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 -mattr=-flat-for-global -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 8 9; SI: v_rcp_iflag_f32_e32 10define void @udiv_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in) { 11 %b_ptr = getelementptr i32, i32 addrspace(1)* %in, i32 1 12 %a = load i32, i32 addrspace(1)* %in 13 %b = load i32, i32 addrspace(1)* %b_ptr 14 %result = udiv i32 %a, %b 15 store i32 %result, i32 addrspace(1)* %out 16 ret void 17} 18 19; FUNC-LABEL: {{^}}s_udiv_i32: 20; SI: v_rcp_iflag_f32_e32 21define void @s_udiv_i32(i32 addrspace(1)* %out, i32 %a, i32 %b) { 22 %result = udiv i32 %a, %b 23 store i32 %result, i32 addrspace(1)* %out 24 ret void 25} 26 27 28; The code generated by udiv is long and complex and may frequently 29; change. The goal of this test is to make sure the ISel doesn't fail 30; when it gets a v4i32 udiv 31 32; FUNC-LABEL: {{^}}udiv_v2i32: 33; EG: CF_END 34 35; SI: v_rcp_iflag_f32_e32 36; SI: v_rcp_iflag_f32_e32 37; SI: s_endpgm 38define void @udiv_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> addrspace(1)* %in) { 39 %b_ptr = getelementptr <2 x i32>, <2 x i32> addrspace(1)* %in, i32 1 40 %a = load <2 x i32>, <2 x i32> addrspace(1) * %in 41 %b = load <2 x i32>, <2 x i32> addrspace(1) * %b_ptr 42 %result = udiv <2 x i32> %a, %b 43 store <2 x i32> %result, <2 x i32> addrspace(1)* %out 44 ret void 45} 46 47; FUNC-LABEL: {{^}}udiv_v4i32: 48; EG: CF_END 49; SI: s_endpgm 50define void @udiv_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %in) { 51 %b_ptr = getelementptr <4 x i32>, <4 x i32> addrspace(1)* %in, i32 1 52 %a = load <4 x i32>, <4 x i32> addrspace(1) * %in 53 %b = load <4 x i32>, <4 x i32> addrspace(1) * %b_ptr 54 %result = udiv <4 x i32> %a, %b 55 store <4 x i32> %result, <4 x i32> addrspace(1)* %out 56 ret void 57} 58 59; FUNC-LABEL: {{^}}udiv_i32_div_pow2: 60; SI: buffer_load_dword [[VAL:v[0-9]+]] 61; SI: v_lshrrev_b32_e32 [[RESULT:v[0-9]+]], 4, [[VAL]] 62; SI: buffer_store_dword [[RESULT]] 63define void @udiv_i32_div_pow2(i32 addrspace(1)* %out, i32 addrspace(1)* %in) { 64 %b_ptr = getelementptr i32, i32 addrspace(1)* %in, i32 1 65 %a = load i32, i32 addrspace(1)* %in 66 %result = udiv i32 %a, 16 67 store i32 %result, i32 addrspace(1)* %out 68 ret void 69} 70 71; FUNC-LABEL: {{^}}udiv_i32_div_k_even: 72; SI-DAG: buffer_load_dword [[VAL:v[0-9]+]] 73; SI-DAG: v_mov_b32_e32 [[K:v[0-9]+]], 0xfabbd9c1 74; SI: v_mul_hi_u32 [[MULHI:v[0-9]+]], [[K]], [[VAL]] 75; SI: v_lshrrev_b32_e32 [[RESULT:v[0-9]+]], 25, [[MULHI]] 76; SI: buffer_store_dword [[RESULT]] 77define void @udiv_i32_div_k_even(i32 addrspace(1)* %out, i32 addrspace(1)* %in) { 78 %b_ptr = getelementptr i32, i32 addrspace(1)* %in, i32 1 79 %a = load i32, i32 addrspace(1)* %in 80 %result = udiv i32 %a, 34259182 81 store i32 %result, i32 addrspace(1)* %out 82 ret void 83} 84 85; FUNC-LABEL: {{^}}udiv_i32_div_k_odd: 86; SI-DAG: buffer_load_dword [[VAL:v[0-9]+]] 87; SI-DAG: v_mov_b32_e32 [[K:v[0-9]+]], 0x7d5deca3 88; SI: v_mul_hi_u32 [[MULHI:v[0-9]+]], [[K]], [[VAL]] 89; SI: v_lshrrev_b32_e32 [[RESULT:v[0-9]+]], 24, [[MULHI]] 90; SI: buffer_store_dword [[RESULT]] 91define void @udiv_i32_div_k_odd(i32 addrspace(1)* %out, i32 addrspace(1)* %in) { 92 %b_ptr = getelementptr i32, i32 addrspace(1)* %in, i32 1 93 %a = load i32, i32 addrspace(1)* %in 94 %result = udiv i32 %a, 34259183 95 store i32 %result, i32 addrspace(1)* %out 96 ret void 97} 98 99; FUNC-LABEL: {{^}}v_udiv_i8: 100; SI: v_rcp_f32 101; SI: v_and_b32_e32 [[TRUNC:v[0-9]+]], 0xff, v{{[0-9]+}} 102; SI: buffer_store_dword [[TRUNC]] 103define void @v_udiv_i8(i32 addrspace(1)* %out, i8 addrspace(1)* %in) { 104 %den_ptr = getelementptr i8, i8 addrspace(1)* %in, i8 1 105 %num = load i8, i8 addrspace(1) * %in 106 %den = load i8, i8 addrspace(1) * %den_ptr 107 %result = udiv i8 %num, %den 108 %result.ext = zext i8 %result to i32 109 store i32 %result.ext, i32 addrspace(1)* %out 110 ret void 111} 112 113; FUNC-LABEL: {{^}}v_udiv_i16: 114; SI: v_rcp_f32 115; SI: v_and_b32_e32 [[TRUNC:v[0-9]+]], 0xffff, v{{[0-9]+}} 116; SI: buffer_store_dword [[TRUNC]] 117define void @v_udiv_i16(i32 addrspace(1)* %out, i16 addrspace(1)* %in) { 118 %den_ptr = getelementptr i16, i16 addrspace(1)* %in, i16 1 119 %num = load i16, i16 addrspace(1) * %in 120 %den = load i16, i16 addrspace(1) * %den_ptr 121 %result = udiv i16 %num, %den 122 %result.ext = zext i16 %result to i32 123 store i32 %result.ext, i32 addrspace(1)* %out 124 ret void 125} 126 127; FUNC-LABEL: {{^}}v_udiv_i23: 128; SI: v_rcp_f32 129; SI: v_and_b32_e32 [[TRUNC:v[0-9]+]], 0x7fffff, v{{[0-9]+}} 130; SI: buffer_store_dword [[TRUNC]] 131define void @v_udiv_i23(i32 addrspace(1)* %out, i23 addrspace(1)* %in) { 132 %den_ptr = getelementptr i23, i23 addrspace(1)* %in, i23 1 133 %num = load i23, i23 addrspace(1) * %in 134 %den = load i23, i23 addrspace(1) * %den_ptr 135 %result = udiv i23 %num, %den 136 %result.ext = zext i23 %result to i32 137 store i32 %result.ext, i32 addrspace(1)* %out 138 ret void 139} 140 141; FUNC-LABEL: {{^}}v_udiv_i24: 142; SI-NOT: v_rcp_f32 143define void @v_udiv_i24(i32 addrspace(1)* %out, i24 addrspace(1)* %in) { 144 %den_ptr = getelementptr i24, i24 addrspace(1)* %in, i24 1 145 %num = load i24, i24 addrspace(1) * %in 146 %den = load i24, i24 addrspace(1) * %den_ptr 147 %result = udiv i24 %num, %den 148 %result.ext = zext i24 %result to i32 149 store i32 %result.ext, i32 addrspace(1)* %out 150 ret void 151} 152 153; FUNC-LABEL: @scalarize_mulhu_4xi32 154; SI: v_mul_hi_u32 155; SI: v_mul_hi_u32 156; SI: v_mul_hi_u32 157; SI: v_mul_hi_u32 158 159define void @scalarize_mulhu_4xi32(<4 x i32> addrspace(1)* nocapture readonly %in, <4 x i32> addrspace(1)* nocapture %out) { 160 %1 = load <4 x i32>, <4 x i32> addrspace(1)* %in, align 16 161 %2 = udiv <4 x i32> %1, <i32 53668, i32 53668, i32 53668, i32 53668> 162 store <4 x i32> %2, <4 x i32> addrspace(1)* %out, align 16 163 ret void 164} 165 166; FUNC-LABEL: {{^}}test_udiv2: 167; SI: s_lshr_b32 s{{[0-9]}}, s{{[0-9]}}, 1 168define void @test_udiv2(i32 %p) { 169 %i = udiv i32 %p, 2 170 store volatile i32 %i, i32 addrspace(1)* undef 171 ret void 172} 173 174; FUNC-LABEL: {{^}}test_udiv_3_mulhu: 175; SI: v_mov_b32_e32 v{{[0-9]+}}, 0xaaaaaaab 176; SI: v_mul_hi_u32 v0, {{v[0-9]+}}, {{s[0-9]+}} 177; SI-NEXT: v_lshrrev_b32_e32 v0, 1, v0 178define void @test_udiv_3_mulhu(i32 %p) { 179 %i = udiv i32 %p, 3 180 store volatile i32 %i, i32 addrspace(1)* undef 181 ret void 182} 183