80 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 22; RUN: llc -mtriple=amdgcn -mcpu=tahiti < %s | FileCheck -check-prefixes=GCN,GFX678,GFX6 %s3; RUN: llc -mtriple=amdgcn -mcpu=hawaii < %s | FileCheck -check-prefixes=GCN,GFX678,GFX7 %s4; RUN: llc -mtriple=amdgcn -mcpu=fiji < %s | FileCheck -check-prefixes=GCN,GFX678,GFX8 %s5; RUN: llc -mtriple=amdgcn -mcpu=gfx900 < %s | FileCheck -check-prefixes=GCN,GFX9 %s6; RUN: llc -mtriple=amdgcn -mcpu=gfx1030 < %s | FileCheck -check-prefixes=GCN,GFX1011,GFX10 %s7; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -amdgpu-enable-delay-alu=0 < %s | FileCheck -check-prefixes=GCN,GFX1011,GFX11 %s8 9declare i32 @llvm.get.rounding()10 11define i32 @func_rounding() {12; GFX678-LABEL: func_rounding:13; GFX678: ; %bb.0:14; GFX678-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)15; GFX678-NEXT: s_getreg_b32 s4, hwreg(HW_REG_MODE, 0, 4)16; GFX678-NEXT: s_lshl_b32 s6, s4, 217; GFX678-NEXT: s_mov_b32 s4, 0xeb24da7118; GFX678-NEXT: s_mov_b32 s5, 0xc96f38519; GFX678-NEXT: s_lshr_b64 s[4:5], s[4:5], s620; GFX678-NEXT: s_and_b32 s4, s4, 1521; GFX678-NEXT: s_add_i32 s5, s4, 422; GFX678-NEXT: s_cmp_lt_u32 s4, 423; GFX678-NEXT: s_cselect_b32 s4, s4, s524; GFX678-NEXT: v_mov_b32_e32 v0, s425; GFX678-NEXT: s_setpc_b64 s[30:31]26;27; GFX9-LABEL: func_rounding:28; GFX9: ; %bb.0:29; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)30; GFX9-NEXT: s_getreg_b32 s4, hwreg(HW_REG_MODE, 0, 4)31; GFX9-NEXT: s_lshl_b32 s6, s4, 232; GFX9-NEXT: s_mov_b32 s4, 0xeb24da7133; GFX9-NEXT: s_mov_b32 s5, 0xc96f38534; GFX9-NEXT: s_lshr_b64 s[4:5], s[4:5], s635; GFX9-NEXT: s_and_b32 s4, s4, 1536; GFX9-NEXT: s_add_i32 s5, s4, 437; GFX9-NEXT: s_cmp_lt_u32 s4, 438; GFX9-NEXT: s_cselect_b32 s4, s4, s539; GFX9-NEXT: v_mov_b32_e32 v0, s440; GFX9-NEXT: s_setpc_b64 s[30:31]41;42; GFX10-LABEL: func_rounding:43; GFX10: ; %bb.0:44; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)45; GFX10-NEXT: s_getreg_b32 s6, hwreg(HW_REG_MODE, 0, 4)46; GFX10-NEXT: s_mov_b32 s4, 0xeb24da7147; GFX10-NEXT: s_mov_b32 s5, 0xc96f38548; GFX10-NEXT: s_lshl_b32 s6, s6, 249; GFX10-NEXT: s_lshr_b64 s[4:5], s[4:5], s650; GFX10-NEXT: s_and_b32 s4, s4, 1551; GFX10-NEXT: s_add_i32 s5, s4, 452; GFX10-NEXT: s_cmp_lt_u32 s4, 453; GFX10-NEXT: s_cselect_b32 s4, s4, s554; GFX10-NEXT: v_mov_b32_e32 v0, s455; GFX10-NEXT: s_setpc_b64 s[30:31]56;57; GFX11-LABEL: func_rounding:58; GFX11: ; %bb.0:59; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)60; GFX11-NEXT: s_getreg_b32 s2, hwreg(HW_REG_MODE, 0, 4)61; GFX11-NEXT: s_mov_b32 s0, 0xeb24da7162; GFX11-NEXT: s_mov_b32 s1, 0xc96f38563; GFX11-NEXT: s_lshl_b32 s2, s2, 264; GFX11-NEXT: s_lshr_b64 s[0:1], s[0:1], s265; GFX11-NEXT: s_and_b32 s0, s0, 1566; GFX11-NEXT: s_add_i32 s1, s0, 467; GFX11-NEXT: s_cmp_lt_u32 s0, 468; GFX11-NEXT: s_cselect_b32 s0, s0, s169; GFX11-NEXT: v_mov_b32_e32 v0, s070; GFX11-NEXT: s_setpc_b64 s[30:31]71 %rounding = call i32 @llvm.get.rounding()72 ret i32 %rounding73}74;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:75; GCN: {{.*}}76; GFX1011: {{.*}}77; GFX6: {{.*}}78; GFX7: {{.*}}79; GFX8: {{.*}}80