1 ; RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s --check-prefix=EG --check-prefix=FUNC 2 ; RUN: llc < %s -march=r600 -mcpu=cayman | FileCheck %s --check-prefix=EG --check-prefix=FUNC 3 ; RUN: llc < %s -march=r600 -mcpu=SI -verify-machineinstrs | FileCheck %s --check-prefix=SI --check-prefix=FUNC 4 5 ; FUNC-LABEL: @u32_mul24 6 ; EG: MUL_UINT24 {{[* ]*}}T{{[0-9]\.[XYZW]}}, KC0[2].Z, KC0[2].W 7 ; SI: V_MUL_U32_U24 8 9 define void @u32_mul24(i32 addrspace(1)* %out, i32 %a, i32 %b) { 10 entry: 11 %0 = shl i32 %a, 8 12 %a_24 = lshr i32 %0, 8 13 %1 = shl i32 %b, 8 14 %b_24 = lshr i32 %1, 8 15 %2 = mul i32 %a_24, %b_24 16 store i32 %2, i32 addrspace(1)* %out 17 ret void 18 } 19 20 ; FUNC-LABEL: @i16_mul24 21 ; EG: MUL_UINT24 {{[* ]*}}T{{[0-9]}}.[[MUL_CHAN:[XYZW]]] 22 ; The result must be sign-extended 23 ; EG: BFE_INT {{[* ]*}}T{{[0-9]}}.{{[XYZW]}}, PV.[[MUL_CHAN]], 0.0, literal.x 24 ; EG: 16 25 ; SI: V_MUL_U32_U24_e{{(32|64)}} [[MUL:v[0-9]]], {{[sv][0-9], [sv][0-9]}} 26 ; SI: V_BFE_I32 v{{[0-9]}}, [[MUL]], 0, 16, 27 define void @i16_mul24(i32 addrspace(1)* %out, i16 %a, i16 %b) { 28 entry: 29 %0 = mul i16 %a, %b 30 %1 = sext i16 %0 to i32 31 store i32 %1, i32 addrspace(1)* %out 32 ret void 33 } 34 35 ; FUNC-LABEL: @i8_mul24 36 ; EG: MUL_UINT24 {{[* ]*}}T{{[0-9]}}.[[MUL_CHAN:[XYZW]]] 37 ; The result must be sign-extended 38 ; EG: BFE_INT {{[* ]*}}T{{[0-9]}}.{{[XYZW]}}, PV.[[MUL_CHAN]], 0.0, literal.x 39 ; SI: V_MUL_U32_U24_e{{(32|64)}} [[MUL:v[0-9]]], {{[sv][0-9], [sv][0-9]}} 40 ; SI: V_BFE_I32 v{{[0-9]}}, [[MUL]], 0, 8, 41 42 define void @i8_mul24(i32 addrspace(1)* %out, i8 %a, i8 %b) { 43 entry: 44 %0 = mul i8 %a, %b 45 %1 = sext i8 %0 to i32 46 store i32 %1, i32 addrspace(1)* %out 47 ret void 48 } 49 50 ; Multiply with 24-bit inputs and 64-bit output 51 ; FUNC_LABEL: @mul24_i64 52 ; EG; MUL_UINT24 53 ; EG: MULHI 54 ; SI: V_MUL_U32_U24 55 ; FIXME: SI support 24-bit mulhi 56 ; SI: V_MUL_HI_U32 57 define void @mul24_i64(i64 addrspace(1)* %out, i64 %a, i64 %b) { 58 entry: 59 %0 = shl i64 %a, 40 60 %a_24 = lshr i64 %0, 40 61 %1 = shl i64 %b, 40 62 %b_24 = lshr i64 %1, 40 63 %2 = mul i64 %a_24, %b_24 64 store i64 %2, i64 addrspace(1)* %out 65 ret void 66 } 67
