77 lines · plain
1; RUN: llc -mtriple=amdgcn < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s2; RUN: llc -mtriple=amdgcn -mcpu=tonga -mattr=-flat-for-global < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s3 4declare double @llvm.maxnum.f64(double, double) #05declare <2 x double> @llvm.maxnum.v2f64(<2 x double>, <2 x double>) #06declare <4 x double> @llvm.maxnum.v4f64(<4 x double>, <4 x double>) #07declare <8 x double> @llvm.maxnum.v8f64(<8 x double>, <8 x double>) #08declare <16 x double> @llvm.maxnum.v16f64(<16 x double>, <16 x double>) #09 10; FUNC-LABEL: @test_fmax_f6411; SI: v_max_f6412define amdgpu_kernel void @test_fmax_f64(ptr addrspace(1) %out, double %a, double %b) nounwind {13 %val = call double @llvm.maxnum.f64(double %a, double %b) #014 store double %val, ptr addrspace(1) %out, align 815 ret void16}17 18; FUNC-LABEL: @test_fmax_v2f6419; SI: v_max_f6420; SI: v_max_f6421define amdgpu_kernel void @test_fmax_v2f64(ptr addrspace(1) %out, <2 x double> %a, <2 x double> %b) nounwind {22 %val = call <2 x double> @llvm.maxnum.v2f64(<2 x double> %a, <2 x double> %b) #023 store <2 x double> %val, ptr addrspace(1) %out, align 1624 ret void25}26 27; FUNC-LABEL: @test_fmax_v4f6428; SI: v_max_f6429; SI: v_max_f6430; SI: v_max_f6431; SI: v_max_f6432define amdgpu_kernel void @test_fmax_v4f64(ptr addrspace(1) %out, <4 x double> %a, <4 x double> %b) nounwind {33 %val = call <4 x double> @llvm.maxnum.v4f64(<4 x double> %a, <4 x double> %b) #034 store <4 x double> %val, ptr addrspace(1) %out, align 3235 ret void36}37 38; FUNC-LABEL: @test_fmax_v8f6439; SI: v_max_f6440; SI: v_max_f6441; SI: v_max_f6442; SI: v_max_f6443; SI: v_max_f6444; SI: v_max_f6445; SI: v_max_f6446; SI: v_max_f6447define amdgpu_kernel void @test_fmax_v8f64(ptr addrspace(1) %out, <8 x double> %a, <8 x double> %b) nounwind {48 %val = call <8 x double> @llvm.maxnum.v8f64(<8 x double> %a, <8 x double> %b) #049 store <8 x double> %val, ptr addrspace(1) %out, align 6450 ret void51}52 53; FUNC-LABEL: @test_fmax_v16f6454; SI: v_max_f6455; SI: v_max_f6456; SI: v_max_f6457; SI: v_max_f6458; SI: v_max_f6459; SI: v_max_f6460; SI: v_max_f6461; SI: v_max_f6462; SI: v_max_f6463; SI: v_max_f6464; SI: v_max_f6465; SI: v_max_f6466; SI: v_max_f6467; SI: v_max_f6468; SI: v_max_f6469; SI: v_max_f6470define amdgpu_kernel void @test_fmax_v16f64(ptr addrspace(1) %out, <16 x double> %a, <16 x double> %b) nounwind {71 %val = call <16 x double> @llvm.maxnum.v16f64(<16 x double> %a, <16 x double> %b) #072 store <16 x double> %val, ptr addrspace(1) %out, align 12873 ret void74}75 76attributes #0 = { nounwind readnone }77