derek/TheRock
1
0 Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity
Files
1a182e2dac58879f215c0fad0f5f78c76dcafe56
TheRock/compiler/amd-llvm/llvm/test/CodeGen/NVPTX/scalarize-non-coalescable-v2f32.ll
T
derek.holloway 7502bf64a4
CodeQL / Analyze (actions) (push) Canceled after 0s
Details
CodeQL / Analyze (python) (push) Canceled after 0s
Details
pre-commit / pre-commit (push) Canceled after 0s
Details
Multi-Arch CI / setup (push) Canceled after 0s
Details
Publish build_manylinux_x86_64 images / publish_build_manylinux_x86_64 (push) Canceled after 0s
Details
Setup (Multi-Arch) / setup (push) Canceled after 0s
Details
Multi-Arch CI / Linux::${{ fromJSON(needs.setup.outputs.linux_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Details
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Details
Publish no_rocm_image_ubi10 image / publish_no_rocm_image_ubi10 (push) Canceled after 0s
Details
Multi-Arch CI / Windows::${{ fromJSON(needs.setup.outputs.windows_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Details
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Details
Multi-Arch CI / Manifest Diff (push) Canceled after 0s
Details
TheRock Manifest Diff Report / Generate Manifest Diff Report (push) Canceled after 0s
Details
Multi-Arch CI / CI Summary (push) Canceled after 0s
Details
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04 (push) Canceled after 0s
Details
Publish no_rocm_image_ubuntu24_04_ocl_rt images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Details
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Details
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Details
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Details
Publish no_rocm_image_ubuntu24_04_rocgdb images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Details
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Details
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Details
Unit Tests / Unit Tests :: ubuntu-24.04 (push) Canceled after 0s
Details
Unit Tests / Unit Tests :: windows-2022 (push) Canceled after 0s
Details
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_media (push) Canceled after 0s
Details
Unit Tests / Unit Tests Summary (push) Canceled after 0s
Details
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_openmpi (push) Canceled after 0s
Details
Flatten subprojects into a single base monorepo
2026-09-02 16:34:47 -07:00

357 lines
15 KiB
LLVM
Raw Blame History

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
; RUN: llc < %s -mtriple=nvptx64 -mcpu=sm_100 -fp-contract=fast | FileCheck %s
; RUN: %if ptxas-sm_100 %{ llc < %s -mtriple=nvptx64 -mcpu=sm_100 -fp-contract=fast | %ptxas-verify -arch sm_100 %}
target triple = "nvptx64-nvidia-cuda"
define <2 x float> @fma_non_coalescable(ptr addrspace(3) %p, <2 x float> %a, <2 x float> %c) {
; CHECK-LABEL: fma_non_coalescable(
; CHECK: {
; CHECK-NEXT: .reg .b32 %r<11>;
; CHECK-NEXT: .reg .b64 %rd<2>;
; CHECK-EMPTY:
; CHECK-NEXT: // %bb.0:
; CHECK-NEXT: ld.param.b64 %rd1, [fma_non_coalescable_param_0];
; CHECK-NEXT: ld.shared.v4.b32 {%r1, %r2, %r3, %r4}, [%rd1];
; CHECK-NEXT: ld.param.v2.b32 {%r5, %r6}, [fma_non_coalescable_param_1];
; CHECK-NEXT: ld.param.v2.b32 {%r7, %r8}, [fma_non_coalescable_param_2];
; CHECK-NEXT: fma.rn.f32 %r9, %r6, %r3, %r8;
; CHECK-NEXT: fma.rn.f32 %r10, %r5, %r1, %r7;
; CHECK-NEXT: st.param.v2.b32 [func_retval0], {%r10, %r9};
; CHECK-NEXT: ret;
%ld = load <4 x float>, ptr addrspace(3) %p, align 16
%e0 = extractelement <4 x float> %ld, i32 0
%e2 = extractelement <4 x float> %ld, i32 2
%bv0 = insertelement <2 x float> poison, float %e0, i32 0
%bv = insertelement <2 x float> %bv0, float %e2, i32 1
%mul = fmul <2 x float> %a, %bv
%res = fadd <2 x float> %mul, %c
ret <2 x float> %res
}
define <2 x float> @fma_constant_and_dynamic_index(i32 %idx, ptr addrspace(3) %p, <2 x float> %a, <2 x float> %c) {
; CHECK-LABEL: fma_constant_and_dynamic_index(
; CHECK: {
; CHECK-NEXT: .local .align 8 .b8 __local_depot1[16];
; CHECK-NEXT: .reg .b64 %SP;
; CHECK-NEXT: .reg .b64 %SPL;
; CHECK-NEXT: .reg .b32 %r<9>;
; CHECK-NEXT: .reg .b64 %rd<9>;
; CHECK-EMPTY:
; CHECK-NEXT: // %bb.0:
; CHECK-NEXT: mov.b64 %SPL, __local_depot1;
; CHECK-NEXT: cvta.local.u64 %SP, %SPL;
; CHECK-NEXT: ld.param.b32 %rd1, [fma_constant_and_dynamic_index_param_0];
; CHECK-NEXT: and.b64 %rd2, %rd1, 3;
; CHECK-NEXT: shl.b64 %rd3, %rd2, 2;
; CHECK-NEXT: add.u64 %rd4, %SP, 0;
; CHECK-NEXT: add.s64 %rd5, %rd4, %rd3;
; CHECK-NEXT: ld.param.b64 %rd6, [fma_constant_and_dynamic_index_param_1];
; CHECK-NEXT: ld.shared.v2.b64 {%rd7, %rd8}, [%rd6];
; CHECK-NEXT: mov.b64 {%r1, _}, %rd7;
; CHECK-NEXT: ld.param.v2.b32 {%r2, %r3}, [fma_constant_and_dynamic_index_param_2];
; CHECK-NEXT: st.b64 [%SP+8], %rd8;
; CHECK-NEXT: st.b64 [%SP], %rd7;
; CHECK-NEXT: ld.b32 %r4, [%rd5];
; CHECK-NEXT: ld.param.v2.b32 {%r5, %r6}, [fma_constant_and_dynamic_index_param_3];
; CHECK-NEXT: fma.rn.f32 %r7, %r2, %r1, %r5;
; CHECK-NEXT: fma.rn.f32 %r8, %r3, %r4, %r6;
; CHECK-NEXT: st.param.v2.b32 [func_retval0], {%r7, %r8};
; CHECK-NEXT: ret;
%ld = load <4 x float>, ptr addrspace(3) %p, align 16
%e0 = extractelement <4 x float> %ld, i32 0
%e1 = extractelement <4 x float> %ld, i32 %idx
%bv0 = insertelement <2 x float> poison, float %e0, i32 0
%bv = insertelement <2 x float> %bv0, float %e1, i32 1
%mul = fmul <2 x float> %a, %bv
%res = fadd <2 x float> %mul, %c
ret <2 x float> %res
}
define <2 x float> @fma_shufflevector_non_coalescable(ptr addrspace(3) %p, <2 x float> %a, <2 x float> %c) {
; CHECK-LABEL: fma_shufflevector_non_coalescable(
; CHECK: {
; CHECK-NEXT: .reg .b32 %r<11>;
; CHECK-NEXT: .reg .b64 %rd<2>;
; CHECK-EMPTY:
; CHECK-NEXT: // %bb.0:
; CHECK-NEXT: ld.param.b64 %rd1, [fma_shufflevector_non_coalescable_param_0];
; CHECK-NEXT: ld.shared.v4.b32 {%r1, %r2, %r3, %r4}, [%rd1];
; CHECK-NEXT: ld.param.v2.b32 {%r5, %r6}, [fma_shufflevector_non_coalescable_param_1];
; CHECK-NEXT: ld.param.v2.b32 {%r7, %r8}, [fma_shufflevector_non_coalescable_param_2];
; CHECK-NEXT: fma.rn.f32 %r9, %r6, %r3, %r8;
; CHECK-NEXT: fma.rn.f32 %r10, %r5, %r1, %r7;
; CHECK-NEXT: st.param.v2.b32 [func_retval0], {%r10, %r9};
; CHECK-NEXT: ret;
%ld = load <4 x float>, ptr addrspace(3) %p, align 16
%bv = shufflevector <4 x float> %ld, <4 x float> poison, <2 x i32> <i32 0, i32 2>
%mul = fmul <2 x float> %a, %bv
%res = fadd <2 x float> %mul, %c
ret <2 x float> %res
}
define <2 x float> @fma_mixed_extract_and_scalar(ptr addrspace(3) %p, float %s, <2 x float> %a, <2 x float> %c) {
; CHECK-LABEL: fma_mixed_extract_and_scalar(
; CHECK: {
; CHECK-NEXT: .reg .b32 %r<12>;
; CHECK-NEXT: .reg .b64 %rd<2>;
; CHECK-EMPTY:
; CHECK-NEXT: // %bb.0:
; CHECK-NEXT: ld.param.b64 %rd1, [fma_mixed_extract_and_scalar_param_0];
; CHECK-NEXT: ld.shared.v4.b32 {%r1, %r2, %r3, %r4}, [%rd1];
; CHECK-NEXT: ld.param.b32 %r5, [fma_mixed_extract_and_scalar_param_1];
; CHECK-NEXT: ld.param.v2.b32 {%r6, %r7}, [fma_mixed_extract_and_scalar_param_2];
; CHECK-NEXT: ld.param.v2.b32 {%r8, %r9}, [fma_mixed_extract_and_scalar_param_3];
; CHECK-NEXT: fma.rn.f32 %r10, %r7, %r5, %r9;
; CHECK-NEXT: fma.rn.f32 %r11, %r6, %r3, %r8;
; CHECK-NEXT: st.param.v2.b32 [func_retval0], {%r11, %r10};
; CHECK-NEXT: ret;
%ld = load <4 x float>, ptr addrspace(3) %p, align 16
%e2 = extractelement <4 x float> %ld, i32 2
%bv0 = insertelement <2 x float> poison, float %e2, i32 0
%bv = insertelement <2 x float> %bv0, float %s, i32 1
%mul = fmul <2 x float> %a, %bv
%res = fadd <2 x float> %mul, %c
ret <2 x float> %res
}
define <2 x float> @fadd_non_coalescable(ptr addrspace(3) %p, <2 x float> %a) {
; CHECK-LABEL: fadd_non_coalescable(
; CHECK: {
; CHECK-NEXT: .reg .b32 %r<9>;
; CHECK-NEXT: .reg .b64 %rd<2>;
; CHECK-EMPTY:
; CHECK-NEXT: // %bb.0:
; CHECK-NEXT: ld.param.b64 %rd1, [fadd_non_coalescable_param_0];
; CHECK-NEXT: ld.shared.v4.b32 {%r1, %r2, %r3, %r4}, [%rd1];
; CHECK-NEXT: ld.param.v2.b32 {%r5, %r6}, [fadd_non_coalescable_param_1];
; CHECK-NEXT: add.f32 %r7, %r6, %r3;
; CHECK-NEXT: add.f32 %r8, %r5, %r1;
; CHECK-NEXT: st.param.v2.b32 [func_retval0], {%r8, %r7};
; CHECK-NEXT: ret;
%ld = load <4 x float>, ptr addrspace(3) %p, align 16
%e0 = extractelement <4 x float> %ld, i32 0
%e2 = extractelement <4 x float> %ld, i32 2
%bv0 = insertelement <2 x float> poison, float %e0, i32 0
%bv = insertelement <2 x float> %bv0, float %e2, i32 1
%add = fadd <2 x float> %a, %bv
ret <2 x float> %add
}
define <2 x float> @fmul_non_coalescable(ptr addrspace(3) %p, <2 x float> %a) {
; CHECK-LABEL: fmul_non_coalescable(
; CHECK: {
; CHECK-NEXT: .reg .b32 %r<9>;
; CHECK-NEXT: .reg .b64 %rd<2>;
; CHECK-EMPTY:
; CHECK-NEXT: // %bb.0:
; CHECK-NEXT: ld.param.b64 %rd1, [fmul_non_coalescable_param_0];
; CHECK-NEXT: ld.shared.v4.b32 {%r1, %r2, %r3, %r4}, [%rd1];
; CHECK-NEXT: ld.param.v2.b32 {%r5, %r6}, [fmul_non_coalescable_param_1];
; CHECK-NEXT: mul.f32 %r7, %r6, %r3;
; CHECK-NEXT: mul.f32 %r8, %r5, %r1;
; CHECK-NEXT: st.param.v2.b32 [func_retval0], {%r8, %r7};
; CHECK-NEXT: ret;
%ld = load <4 x float>, ptr addrspace(3) %p, align 16
%e0 = extractelement <4 x float> %ld, i32 0
%e2 = extractelement <4 x float> %ld, i32 2
%bv0 = insertelement <2 x float> poison, float %e0, i32 0
%bv = insertelement <2 x float> %bv0, float %e2, i32 1
%mul = fmul <2 x float> %a, %bv
ret <2 x float> %mul
}
define <2 x float> @fsub_non_coalescable(ptr addrspace(3) %p, <2 x float> %a) {
; CHECK-LABEL: fsub_non_coalescable(
; CHECK: {
; CHECK-NEXT: .reg .b32 %r<9>;
; CHECK-NEXT: .reg .b64 %rd<2>;
; CHECK-EMPTY:
; CHECK-NEXT: // %bb.0:
; CHECK-NEXT: ld.param.b64 %rd1, [fsub_non_coalescable_param_0];
; CHECK-NEXT: ld.shared.v4.b32 {%r1, %r2, %r3, %r4}, [%rd1];
; CHECK-NEXT: ld.param.v2.b32 {%r5, %r6}, [fsub_non_coalescable_param_1];
; CHECK-NEXT: sub.f32 %r7, %r6, %r3;
; CHECK-NEXT: sub.f32 %r8, %r5, %r1;
; CHECK-NEXT: st.param.v2.b32 [func_retval0], {%r8, %r7};
; CHECK-NEXT: ret;
%ld = load <4 x float>, ptr addrspace(3) %p, align 16
%e0 = extractelement <4 x float> %ld, i32 0
%e2 = extractelement <4 x float> %ld, i32 2
%bv0 = insertelement <2 x float> poison, float %e0, i32 0
%bv = insertelement <2 x float> %bv0, float %e2, i32 1
%sub = fsub <2 x float> %a, %bv
ret <2 x float> %sub
}
; The rest of these tests should remain vectorized
define <4 x float> @fma_adjacent_elements(ptr addrspace(3) %p, <2 x float> %a, <2 x float> %c) {
; CHECK-LABEL: fma_adjacent_elements(
; CHECK: {
; CHECK-NEXT: .reg .b64 %rd<8>;
; CHECK-EMPTY:
; CHECK-NEXT: // %bb.0:
; CHECK-NEXT: ld.param.b64 %rd1, [fma_adjacent_elements_param_0];
; CHECK-NEXT: ld.shared.v2.b64 {%rd2, %rd3}, [%rd1];
; CHECK-NEXT: ld.param.b64 %rd4, [fma_adjacent_elements_param_1];
; CHECK-NEXT: ld.param.b64 %rd5, [fma_adjacent_elements_param_2];
; CHECK-NEXT: fma.rn.f32x2 %rd6, %rd4, %rd2, %rd5;
; CHECK-NEXT: fma.rn.f32x2 %rd7, %rd4, %rd3, %rd5;
; CHECK-NEXT: st.param.v2.b64 [func_retval0], {%rd6, %rd7};
; CHECK-NEXT: ret;
%ld = load <4 x float>, ptr addrspace(3) %p, align 16
%e0 = extractelement <4 x float> %ld, i32 0
%e1 = extractelement <4 x float> %ld, i32 1
%e2 = extractelement <4 x float> %ld, i32 2
%e3 = extractelement <4 x float> %ld, i32 3
%lo0 = insertelement <2 x float> poison, float %e0, i32 0
%lo = insertelement <2 x float> %lo0, float %e1, i32 1
%hi0 = insertelement <2 x float> poison, float %e2, i32 0
%hi = insertelement <2 x float> %hi0, float %e3, i32 1
%mul_lo = fmul <2 x float> %a, %lo
%res_lo = fadd <2 x float> %mul_lo, %c
%mul_hi = fmul <2 x float> %a, %hi
%res_hi = fadd <2 x float> %mul_hi, %c
%out = shufflevector <2 x float> %res_lo, <2 x float> %res_hi, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
ret <4 x float> %out
}
define <4 x float> @fma_adjacent_swapped_elements(ptr addrspace(3) %p, <2 x float> %a, <2 x float> %c) {
; CHECK-LABEL: fma_adjacent_swapped_elements(
; CHECK: {
; CHECK-NEXT: .reg .b32 %r<5>;
; CHECK-NEXT: .reg .b64 %rd<8>;
; CHECK-EMPTY:
; CHECK-NEXT: // %bb.0:
; CHECK-NEXT: ld.param.b64 %rd1, [fma_adjacent_swapped_elements_param_0];
; CHECK-NEXT: ld.shared.v4.b32 {%r1, %r2, %r3, %r4}, [%rd1];
; CHECK-NEXT: ld.param.b64 %rd2, [fma_adjacent_swapped_elements_param_1];
; CHECK-NEXT: ld.param.b64 %rd3, [fma_adjacent_swapped_elements_param_2];
; CHECK-NEXT: mov.b64 %rd4, {%r2, %r1};
; CHECK-NEXT: mov.b64 %rd5, {%r4, %r3};
; CHECK-NEXT: fma.rn.f32x2 %rd6, %rd2, %rd4, %rd3;
; CHECK-NEXT: fma.rn.f32x2 %rd7, %rd2, %rd5, %rd3;
; CHECK-NEXT: st.param.v2.b64 [func_retval0], {%rd6, %rd7};
; CHECK-NEXT: ret;
%ld = load <4 x float>, ptr addrspace(3) %p, align 16
%e0 = extractelement <4 x float> %ld, i32 0
%e1 = extractelement <4 x float> %ld, i32 1
%e2 = extractelement <4 x float> %ld, i32 2
%e3 = extractelement <4 x float> %ld, i32 3
%lo0 = insertelement <2 x float> poison, float %e1, i32 0
%lo = insertelement <2 x float> %lo0, float %e0, i32 1
%hi0 = insertelement <2 x float> poison, float %e3, i32 0
%hi = insertelement <2 x float> %hi0, float %e2, i32 1
%mul_lo = fmul <2 x float> %a, %lo
%res_lo = fadd <2 x float> %mul_lo, %c
%mul_hi = fmul <2 x float> %a, %hi
%res_hi = fadd <2 x float> %mul_hi, %c
%out = shufflevector <2 x float> %res_lo, <2 x float> %res_hi, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
ret <4 x float> %out
}
define <2 x float> @fma_both_dynamic_indices(i32 %i0, i32 %i1, ptr addrspace(3) %p, <2 x float> %a, <2 x float> %c) {
; CHECK-LABEL: fma_both_dynamic_indices(
; CHECK: {
; CHECK-NEXT: .local .align 8 .b8 __local_depot9[32];
; CHECK-NEXT: .reg .b64 %SP;
; CHECK-NEXT: .reg .b64 %SPL;
; CHECK-NEXT: .reg .b32 %r<3>;
; CHECK-NEXT: .reg .b64 %rd<18>;
; CHECK-EMPTY:
; CHECK-NEXT: // %bb.0:
; CHECK-NEXT: mov.b64 %SPL, __local_depot9;
; CHECK-NEXT: cvta.local.u64 %SP, %SPL;
; CHECK-NEXT: ld.param.b32 %rd1, [fma_both_dynamic_indices_param_0];
; CHECK-NEXT: and.b64 %rd2, %rd1, 3;
; CHECK-NEXT: shl.b64 %rd3, %rd2, 2;
; CHECK-NEXT: add.u64 %rd4, %SP, 16;
; CHECK-NEXT: add.s64 %rd5, %rd4, %rd3;
; CHECK-NEXT: ld.param.b32 %rd6, [fma_both_dynamic_indices_param_1];
; CHECK-NEXT: and.b64 %rd7, %rd6, 3;
; CHECK-NEXT: shl.b64 %rd8, %rd7, 2;
; CHECK-NEXT: add.u64 %rd9, %SP, 0;
; CHECK-NEXT: add.s64 %rd10, %rd9, %rd8;
; CHECK-NEXT: ld.param.b64 %rd11, [fma_both_dynamic_indices_param_2];
; CHECK-NEXT: ld.shared.v2.b64 {%rd12, %rd13}, [%rd11];
; CHECK-NEXT: st.b64 [%SP+24], %rd13;
; CHECK-NEXT: st.b64 [%SP+16], %rd12;
; CHECK-NEXT: ld.b32 %r1, [%rd5];
; CHECK-NEXT: st.b64 [%SP+8], %rd13;
; CHECK-NEXT: st.b64 [%SP], %rd12;
; CHECK-NEXT: ld.b32 %r2, [%rd10];
; CHECK-NEXT: ld.param.b64 %rd14, [fma_both_dynamic_indices_param_3];
; CHECK-NEXT: ld.param.b64 %rd15, [fma_both_dynamic_indices_param_4];
; CHECK-NEXT: mov.b64 %rd16, {%r1, %r2};
; CHECK-NEXT: fma.rn.f32x2 %rd17, %rd14, %rd16, %rd15;
; CHECK-NEXT: st.param.b64 [func_retval0], %rd17;
; CHECK-NEXT: ret;
%ld = load <4 x float>, ptr addrspace(3) %p, align 16
%e0 = extractelement <4 x float> %ld, i32 %i0
%e1 = extractelement <4 x float> %ld, i32 %i1
%bv0 = insertelement <2 x float> poison, float %e0, i32 0
%bv = insertelement <2 x float> %bv0, float %e1, i32 1
%mul = fmul <2 x float> %a, %bv
%res = fadd <2 x float> %mul, %c
ret <2 x float> %res
}
define <2 x float> @fma_shufflevector_adjacent_elements(ptr addrspace(3) %p, <2 x float> %a, <2 x float> %c) {
; CHECK-LABEL: fma_shufflevector_adjacent_elements(
; CHECK: {
; CHECK-NEXT: .reg .b64 %rd<6>;
; CHECK-EMPTY:
; CHECK-NEXT: // %bb.0:
; CHECK-NEXT: ld.param.b64 %rd1, [fma_shufflevector_adjacent_elements_param_0];
; CHECK-NEXT: ld.param.b64 %rd2, [fma_shufflevector_adjacent_elements_param_1];
; CHECK-NEXT: ld.shared.b64 %rd3, [%rd1];
; CHECK-NEXT: ld.param.b64 %rd4, [fma_shufflevector_adjacent_elements_param_2];
; CHECK-NEXT: fma.rn.f32x2 %rd5, %rd2, %rd3, %rd4;
; CHECK-NEXT: st.param.b64 [func_retval0], %rd5;
; CHECK-NEXT: ret;
%ld = load <4 x float>, ptr addrspace(3) %p, align 16
%bv = shufflevector <4 x float> %ld, <4 x float> poison, <2 x i32> <i32 0, i32 1>
%mul = fmul <2 x float> %a, %bv
%res = fadd <2 x float> %mul, %c
ret <2 x float> %res
}
define <2 x float> @fma_naturally_paired(<2 x float> %a, <2 x float> %b, <2 x float> %c) {
; CHECK-LABEL: fma_naturally_paired(
; CHECK: {
; CHECK-NEXT: .reg .b64 %rd<5>;
; CHECK-EMPTY:
; CHECK-NEXT: // %bb.0:
; CHECK-NEXT: ld.param.b64 %rd1, [fma_naturally_paired_param_0];
; CHECK-NEXT: ld.param.b64 %rd2, [fma_naturally_paired_param_1];
; CHECK-NEXT: ld.param.b64 %rd3, [fma_naturally_paired_param_2];
; CHECK-NEXT: fma.rn.f32x2 %rd4, %rd1, %rd2, %rd3;
; CHECK-NEXT: st.param.b64 [func_retval0], %rd4;
; CHECK-NEXT: ret;
%mul = fmul <2 x float> %a, %b
%res = fadd <2 x float> %mul, %c
ret <2 x float> %res
}
define <2 x float> @fma_coalescable_scalars(float %x, float %y, <2 x float> %b, <2 x float> %c) {
; CHECK-LABEL: fma_coalescable_scalars(
; CHECK: {
; CHECK-NEXT: .reg .b32 %r<3>;
; CHECK-NEXT: .reg .b64 %rd<5>;
; CHECK-EMPTY:
; CHECK-NEXT: // %bb.0:
; CHECK-NEXT: ld.param.b32 %r1, [fma_coalescable_scalars_param_0];
; CHECK-NEXT: ld.param.b32 %r2, [fma_coalescable_scalars_param_1];
; CHECK-NEXT: mov.b64 %rd1, {%r1, %r2};
; CHECK-NEXT: ld.param.b64 %rd2, [fma_coalescable_scalars_param_2];
; CHECK-NEXT: ld.param.b64 %rd3, [fma_coalescable_scalars_param_3];
; CHECK-NEXT: fma.rn.f32x2 %rd4, %rd1, %rd2, %rd3;
; CHECK-NEXT: st.param.b64 [func_retval0], %rd4;
; CHECK-NEXT: ret;
%v0 = insertelement <2 x float> poison, float %x, i32 0
%v = insertelement <2 x float> %v0, float %y, i32 1
%mul = fmul <2 x float> %v, %b
%res = fadd <2 x float> %mul, %c
ret <2 x float> %res
}
Reference in new issue View Git Blame Copy Permalink