diff options
Diffstat (limited to 'test/CodeGen/R600/mad_int24.ll')
| -rw-r--r-- | test/CodeGen/R600/mad_int24.ll | 17 |
1 files changed, 14 insertions, 3 deletions
diff --git a/test/CodeGen/R600/mad_int24.ll b/test/CodeGen/R600/mad_int24.ll index abb5290..c8dd377 100644 --- a/test/CodeGen/R600/mad_int24.ll +++ b/test/CodeGen/R600/mad_int24.ll @@ -2,14 +2,16 @@ ; RUN: llc < %s -march=r600 -mcpu=cayman | FileCheck %s --check-prefix=CM --check-prefix=FUNC ; RUN: llc < %s -march=r600 -mcpu=SI -verify-machineinstrs | FileCheck %s --check-prefix=SI --check-prefix=FUNC -; FUNC-LABEL: @i32_mad24 +declare i32 @llvm.AMDGPU.imul24(i32, i32) nounwind readnone + +; FUNC-LABEL: {{^}}i32_mad24: ; Signed 24-bit multiply is not supported on pre-Cayman GPUs. ; EG: MULLO_INT ; Make sure we aren't masking the inputs. ; CM-NOT: AND ; CM: MULADD_INT24 -; SI-NOT: AND -; SI: V_MAD_I32_I24 +; SI-NOT: and +; SI: v_mad_i32_i24 define void @i32_mad24(i32 addrspace(1)* %out, i32 %a, i32 %b, i32 %c) { entry: %0 = shl i32 %a, 8 @@ -21,3 +23,12 @@ entry: store i32 %3, i32 addrspace(1)* %out ret void } + +; FUNC-LABEL: @test_imul24 +; SI: v_mad_i32_i24 +define void @test_imul24(i32 addrspace(1)* %out, i32 %src0, i32 %src1, i32 %src2) nounwind { + %mul = call i32 @llvm.AMDGPU.imul24(i32 %src0, i32 %src1) nounwind readnone + %add = add i32 %mul, %src2 + store i32 %add, i32 addrspace(1)* %out, align 4 + ret void +} |