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
5 ; FUNC-LABEL: @u32_mad24
9 define void @u32_mad24(i32 addrspace(1)* %out, i32 %a, i32 %b, i32 %c) {
12 %a_24 = lshr i32 %0, 8
14 %b_24 = lshr i32 %1, 8
15 %2 = mul i32 %a_24, %b_24
17 store i32 %3, i32 addrspace(1)* %out
21 ; FUNC-LABEL: @i16_mad24
22 ; The order of A and B does not matter.
23 ; EG: MULADD_UINT24 {{[* ]*}}T{{[0-9]}}.[[MAD_CHAN:[XYZW]]]
24 ; The result must be sign-extended
25 ; EG: BFE_INT {{[* ]*}}T{{[0-9]\.[XYZW]}}, PV.[[MAD_CHAN]], 0.0, literal.x
27 ; SI: V_MAD_U32_U24 [[MAD:v[0-9]]], {{[sv][0-9], [sv][0-9]}}
28 ; SI: V_BFE_I32 v{{[0-9]}}, [[MAD]], 0, 16
30 define void @i16_mad24(i32 addrspace(1)* %out, i16 %a, i16 %b, i16 %c) {
34 %2 = sext i16 %1 to i32
35 store i32 %2, i32 addrspace(1)* %out
39 ; FUNC-LABEL: @i8_mad24
40 ; EG: MULADD_UINT24 {{[* ]*}}T{{[0-9]}}.[[MAD_CHAN:[XYZW]]]
41 ; The result must be sign-extended
42 ; EG: BFE_INT {{[* ]*}}T{{[0-9]\.[XYZW]}}, PV.[[MAD_CHAN]], 0.0, literal.x
44 ; SI: V_MAD_U32_U24 [[MUL:v[0-9]]], {{[sv][0-9], [sv][0-9]}}
45 ; SI: V_BFE_I32 v{{[0-9]}}, [[MUL]], 0, 8
47 define void @i8_mad24(i32 addrspace(1)* %out, i8 %a, i8 %b, i8 %c) {
51 %2 = sext i8 %1 to i32
52 store i32 %2, i32 addrspace(1)* %out
56 ; This tests for a bug where the mad_u24 pattern matcher would call
57 ; SimplifyDemandedBits on the first operand of the mul instruction
58 ; assuming that the pattern would be matched to a 24-bit mad. This
59 ; led to some instructions being incorrectly erased when the entire
60 ; 24-bit mad pattern wasn't being matched.
62 ; Check that the select instruction is not deleted.
63 ; FUNC-LABEL: @i24_i32_i32_mad
66 define void @i24_i32_i32_mad(i32 addrspace(1)* %out, i32 %a, i32 %b, i32 %c, i32 %d) {
69 %1 = icmp ne i32 %c, 0
70 %2 = select i1 %1, i32 %0, i32 34
73 store i32 %4, i32 addrspace(1)* %out