFX-950-NEXT: [[GLOBAL_LOAD_DWORDX4_:%[0-9]+]]:vreg_128_align2 = GLOBAL_LOAD_DWORDX4 [[REG_SEQUENCE]], 0, 0, implicit $exec :: (load (<4 x s32>) from %ir.src, align 8, addrspace 1)
|
|
; GFX-950-NEXT: [[GLOBAL_LOAD_DWORDX2_:%[0-9]+]]:vreg_64_align2 = GLOBAL_LOAD_DWORDX2 [[REG_SEQUENCE]], 16, 0, implicit $exec :: (load (<2 x s32>) from %ir.src + 16, addrspace 1)
|
|
; GFX-950-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[GLOBAL_LOAD_DWORDX4_]].sub0
|
|
; GFX-950-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[GLOBAL_LOAD_DWORDX4_]].sub1
|
|
; GFX-950-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[GLOBAL_LOAD_DWORDX4_]].sub2
|
|
; GFX-950-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[GLOBAL_LOAD_DWORDX4_]].sub3
|
|
; GFX-950-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[GLOBAL_LOAD_DWORDX2_]].sub0
|
|
; GFX-950-NEXT: [[COPY9:%[0-9]+]]:vgpr_32 = COPY [[GLOBAL_LOAD_DWORDX2_]].sub1
|
|
; GFX-950-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_192_align2 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY6]], %subreg.sub2, [[COPY7]], %subreg.sub3, [[COPY8]], %subreg.sub4, [[COPY9]], %subreg.sub5
|
|
; GFX-950-NEXT: [[COPY10:%[0-9]+]]:vreg_160_align2 = COPY [[REG_SEQUENCE2]].sub0_sub1_sub2_sub3_sub4, debug-location !4
|
|
%val = load i192, ptr addrspace(1) %src
|
|
%trunc = trunc i192 %val to i160, !dbg !4
|
|
store i160 %trunc, ptr addrspace(1) %dst
|
|
ret void
|
|
}
|
|
|
|
; --- sub0_sub1_sub2_sub3_sub4_sub5: 192-bit from 224-bit ---
|
|
|
|
define amdgpu_ps void @s_trunc_i224_to_i192(ptr addrspace(1) inreg %src, ptr addrspace(1) inreg %dst) {
|
|
; GFX-950-LABEL: name: s_trunc_i224_to_i192
|
|
; GFX-950: bb.1 (%ir-block.0):
|
|
; GFX-950-NEXT: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3
|
|
; GFX-950-NEXT: {{ $}}
|
|
; GFX-950-NEXT: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr0
|
|
; GFX-950-NEXT: [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr1
|
|
; GFX-950-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
|
|
; GFX-950-NEXT: [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr2
|
|
; GFX-950-NEXT: [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr3
|
|
; GFX-950-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64_xexec_xnull = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY3]], %subreg.sub1
|
|
; GFX-950-NEXT: early-clobber %16:sgpr_128 = S_LOAD_DWORDX4_IMM_ec [[REG_SEQUENCE]], 0, 0 :: ("amdgpu-noclobber" load (<4 x s32>) from %ir.src, align 8, addrspace 1)
|
|
; GFX-950-NEXT: [[S_LOAD_DWORDX2_IMM:%[0-9]+]]:sreg_64_xexec = S_LOAD_DWORDX2_IMM [[REG_SEQUENCE]], 16, 0 :: ("amdgpu-noclobber" load (<2 x s32>) from %ir.src + 16, addrspace 1)
|
|
; GFX-950-NEXT: [[S_LOAD_DWORD_IMM:%[0-9]+]]:sreg_32_xm0_xexec = S_LOAD_DWORD_IMM [[REG_SEQUENCE]], 24, 0 :: ("amdgpu-noclobber" load (s32) from %ir.src + 24, align 8, addrspace 1)
|
|
; GFX-950-NEXT: [[COPY4:%[0-9]+]]:sreg_32 = COPY [[S_LOAD_DWORDX2_IMM]].sub0
|
|
; GFX-950-NEXT: [[COPY5:%[0-9]+]]:sreg_32 = COPY [[S_LOAD_DWORDX2_IMM]].sub1
|
|
; GFX-950-NEXT: [[COPY6:%[0-9]+]]:sreg_32 = COPY %16.sub0
|
|
; GFX-950-NEXT: [[COPY7:%[0-9]+]]:sreg_32 = COPY %16.sub1
|
|
; GFX-950-NEXT: [[COPY8:%[0-9]+]]:sreg_32 = COPY %16.sub2
|
|
; GFX-950-NEXT: [[COPY9:%[0-9]+]]:sreg_32 = COPY %16.sub3
|
|
; GFX-950-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_224 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY7]], %subreg.sub1, [[COPY8]], %subreg.sub2, [[COPY9]], %subreg.sub3, [[COPY4]], %subreg.sub4, [[COPY5]], %subreg.sub5, [[S_LOAD_DWORD_IMM]], %subreg.sub6
|
|
; GFX-950-NEXT: [[COPY10:%[0-9]+]]:sgpr_192 = COPY [[REG_SEQUENCE2]].sub0_sub1_sub2_sub3_sub4_sub5, debug-location !4
|
|
%val = load i224, ptr addrspace(1) %src
|
|
%trunc = trunc i224 %val to i192, !dbg !4
|
|
store i192 %trunc, ptr addrspace(1) %dst
|
|
ret void
|
|
}
|
|
|
|
define void @v_trunc_i224_to_i192(ptr addrspace(1) %src, ptr addrspace(1) %dst) {
|
|
; GFX-950-LABEL: name: v_trunc_i224_to_i192
|
|
; GFX-950: bb.1 (%ir-block.0):
|
|
; GFX-950-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
|
|
; GFX-950-NEXT: {{ $}}
|
|
; GFX-950-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
|
|
; GFX-950-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
|
|
; GFX-950-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
|
|
; GFX-950-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
|
|
; GFX-950-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
|
|
; GFX-950-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY3]], %subreg.sub1
|
|
; GFX-950-NEXT: [[GLOBAL_LOAD_DWORDX4_:%[0-9]+]]:vreg_128_align2 = GLOBAL_LOAD_DWORDX4 [[REG_SEQUENCE]], 0, 0, implicit $exec :: (load (<4 x s32>) from %ir.src, align 8, addrspace 1)
|
|
; GFX-950-NEXT: [[GLOBAL_LOAD_DWORDX3_:%[0-9]+]]:vreg_96_align2 = GLOBAL_LOAD_DWORDX3 [[REG_SEQUENCE]], 16, 0, implicit $exec :: (load (<3 x s32>) from %ir.src + 16, align 8, addrspace 1)
|
|
; GFX-950-NEXT: [[COPY4:%[0-9]+]]:vgpr_32 = COPY [[GLOBAL_LOAD_DWORDX4_]].sub0
|
|
; GFX-950-NEXT: [[COPY5:%[0-9]+]]:vgpr_32 = COPY [[GLOBAL_LOAD_DWORDX4_]].sub1
|
|
; GFX-950-NEXT: [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[GLOBAL_LOAD_DWORDX4_]].sub2
|
|
; GFX-950-NEXT: [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[GLOBAL_LOAD_DWORDX4_]].sub3
|
|
; GFX-950-NEXT: [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[GLOBAL_LOAD_DWORDX3_]].sub0
|
|
; GFX-950-NEXT: [[COPY9:%[0-9]+]]:vgpr_32 = COPY [[GLOBAL_LOAD_DWORDX3_]].sub1
|
|
; GFX-950-NEXT: [[COPY10:%[0-9]+]]:vgpr_32 = COPY [[GLOBAL_LOAD_DWORDX3_]].sub2
|
|
; GFX-950-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_224_align2 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1, [[COPY6]], %subreg.sub2, [[COPY7]], %subreg.sub3, [[COPY8]], %subreg.sub4, [[COPY9]], %subreg.sub5, [[COPY10]], %subreg.sub6
|
|
; GFX-950-NEXT: [[COPY11:%[0-9]+]]:vreg_192_align2 = COPY [[REG_SEQUENCE2]].sub0_sub1_sub2_sub3_sub4_sub5, debug-location !4
|
|
%val = load i224, ptr addrspace(1) %src
|
|
%trunc = trunc i224 %val to i192, !dbg !4
|
|
store i192 %trunc, ptr addrspace(1) %dst
|
|
ret void
|
|
}
|
|
|
|
; --- sub0_sub1_sub2_sub3_sub4_sub5_sub6: 224-bit from 256-bit ---
|
|
|
|
define amdgpu_ps void @s_trunc_i256_to_i224(ptr addrspace(1) inreg %src, ptr addrspace(1) inreg %dst) {
|
|
; GFX-950-LABEL: name: s_trunc_i256_to_i224
|
|
; GFX-950: bb.1 (%ir-block.0):
|
|
; GFX-950-NEXT: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3
|
|
; GFX-950-NEXT: {{ $}}
|
|
; GFX-950-NEXT: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr0
|
|
; GFX-950-NEXT: [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr1
|
|
; GFX-950-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
|
|
; GFX-950-NEXT: [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr2
|
|
; GFX-950-NEXT: [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr3
|
|
; GFX-950-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64_xexec_xnull = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY3]], %subreg.sub1
|
|
; GFX-950-NEXT: early-clobber %20:sgpr_256 = S_LOAD_DWORDX8_IMM_ec [[REG_SEQUENCE]], 0, 0 :: ("amdgpu-noclobber" load (<8 x s32>) from %ir.src, align 8, addrspace 1)
|
|
; GFX-950-NEXT: [[COPY4:%[0-9]+]]:sgpr_224 = COPY %20.lo16_hi16_sub1_lo16_sub1_hi16_sub2_lo16_sub2_hi16_sub3_lo16_sub3_hi16_sub4_lo16_sub4_hi16_sub5_lo16_sub5_hi16_sub6_lo16_sub6_hi16, debug-location !4
|
|
%val = load i256, ptr addrspace(1) %src
|
|
%trunc = trunc i256 %val to i224, !dbg !4
|
|
store i224 %trunc, ptr addrspace(1) %dst
|
|
ret void
|
|
}
|
|
|
|
define void @v_trunc_i256_to_i224(ptr addrspace(1) %src, ptr addrspace(1) %dst) {
|
|
; GFX-950-LABEL: name: v_trunc_i256_to_i224
|
|
; GFX-950: bb.1 (%ir-block.0):
|
|
; GFX-950-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
|
|
; GFX-950-NEXT: {{ $}}
|
|
; GFX-950-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
|
|
; GFX-950-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
|
|
; GFX-950-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
|
|
; GFX-950-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
|
|
; GFX-950-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
|
|
; GFX-950-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY3]], %subreg.sub1
|
|
; GFX-950-NEXT: [[GLOBAL_LOAD_DWORDX4_:%[0-9]+]]:vreg_128_align2 = GLOBAL_LOAD_DWORDX4 [[REG_SEQUENCE]], 0, 0, implicit $exec :: (load (<4 x s32>) from %ir.src, align 8, addrspace 1)
|
|
; GFX-950-NEXT: [[GLOBAL_LOAD_DWORDX4_1:%[0-9]+]]:vreg_128_align2 = GLOBAL_LOAD_DWORDX4 [[REG_SEQUENCE]], 16, 0, implicit $exec :: (load (<4 x s32>) from %ir.src + 16, align 8, addrspace 1)
|
|
; GFX-950-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_256_align2 = REG_SEQUENCE [[GLOBAL_LOAD_DWORDX4_]], %subreg.sub0_sub1_sub2_sub3, [[GLOBAL_LOAD_DWORDX4_1]], %subreg.sub4_sub5_sub6_sub7
|
|
; GFX-950-NEXT: [[COPY4:%[0-9]+]]:vreg_224_align2 = COPY [[REG_SEQUENCE2]].lo16_hi16_sub1_lo16_sub1_hi16_sub2_lo16_sub2_hi16_sub3_lo16_sub3_hi16_sub4_lo16_sub4_hi16_sub5_lo16_sub5_hi16_sub6_lo16_sub6_hi16, debug-location !4
|
|
%val = load i256, ptr addrspace(1) %src
|
|
%trunc = trunc i256 %val to i224, !dbg !4
|
|
store i224 %trunc, ptr addrspace(1) %dst
|
|
ret void
|
|
}
|
|
|
|
; --- sub0..sub15: 512-bit from 1024-bit ---
|
|
|
|
define amdgpu_ps void @s_trunc_i1024_to_i512(ptr addrspace(1) inreg %src, ptr addrspace(1) inreg %dst) {
|
|
; GFX-950-LABEL: name: s_trunc_i1024_to_i512
|
|
; GFX-950: bb.1 (%ir-block.0):
|
|
; GFX-950-NEXT: liveins: $sgpr0, $sgpr1, $sgpr2, $sgpr3
|
|
; GFX-950-NEXT: {{ $}}
|
|
; GFX-950-NEXT: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr0
|
|
; GFX-950-NEXT: [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr1
|
|
; GFX-950-NEXT: [[REG_SEQUENCE:%[0-9]+]]:sreg_64 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
|
|
; GFX-950-NEXT: [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr2
|
|
; GFX-950-NEXT: [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr3
|
|
; GFX-950-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64_xexec_xnull = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY3]], %subreg.sub1
|
|
; GFX-950-NEXT: early-clobber %20:sgpr_512 = S_LOAD_DWORDX16_IMM_ec [[REG_SEQUENCE]], 0, 0 :: ("amdgpu-noclobber" load (<16 x s32>) from %ir.src, align 8, addrspace 1)
|
|
; GFX-950-NEXT: early-clobber %23:sgpr_512 = S_LOAD_DWORDX16_IMM_ec [[REG_SEQUENCE]], 64, 0 :: ("amdgpu-noclobber" load (<16 x s32>) from %ir.src + 64, align 8, addrspace 1)
|
|
; GFX-950-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:sgpr_1024 = REG_SEQUENCE %20, %subreg.sub0_sub1_sub2_sub3_sub4_sub5_sub6_sub7_sub8_sub9_sub10_sub11_sub12_sub13_sub14_sub15, %23, %subreg.sub16_sub17_sub18_sub19_sub20_sub21_sub22_sub23_sub24_sub25_sub26_sub27_sub28_sub29_sub30_sub31
|
|
; GFX-950-NEXT: [[COPY4:%[0-9]+]]:sgpr_512 = COPY [[REG_SEQUENCE2]].sub0_sub1_sub2_sub3_sub4_sub5_sub6_sub7_sub8_sub9_sub10_sub11_sub12_sub13_sub14_sub15, debug-location !4
|
|
%val = load i1024, ptr addrspace(1) %src
|
|
%trunc = trunc i1024 %val to i512, !dbg !4
|
|
store i512 %trunc, ptr addrspace(1) %dst
|
|
ret void
|
|
}
|
|
|
|
define void @v_trunc_i1024_to_i512(ptr addrspace(1) %src, ptr addrspace(1) %dst) {
|
|
; GFX-950-LABEL: name: v_trunc_i1024_to_i512
|
|
; GFX-950: bb.1 (%ir-block.0):
|
|
; GFX-950-NEXT: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
|
|
; GFX-950-NEXT: {{ $}}
|
|
; GFX-950-NEXT: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
|
|
; GFX-950-NEXT: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
|
|
; GFX-950-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
|
|
; GFX-950-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2
|
|
; GFX-950-NEXT: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3
|
|
; GFX-950-NEXT: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64_align2 = REG_SEQUENCE [[COPY2]], %subreg.sub0, [[COPY3]], %subreg.sub1
|
|
; GFX-950-NEXT: [[GLOBAL_LOAD_DWORDX4_:%[0-9]+]]:vreg_128_align2 = GLOBAL_LOAD_DWORDX4 [[REG_SEQUENCE]], 0, 0, implicit $exec :: (load (<4 x s32>) from %ir.src, align 8, addrspace 1)
|
|
; GFX-950-NEXT: [[GLOBAL_LOAD_DWORDX4_1:%[0-9]+]]:vreg_128_align2 = GLOBAL_LOAD_DWORDX4 [[REG_SEQUENCE]], 16, 0, implicit $exec :: (load (<4 x s32>) from %ir.src + 16, align 8, addrspace 1)
|
|
; GFX-950-NEXT: [[GLOBAL_LOAD_DWORDX4_2:%[0-9]+]]:vreg_128_align2 = GLOBAL_LOAD_DWORDX4 [[REG_SEQUENCE]], 32, 0, implicit $exec :: (load (<4 x s32>) from %ir.src + 32, align 8, addrspace 1)
|
|
; GFX-950-NEXT: [[GLOBAL_LOAD_DWORDX4_3:%[0-9]+]]:vreg_128_align2 = GLOBAL_LOAD_DWORDX4 [[REG_SEQUENCE]], 48, 0, implicit $exec :: (load (<4 x s32>) from %ir.src + 48, align 8, addrspace 1)
|
|
; GFX-950-NEXT: [[REG_SEQUENCE2:%[0-9]+]]:vreg_512_align2 = REG_SEQUENCE [[GLOBAL_LOAD_DWORDX4_]], %subreg.sub0_sub1_sub2_sub3, [[GLOBAL_LOAD_DWORDX4_1]], %subreg.sub4_sub5_sub6_sub7, [[GLOBAL_LOAD_DWORDX4_2]], %subreg.sub8_sub9_sub10_sub11, [[GLOBAL_LOAD_DWORDX4_3]], %subreg.sub12_sub13_sub14_sub15
|
|
; GFX-950-NEXT: [[GLOBAL_LOAD_DWORDX4_4:%[0-9]+]]:vreg_128_align2 = GLOBAL_LOAD_DWORDX4 [[REG_SEQUENCE]], 64, 0, implicit $exec :: (load (<4 x s32>) from %ir.src + 64, align 8, addrspace 1)
|
|
; GFX-950-NEXT: [[GLOBAL_LOAD_DWORDX4_5:%[0-9]+]]:vreg_128_align2 = GLOBAL_LOAD_DWORDX4 [[REG_SEQUENCE]], 80, 0, implicit $exec :: (load (<4 x s32>) from %ir.src + 80, align 8, addrspace 1)
|
|
; GFX-950-NEXT: [[GLOBAL_LOAD_DWORDX4_6:%[0-9]+]]:vreg_128_align2 = GLOBAL_LOAD_DWORDX4 [[REG_SEQUENCE]], 96, 0, implicit $exec :: (load (<4 x s32>) from %ir.src + 96, align 8, addrspace 1)
|
|
; GFX-950-NEXT: [[GLOBAL_LOAD_DWORDX4_7:%[0-9]+]]:vreg_128_align2 = GLOBAL_LOAD_DWORDX4 [[REG_SEQUENCE]], 112, 0, implicit $exec :: (load (<4 x s32>) from %ir.src + 112, align 8, addrspace 1)
|
|
; GFX-950-NEXT: [[REG_SEQUENCE3:%[0-9]+]]:vreg_512_align2 = REG_SEQUENCE [[GLOBAL_LOAD_DWORDX4_4]], %subreg.sub0_sub1_sub2_sub3, [[GLOBAL_LOAD_DWORDX4_5]], %subreg.sub4_sub5_sub6_sub7, [[GLOBAL_LOAD_DWORDX4_6]], %subreg.sub8_sub9_sub10_sub11, [[GLOBAL_LOAD_DWORDX4_7]], %subreg.sub12_sub13_sub14_sub15
|
|
; GFX-950-NEXT: [[REG_SEQUENCE4:%[0-9]+]]:vreg_1024_align2 = REG_SEQUENCE [[REG_SEQUENCE2]], %subreg.sub0_sub1_sub2_sub3_sub4_sub5_sub6_sub7_sub8_sub9_sub10_sub11_sub12_sub13_sub14_sub15, [[REG_SEQUENCE3]], %subreg.sub16_sub17_sub18_sub19_sub20_sub21_sub22_sub23_sub24_sub25_sub26_sub27_sub28_sub29_sub30_sub31
|
|
; GFX-950-NEXT: [[COPY4:%[0-9]+]]:vreg_512_align2 = COPY [[REG_SEQUENCE4]].sub0_sub1_sub2_sub3_sub4_sub5_sub6_sub7_sub8_sub9_sub10_sub11_sub12_sub13_sub14_sub15, debug-location !4
|
|
%val = load i1024, ptr addrspace(1) %src
|
|
%trunc = trunc i1024 %val to i512, !dbg !4
|
|
store i512 %trunc, ptr addrspace(1) %dst
|
|
ret void
|
|
}
|
|
|
|
!llvm.dbg.cu = !{!0}
|
|
!5 = !{null}
|
|
!6 = !DISubroutineType(types: !5)
|
|
!llvm.module.flags = !{!2}
|
|
|
|
!0 = distinct !DICompileUnit(language: DW_LANG_C99, file: !1)
|
|
!1 = !DIFile(filename: ".", directory: ".")
|
|
!2 = !{i32 2, !"Debug Info Version", i32 3}
|
|
!3 = distinct !DISubprogram(name: "trunc", scope: !1, file: !1, line: 1, type: !6, unit: !0)
|
|
!4 = !DILocation(line: 1, scope: !3)
|