• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1; RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s --check-prefix=EG-CHECK
2; RUN: llc < %s -march=r600 -mcpu=cayman | FileCheck %s --check-prefix=EG-CHECK
3; RUN: llc < %s -march=r600 -mcpu=SI | FileCheck %s --check-prefix=SI-CHECK
4
5; EG-CHECK: @u32_mul24
6; EG-CHECK: MUL_UINT24 {{[* ]*}}T{{[0-9]\.[XYZW]}}, KC0[2].Z, KC0[2].W
7; SI-CHECK: @u32_mul24
8; SI-CHECK: V_MUL_U32_U24
9
10define void @u32_mul24(i32 addrspace(1)* %out, i32 %a, i32 %b) {
11entry:
12  %0 = shl i32 %a, 8
13  %a_24 = lshr i32 %0, 8
14  %1 = shl i32 %b, 8
15  %b_24 = lshr i32 %1, 8
16  %2 = mul i32 %a_24, %b_24
17  store i32 %2, i32 addrspace(1)* %out
18  ret void
19}
20
21; EG-CHECK: @i16_mul24
22; EG-CHECK-DAG: VTX_READ_16 [[A:T[0-9]\.X]], T{{[0-9]}}.X, 40
23; EG-CHECK-DAG: VTX_READ_16 [[B:T[0-9]\.X]], T{{[0-9]}}.X, 44
24; The order of A and B does not matter.
25; EG-CHECK: MUL_UINT24 {{[* ]*}}T{{[0-9]}}.[[MUL_CHAN:[XYZW]]], [[A]], [[B]]
26; The result must be sign-extended
27; EG-CHECK: LSHL {{[* ]*}}T{{[0-9]}}.[[LSHL_CHAN:[XYZW]]], PV.[[MUL_CHAN]], literal.x
28; EG-CHECK: 16
29; EG-CHECK: ASHR {{[* ]*}}T{{[0-9]\.[XYZW]}}, PV.[[LSHL_CHAN]], literal.x
30; EG-CHECK: 16
31; SI-CHECK: @i16_mul24
32; SI-CHECK: V_MUL_U32_U24_e{{(32|64)}} [[MUL:VGPR[0-9]]], {{[SV]GPR[0-9], [SV]GPR[0-9]}}
33; SI-CHECK: V_LSHLREV_B32_e32 [[LSHL:VGPR[0-9]]], 16, [[MUL]]
34; SI-CHECK: V_ASHRREV_I32_e32 VGPR{{[0-9]}}, 16, [[LSHL]]
35
36define void @i16_mul24(i32 addrspace(1)* %out, i16 %a, i16 %b) {
37entry:
38  %0 = mul i16 %a, %b
39  %1 = sext i16 %0 to i32
40  store i32 %1, i32 addrspace(1)* %out
41  ret void
42}
43
44; EG-CHECK: @i8_mul24
45; EG-CHECK-DAG: VTX_READ_8 [[A:T[0-9]\.X]], T{{[0-9]}}.X, 40
46; EG-CHECK-DAG: VTX_READ_8 [[B:T[0-9]\.X]], T{{[0-9]}}.X, 44
47; The order of A and B does not matter.
48; EG-CHECK: MUL_UINT24 {{[* ]*}}T{{[0-9]}}.[[MUL_CHAN:[XYZW]]], [[A]], [[B]]
49; The result must be sign-extended
50; EG-CHECK: LSHL {{[* ]*}}T{{[0-9]}}.[[LSHL_CHAN:[XYZW]]], PV.[[MUL_CHAN]], literal.x
51; EG-CHECK: 24
52; EG-CHECK: ASHR {{[* ]*}}T{{[0-9]\.[XYZW]}}, PV.[[LSHL_CHAN]], literal.x
53; EG-CHECK: 24
54; SI-CHECK: @i8_mul24
55; SI-CHECK: V_MUL_U32_U24_e{{(32|64)}} [[MUL:VGPR[0-9]]], {{[SV]GPR[0-9], [SV]GPR[0-9]}}
56; SI-CHECK: V_LSHLREV_B32_e32 [[LSHL:VGPR[0-9]]], 24, [[MUL]]
57; SI-CHECK: V_ASHRREV_I32_e32 VGPR{{[0-9]}}, 24, [[LSHL]]
58
59define void @i8_mul24(i32 addrspace(1)* %out, i8 %a, i8 %b) {
60entry:
61  %0 = mul i8 %a, %b
62  %1 = sext i8 %0 to i32
63  store i32 %1, i32 addrspace(1)* %out
64  ret void
65}
66