xref: /aosp_15_r20/external/llvm/test/CodeGen/AMDGPU/mul_int24.ll (revision 9880d6810fe72a1726cb53787c6711e909410d58)
1*9880d681SAndroid Build Coastguard Worker; RUN: llc < %s -march=amdgcn -verify-machineinstrs | FileCheck %s --check-prefix=SI --check-prefix=FUNC
2*9880d681SAndroid Build Coastguard Worker; RUN: llc < %s -march=amdgcn -mcpu=tonga -verify-machineinstrs | FileCheck %s --check-prefix=SI --check-prefix=FUNC
3*9880d681SAndroid Build Coastguard Worker; RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s --check-prefix=EG --check-prefix=FUNC
4*9880d681SAndroid Build Coastguard Worker; RUN: llc < %s -march=r600 -mcpu=cayman | FileCheck %s --check-prefix=CM --check-prefix=FUNC
5*9880d681SAndroid Build Coastguard Worker
6*9880d681SAndroid Build Coastguard Worker; FUNC-LABEL: {{^}}i32_mul24:
7*9880d681SAndroid Build Coastguard Worker; Signed 24-bit multiply is not supported on pre-Cayman GPUs.
8*9880d681SAndroid Build Coastguard Worker; EG: MULLO_INT
9*9880d681SAndroid Build Coastguard Worker; Make sure we are not masking the inputs
10*9880d681SAndroid Build Coastguard Worker; CM-NOT: AND
11*9880d681SAndroid Build Coastguard Worker; CM: MUL_INT24
12*9880d681SAndroid Build Coastguard Worker; SI-NOT: and
13*9880d681SAndroid Build Coastguard Worker; SI: v_mul_i32_i24
14*9880d681SAndroid Build Coastguard Workerdefine void @i32_mul24(i32 addrspace(1)* %out, i32 %a, i32 %b) {
15*9880d681SAndroid Build Coastguard Workerentry:
16*9880d681SAndroid Build Coastguard Worker  %0 = shl i32 %a, 8
17*9880d681SAndroid Build Coastguard Worker  %a_24 = ashr i32 %0, 8
18*9880d681SAndroid Build Coastguard Worker  %1 = shl i32 %b, 8
19*9880d681SAndroid Build Coastguard Worker  %b_24 = ashr i32 %1, 8
20*9880d681SAndroid Build Coastguard Worker  %2 = mul i32 %a_24, %b_24
21*9880d681SAndroid Build Coastguard Worker  store i32 %2, i32 addrspace(1)* %out
22*9880d681SAndroid Build Coastguard Worker  ret void
23*9880d681SAndroid Build Coastguard Worker}
24