Skip to content

Commit

Permalink
[X86][AVX512] small fix in ptestm intrinsics
Browse files Browse the repository at this point in the history 
move ptestm{q|d} intrinsics from patterns form (in td file) to the intrinsics table

Differential Revision: http://reviews.llvm.org/D16633



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@259029 91177308-0d34-0410-b5e6-96231b3b80d8
  • Loading branch information
Asaf Badouh committed Jan 28, 2016
1 parent f2fb000 commit 665e71a
Show file tree
Hide file tree
Showing 4 changed files with 24 additions and 17 deletions.
4 changes: 2 additions & 2 deletions include/llvm/IR/IntrinsicsX86.td
View file
Original file line number Diff line number Diff line change
Expand Up @@ -1848,10 +1848,10 @@ let TargetPrefix = "x86" in { // All intrinsics start with "llvm.x86.".
def int_x86_avx_ptestnzc_256 : GCCBuiltin<"__builtin_ia32_ptestnzc256">,
Intrinsic<[llvm_i32_ty], [llvm_v4i64_ty,
llvm_v4i64_ty], [IntrNoMem]>;
def int_x86_avx512_mask_ptestm_d_512 : GCCBuiltin<"__builtin_ia32_ptestmd512">,
def int_x86_avx512_ptestm_d_512 : GCCBuiltin<"__builtin_ia32_ptestmd512">,
Intrinsic<[llvm_i16_ty], [llvm_v16i32_ty, llvm_v16i32_ty,
llvm_i16_ty], [IntrNoMem]>;
def int_x86_avx512_mask_ptestm_q_512 : GCCBuiltin<"__builtin_ia32_ptestmq512">,
def int_x86_avx512_ptestm_q_512 : GCCBuiltin<"__builtin_ia32_ptestmq512">,
Intrinsic<[llvm_i8_ty], [llvm_v8i64_ty, llvm_v8i64_ty,
llvm_i8_ty], [IntrNoMem]>;

Expand Down
7 changes: 0 additions & 7 deletions lib/Target/X86/X86InstrAVX512.td
View file
Original file line number Diff line number Diff line change
Expand Up @@ -3908,13 +3908,6 @@ multiclass avx512_vptest_all_forms<bits<8> opc_wb, bits<8> opc_dq, string Opcode
defm VPTESTM : avx512_vptest_all_forms<0x26, 0x27, "vptestm", X86testm>, T8PD;
defm VPTESTNM : avx512_vptest_all_forms<0x26, 0x27, "vptestnm", X86testnm>, T8XS;

def : Pat <(i16 (int_x86_avx512_mask_ptestm_d_512 (v16i32 VR512:$src1),
(v16i32 VR512:$src2), (i16 -1))),
(COPY_TO_REGCLASS (VPTESTMDZrr VR512:$src1, VR512:$src2), GR16)>;

def : Pat <(i8 (int_x86_avx512_mask_ptestm_q_512 (v8i64 VR512:$src1),
(v8i64 VR512:$src2), (i8 -1))),
(COPY_TO_REGCLASS (VPTESTMQZrr VR512:$src1, VR512:$src2), GR8)>;

//===----------------------------------------------------------------------===//
// AVX-512 Shift instructions
Expand Down
2 changes: 2 additions & 0 deletions lib/Target/X86/X86IntrinsicsInfo.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -2053,8 +2053,10 @@ static const IntrinsicData IntrinsicsWithoutChain[] = {
X86_INTRINSIC_DATA(avx512_ptestm_b_512, CMP_MASK, X86ISD::TESTM, 0),
X86_INTRINSIC_DATA(avx512_ptestm_d_128, CMP_MASK, X86ISD::TESTM, 0),
X86_INTRINSIC_DATA(avx512_ptestm_d_256, CMP_MASK, X86ISD::TESTM, 0),
X86_INTRINSIC_DATA(avx512_ptestm_d_512, CMP_MASK, X86ISD::TESTM, 0),
X86_INTRINSIC_DATA(avx512_ptestm_q_128, CMP_MASK, X86ISD::TESTM, 0),
X86_INTRINSIC_DATA(avx512_ptestm_q_256, CMP_MASK, X86ISD::TESTM, 0),
X86_INTRINSIC_DATA(avx512_ptestm_q_512, CMP_MASK, X86ISD::TESTM, 0),
X86_INTRINSIC_DATA(avx512_ptestm_w_128, CMP_MASK, X86ISD::TESTM, 0),
X86_INTRINSIC_DATA(avx512_ptestm_w_256, CMP_MASK, X86ISD::TESTM, 0),
X86_INTRINSIC_DATA(avx512_ptestm_w_512, CMP_MASK, X86ISD::TESTM, 0),
Expand Down
28 changes: 20 additions & 8 deletions test/CodeGen/X86/avx512-intrinsics.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -836,27 +836,39 @@ define <8 x i64>@test_int_x86_avx512_mask_pabs_q_512(<8 x i64> %x0, <8 x i64> %x
ret <8 x i64> %res2
}

define i8 @test_vptestmq(<8 x i64> %a0, <8 x i64> %a1) {
define i8 @test_vptestmq(<8 x i64> %a0, <8 x i64> %a1, i8 %m) {
; CHECK-LABEL: test_vptestmq:
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vptestmq %zmm1, %zmm0, %k0 {%k1}
; CHECK-NEXT: kmovw %k0, %ecx
; CHECK-NEXT: vptestmq %zmm1, %zmm0, %k0
; CHECK-NEXT: kmovw %k0, %eax
; CHECK-NEXT: addb %cl, %al
; CHECK-NEXT: retq
%res = call i8 @llvm.x86.avx512.mask.ptestm.q.512(<8 x i64> %a0, <8 x i64> %a1, i8 -1)
ret i8 %res
%res = call i8 @llvm.x86.avx512.ptestm.q.512(<8 x i64> %a0, <8 x i64> %a1, i8 -1)
%res1 = call i8 @llvm.x86.avx512.ptestm.q.512(<8 x i64> %a0, <8 x i64> %a1, i8 %m)
%res2 = add i8 %res1, %res
ret i8 %res2
}
declare i8 @llvm.x86.avx512.mask.ptestm.q.512(<8 x i64>, <8 x i64>, i8)
declare i8 @llvm.x86.avx512.ptestm.q.512(<8 x i64>, <8 x i64>, i8)

define i16 @test_vptestmd(<16 x i32> %a0, <16 x i32> %a1) {
define i16 @test_vptestmd(<16 x i32> %a0, <16 x i32> %a1, i16 %m) {
; CHECK-LABEL: test_vptestmd:
; CHECK: ## BB#0:
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vptestmd %zmm1, %zmm0, %k0 {%k1}
; CHECK-NEXT: kmovw %k0, %ecx
; CHECK-NEXT: vptestmd %zmm1, %zmm0, %k0
; CHECK-NEXT: kmovw %k0, %eax
; CHECK-NEXT: addl %ecx, %eax
; CHECK-NEXT: retq
%res = call i16 @llvm.x86.avx512.mask.ptestm.d.512(<16 x i32> %a0, <16 x i32> %a1, i16 -1)
ret i16 %res
%res = call i16 @llvm.x86.avx512.ptestm.d.512(<16 x i32> %a0, <16 x i32> %a1, i16 -1)
%res1 = call i16 @llvm.x86.avx512.ptestm.d.512(<16 x i32> %a0, <16 x i32> %a1, i16 %m)
%res2 = add i16 %res1, %res
ret i16 %res2
}
declare i16 @llvm.x86.avx512.mask.ptestm.d.512(<16 x i32>, <16 x i32>, i16)
declare i16 @llvm.x86.avx512.ptestm.d.512(<16 x i32>, <16 x i32>, i16)

define void @test_store1(<16 x float> %data, i8* %ptr, i8* %ptr2, i16 %mask) {
; CHECK-LABEL: test_store1:
Expand Down

0 comments on commit 665e71a

Please sign in to comment.