1; RUN: llc -march=amdgcn -mcpu=verde -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s 2; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s 3; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s 4 5; FUNC-LABEL: {{^}}udiv_i32: 6; EG-NOT: SETGE_INT 7; EG: CF_END 8define void @udiv_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in) { 9 %b_ptr = getelementptr i32, i32 addrspace(1)* %in, i32 1 10 %a = load i32, i32 addrspace(1) * %in 11 %b = load i32, i32 addrspace(1) * %b_ptr 12 %result = udiv i32 %a, %b 13 store i32 %result, i32 addrspace(1)* %out 14 ret void 15} 16 17; FUNC-LABEL: {{^}}s_udiv_i32: 18 19define void @s_udiv_i32(i32 addrspace(1)* %out, i32 %a, i32 %b) { 20 %result = udiv i32 %a, %b 21 store i32 %result, i32 addrspace(1)* %out 22 ret void 23} 24 25 26; The code generated by udiv is long and complex and may frequently 27; change. The goal of this test is to make sure the ISel doesn't fail 28; when it gets a v4i32 udiv 29 30; FUNC-LABEL: {{^}}udiv_v2i32: 31; EG: CF_END 32 33; SI: s_endpgm 34define void @udiv_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> addrspace(1)* %in) { 35 %b_ptr = getelementptr <2 x i32>, <2 x i32> addrspace(1)* %in, i32 1 36 %a = load <2 x i32>, <2 x i32> addrspace(1) * %in 37 %b = load <2 x i32>, <2 x i32> addrspace(1) * %b_ptr 38 %result = udiv <2 x i32> %a, %b 39 store <2 x i32> %result, <2 x i32> addrspace(1)* %out 40 ret void 41} 42 43; FUNC-LABEL: {{^}}udiv_v4i32: 44; EG: CF_END 45; SI: s_endpgm 46define void @udiv_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %in) { 47 %b_ptr = getelementptr <4 x i32>, <4 x i32> addrspace(1)* %in, i32 1 48 %a = load <4 x i32>, <4 x i32> addrspace(1) * %in 49 %b = load <4 x i32>, <4 x i32> addrspace(1) * %b_ptr 50 %result = udiv <4 x i32> %a, %b 51 store <4 x i32> %result, <4 x i32> addrspace(1)* %out 52 ret void 53} 54 55; FUNC-LABEL: {{^}}udiv_i32_div_pow2: 56; SI: buffer_load_dword [[VAL:v[0-9]+]] 57; SI: v_lshrrev_b32_e32 [[RESULT:v[0-9]+]], 4, [[VAL]] 58; SI: buffer_store_dword [[RESULT]] 59define void @udiv_i32_div_pow2(i32 addrspace(1)* %out, i32 addrspace(1)* %in) { 60 %b_ptr = getelementptr i32, i32 addrspace(1)* %in, i32 1 61 %a = load i32, i32 addrspace(1)* %in 62 %result = udiv i32 %a, 16 63 store i32 %result, i32 addrspace(1)* %out 64 ret void 65} 66 67; FUNC-LABEL: {{^}}udiv_i32_div_k_even: 68; SI-DAG: buffer_load_dword [[VAL:v[0-9]+]] 69; SI-DAG: v_mov_b32_e32 [[K:v[0-9]+]], 0xfabbd9c1 70; SI: v_mul_hi_u32 [[MULHI:v[0-9]+]], [[K]], [[VAL]] 71; SI: v_lshrrev_b32_e32 [[RESULT:v[0-9]+]], 25, [[MULHI]] 72; SI: buffer_store_dword [[RESULT]] 73define void @udiv_i32_div_k_even(i32 addrspace(1)* %out, i32 addrspace(1)* %in) { 74 %b_ptr = getelementptr i32, i32 addrspace(1)* %in, i32 1 75 %a = load i32, i32 addrspace(1)* %in 76 %result = udiv i32 %a, 34259182 77 store i32 %result, i32 addrspace(1)* %out 78 ret void 79} 80 81; FUNC-LABEL: {{^}}udiv_i32_div_k_odd: 82; SI-DAG: buffer_load_dword [[VAL:v[0-9]+]] 83; SI-DAG: v_mov_b32_e32 [[K:v[0-9]+]], 0x7d5deca3 84; SI: v_mul_hi_u32 [[MULHI:v[0-9]+]], [[K]], [[VAL]] 85; SI: v_lshrrev_b32_e32 [[RESULT:v[0-9]+]], 24, [[MULHI]] 86; SI: buffer_store_dword [[RESULT]] 87define void @udiv_i32_div_k_odd(i32 addrspace(1)* %out, i32 addrspace(1)* %in) { 88 %b_ptr = getelementptr i32, i32 addrspace(1)* %in, i32 1 89 %a = load i32, i32 addrspace(1)* %in 90 %result = udiv i32 %a, 34259183 91 store i32 %result, i32 addrspace(1)* %out 92 ret void 93} 94 95; FUNC-LABEL: {{^}}v_udiv_i8: 96; SI: v_rcp_f32 97; SI: v_and_b32_e32 [[TRUNC:v[0-9]+]], 0xff, v{{[0-9]+}} 98; SI: buffer_store_dword [[TRUNC]] 99define void @v_udiv_i8(i32 addrspace(1)* %out, i8 addrspace(1)* %in) { 100 %den_ptr = getelementptr i8, i8 addrspace(1)* %in, i8 1 101 %num = load i8, i8 addrspace(1) * %in 102 %den = load i8, i8 addrspace(1) * %den_ptr 103 %result = udiv i8 %num, %den 104 %result.ext = zext i8 %result to i32 105 store i32 %result.ext, i32 addrspace(1)* %out 106 ret void 107} 108 109; FUNC-LABEL: {{^}}v_udiv_i16: 110; SI: v_rcp_f32 111; SI: v_and_b32_e32 [[TRUNC:v[0-9]+]], 0xffff, v{{[0-9]+}} 112; SI: buffer_store_dword [[TRUNC]] 113define void @v_udiv_i16(i32 addrspace(1)* %out, i16 addrspace(1)* %in) { 114 %den_ptr = getelementptr i16, i16 addrspace(1)* %in, i16 1 115 %num = load i16, i16 addrspace(1) * %in 116 %den = load i16, i16 addrspace(1) * %den_ptr 117 %result = udiv i16 %num, %den 118 %result.ext = zext i16 %result to i32 119 store i32 %result.ext, i32 addrspace(1)* %out 120 ret void 121} 122 123; FUNC-LABEL: {{^}}v_udiv_i23: 124; SI: v_rcp_f32 125; SI: v_and_b32_e32 [[TRUNC:v[0-9]+]], 0x7fffff, v{{[0-9]+}} 126; SI: buffer_store_dword [[TRUNC]] 127define void @v_udiv_i23(i32 addrspace(1)* %out, i23 addrspace(1)* %in) { 128 %den_ptr = getelementptr i23, i23 addrspace(1)* %in, i23 1 129 %num = load i23, i23 addrspace(1) * %in 130 %den = load i23, i23 addrspace(1) * %den_ptr 131 %result = udiv i23 %num, %den 132 %result.ext = zext i23 %result to i32 133 store i32 %result.ext, i32 addrspace(1)* %out 134 ret void 135} 136 137; FUNC-LABEL: {{^}}v_udiv_i24: 138; SI-NOT: v_rcp_f32 139define void @v_udiv_i24(i32 addrspace(1)* %out, i24 addrspace(1)* %in) { 140 %den_ptr = getelementptr i24, i24 addrspace(1)* %in, i24 1 141 %num = load i24, i24 addrspace(1) * %in 142 %den = load i24, i24 addrspace(1) * %den_ptr 143 %result = udiv i24 %num, %den 144 %result.ext = zext i24 %result to i32 145 store i32 %result.ext, i32 addrspace(1)* %out 146 ret void 147} 148