1 ; RUN: llc -march=r600 -mcpu=SI < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s 2 ; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s 3 4 ; The code generated by sdiv is long and complex and may frequently change. 5 ; The goal of this test is to make sure the ISel doesn't fail. 6 ; 7 ; This program was previously failing to compile when one of the selectcc 8 ; opcodes generated by the sdiv lowering was being legalized and optimized to: 9 ; selectcc Remainder -1, 0, -1, SETGT 10 ; This was fixed by adding an additional pattern in R600Instructions.td to 11 ; match this pattern with a CNDGE_INT. 12 13 ; FUNC-LABEL: @sdiv_i32 14 ; EG: CF_END 15 define void @sdiv_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in) { 16 %den_ptr = getelementptr i32 addrspace(1)* %in, i32 1 17 %num = load i32 addrspace(1) * %in 18 %den = load i32 addrspace(1) * %den_ptr 19 %result = sdiv i32 %num, %den 20 store i32 %result, i32 addrspace(1)* %out 21 ret void 22 } 23 24 ; FUNC-LABEL: @sdiv_i32_4 25 define void @sdiv_i32_4(i32 addrspace(1)* %out, i32 addrspace(1)* %in) { 26 %num = load i32 addrspace(1) * %in 27 %result = sdiv i32 %num, 4 28 store i32 %result, i32 addrspace(1)* %out 29 ret void 30 } 31 32 ; Multiply by a weird constant to make sure setIntDivIsCheap is 33 ; working. 34 35 ; FUNC-LABEL: @slow_sdiv_i32_3435 36 ; SI: BUFFER_LOAD_DWORD [[VAL:v[0-9]+]], 37 ; SI: V_MOV_B32_e32 [[MAGIC:v[0-9]+]], 0x98a1930b 38 ; SI: V_MUL_HI_I32 [[TMP:v[0-9]+]], [[VAL]], [[MAGIC]] 39 ; SI: V_ADD_I32 40 ; SI: V_LSHRREV_B32 41 ; SI: V_ASHRREV_I32 42 ; SI: V_ADD_I32 43 ; SI: BUFFER_STORE_DWORD 44 ; SI: S_ENDPGM 45 define void @slow_sdiv_i32_3435(i32 addrspace(1)* %out, i32 addrspace(1)* %in) { 46 %num = load i32 addrspace(1) * %in 47 %result = sdiv i32 %num, 3435 48 store i32 %result, i32 addrspace(1)* %out 49 ret void 50 } 51 52 define void @sdiv_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> addrspace(1)* %in) { 53 %den_ptr = getelementptr <2 x i32> addrspace(1)* %in, i32 1 54 %num = load <2 x i32> addrspace(1) * %in 55 %den = load <2 x i32> addrspace(1) * %den_ptr 56 %result = sdiv <2 x i32> %num, %den 57 store <2 x i32> %result, <2 x i32> addrspace(1)* %out 58 ret void 59 } 60 61 define void @sdiv_v2i32_4(<2 x i32> addrspace(1)* %out, <2 x i32> addrspace(1)* %in) { 62 %num = load <2 x i32> addrspace(1) * %in 63 %result = sdiv <2 x i32> %num, <i32 4, i32 4> 64 store <2 x i32> %result, <2 x i32> addrspace(1)* %out 65 ret void 66 } 67 68 define void @sdiv_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %in) { 69 %den_ptr = getelementptr <4 x i32> addrspace(1)* %in, i32 1 70 %num = load <4 x i32> addrspace(1) * %in 71 %den = load <4 x i32> addrspace(1) * %den_ptr 72 %result = sdiv <4 x i32> %num, %den 73 store <4 x i32> %result, <4 x i32> addrspace(1)* %out 74 ret void 75 } 76 77 define void @sdiv_v4i32_4(<4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %in) { 78 %num = load <4 x i32> addrspace(1) * %in 79 %result = sdiv <4 x i32> %num, <i32 4, i32 4, i32 4, i32 4> 80 store <4 x i32> %result, <4 x i32> addrspace(1)* %out 81 ret void 82 } 83 84 ; Tests for 64-bit divide bypass. 85 ; define void @test_get_quotient(i64 addrspace(1)* %out, i64 %a, i64 %b) nounwind { 86 ; %result = sdiv i64 %a, %b 87 ; store i64 %result, i64 addrspace(1)* %out, align 8 88 ; ret void 89 ; } 90 91 ; define void @test_get_remainder(i64 addrspace(1)* %out, i64 %a, i64 %b) nounwind { 92 ; %result = srem i64 %a, %b 93 ; store i64 %result, i64 addrspace(1)* %out, align 8 94 ; ret void 95 ; } 96 97 ; define void @test_get_quotient_and_remainder(i64 addrspace(1)* %out, i64 %a, i64 %b) nounwind { 98 ; %resultdiv = sdiv i64 %a, %b 99 ; %resultrem = srem i64 %a, %b 100 ; %result = add i64 %resultdiv, %resultrem 101 ; store i64 %result, i64 addrspace(1)* %out, align 8 102 ; ret void 103 ; } 104
