switch (IntNo) {
default: return SDValue(); // Don't custom lower most intrinsics.
+ case Intrinsic::x86_avx2_permd:
+ case Intrinsic::x86_avx2_permps:
+ // Operands intentionally swapped. Mask is last operand to intrinsic,
+ // but second operand for node/instruction.
+ return DAG.getNode(X86ISD::VPERMV, dl, Op.getValueType(),
+ Op.getOperand(2), Op.getOperand(1));
+
case Intrinsic::x86_avx512_mask_valign_q_512:
case Intrinsic::x86_avx512_mask_valign_d_512:
// Vector source operands are swapped.
X86_INTRINSIC_DATA(avx2_packsswb, INTR_TYPE_2OP, X86ISD::PACKSS, 0),
X86_INTRINSIC_DATA(avx2_packusdw, INTR_TYPE_2OP, X86ISD::PACKUS, 0),
X86_INTRINSIC_DATA(avx2_packuswb, INTR_TYPE_2OP, X86ISD::PACKUS, 0),
- X86_INTRINSIC_DATA(avx2_permd, INTR_TYPE_2OP, X86ISD::VPERMV, 0),
- X86_INTRINSIC_DATA(avx2_permps, INTR_TYPE_2OP, X86ISD::VPERMV, 0),
X86_INTRINSIC_DATA(avx2_phadd_d, INTR_TYPE_2OP, X86ISD::HADD, 0),
X86_INTRINSIC_DATA(avx2_phadd_w, INTR_TYPE_2OP, X86ISD::HADD, 0),
X86_INTRINSIC_DATA(avx2_phsub_d, INTR_TYPE_2OP, X86ISD::HSUB, 0),
define <8 x i32> @test_x86_avx2_permd(<8 x i32> %a0, <8 x i32> %a1) {
- ; CHECK: vpermd
+ ; Check that the arguments are swapped between the intrinsic definition
+ ; and its lowering. Indeed, the offsets are the first source in
+ ; the instruction.
+ ; CHECK: vpermd %ymm0, %ymm1, %ymm0
%res = call <8 x i32> @llvm.x86.avx2.permd(<8 x i32> %a0, <8 x i32> %a1) ; <<8 x i32>> [#uses=1]
ret <8 x i32> %res
}
define <8 x float> @test_x86_avx2_permps(<8 x float> %a0, <8 x float> %a1) {
- ; CHECK: vpermps
+ ; Check that the arguments are swapped between the intrinsic definition
+ ; and its lowering. Indeed, the offsets are the first source in
+ ; the instruction.
+ ; CHECK: vpermps %ymm0, %ymm1, %ymm0
%res = call <8 x float> @llvm.x86.avx2.permps(<8 x float> %a0, <8 x float> %a1) ; <<8 x float>> [#uses=1]
ret <8 x float> %res
}
;CHECK-LABEL: stack_fold_permd
;CHECK: vpermd {{-?[0-9]*}}(%rsp), {{%ymm[0-9][0-9]*}}, {{%ymm[0-9][0-9]*}} {{.*#+}} 32-byte Folded Reload
%1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
- %2 = call <8 x i32> @llvm.x86.avx2.permd(<8 x i32> %a0, <8 x i32> %a1)
+ %2 = call <8 x i32> @llvm.x86.avx2.permd(<8 x i32> %a1, <8 x i32> %a0)
ret <8 x i32> %2
}
declare <8 x i32> @llvm.x86.avx2.permd(<8 x i32>, <8 x i32>) nounwind readonly
;CHECK-LABEL: stack_fold_permps
;CHECK: vpermps {{-?[0-9]*}}(%rsp), {{%ymm[0-9][0-9]*}}, {{%ymm[0-9][0-9]*}} {{.*#+}} 32-byte Folded Reload
%1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm2},~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
- %2 = call <8 x float> @llvm.x86.avx2.permps(<8 x float> %a0, <8 x float> %a1)
+ %2 = call <8 x float> @llvm.x86.avx2.permps(<8 x float> %a1, <8 x float> %a0)
ret <8 x float> %2
}
declare <8 x float> @llvm.x86.avx2.permps(<8 x float>, <8 x float>) nounwind readonly