1 ; RUN: llc -mtriple=amdgcn--amdhsa -mcpu=fiji -verify-machineinstrs < %s | FileCheck %s 2 3 declare i32 @llvm.amdgcn.readfirstlane(i32) #0 4 5 ; CHECK-LABEL: {{^}}test_readfirstlane: 6 ; CHECK: v_readfirstlane_b32 s{{[0-9]+}}, v{{[0-9]+}} 7 define amdgpu_kernel void @test_readfirstlane(i32 addrspace(1)* %out, i32 %src) #1 { 8 %readfirstlane = call i32 @llvm.amdgcn.readfirstlane(i32 %src) 9 store i32 %readfirstlane, i32 addrspace(1)* %out, align 4 10 ret void 11 } 12 13 ; CHECK-LABEL: {{^}}test_readfirstlane_imm: 14 ; CHECK: v_mov_b32_e32 [[VVAL:v[0-9]]], 32 15 ; CHECK: v_readfirstlane_b32 s{{[0-9]+}}, [[VVAL]] 16 define amdgpu_kernel void @test_readfirstlane_imm(i32 addrspace(1)* %out) #1 { 17 %readfirstlane = call i32 @llvm.amdgcn.readfirstlane(i32 32) 18 store i32 %readfirstlane, i32 addrspace(1)* %out, align 4 19 ret void 20 } 21 22 ; TODO: m0 should be folded. 23 ; CHECK-LABEL: {{^}}test_readfirstlane_m0: 24 ; CHECK: s_mov_b32 m0, -1 25 ; CHECK: s_mov_b32 [[COPY_M0:s[0-9]+]], m0 26 ; CHECK: v_mov_b32_e32 [[VVAL:v[0-9]]], [[COPY_M0]] 27 ; CHECK: v_readfirstlane_b32 s{{[0-9]+}}, [[VVAL]] 28 define amdgpu_kernel void @test_readfirstlane_m0(i32 addrspace(1)* %out) #1 { 29 %m0 = call i32 asm "s_mov_b32 m0, -1", "={M0}"() 30 %readfirstlane = call i32 @llvm.amdgcn.readfirstlane(i32 %m0) 31 store i32 %readfirstlane, i32 addrspace(1)* %out, align 4 32 ret void 33 } 34 35 attributes #0 = { nounwind readnone convergent } 36 attributes #1 = { nounwind } 37
