// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --version 6
// RUN: %clang_cc1 -O3 -triple powerpc64le-unknown-unknown -target-cpu pwr10 \
// RUN:  -emit-llvm %s -o - | FileCheck %s
// RUN: %clang_cc1 -O3 -triple powerpc64-unknown-unknown -target-cpu pwr10 \
// RUN: -emit-llvm %s -o - | FileCheck %s

// CHECK-LABEL: define dso_local void @test1(
// CHECK-SAME: ptr noundef readnone captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = tail call <512 x i1> @llvm.ppc.mma.assemble.acc(<16 x i8> [[VC]], <16 x i8> [[VC]], <16 x i8> [[VC]], <16 x i8> [[VC]])
// CHECK-NEXT:    store <512 x i1> [[TMP0]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6:![0-9]+]]
// CHECK-NEXT:    ret void
//
void test1(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __vector_quad res;
  __builtin_mma_assemble_acc(&res, vc, vc, vc, vc);
  *((__vector_quad *)resp) = res;
}

// CHECK-LABEL: define dso_local void @test2(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2:[0-9]+]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64
// CHECK-NEXT:    [[TMP1:%.*]] = tail call { <16 x i8>, <16 x i8>, <16 x i8>, <16 x i8> } @llvm.ppc.mma.disassemble.acc(<512 x i1> [[TMP0]])
// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <16 x i8>, <16 x i8>, <16 x i8>, <16 x i8> } [[TMP1]], 0
// CHECK-NEXT:    store <16 x i8> [[TMP2]], ptr [[RESP]], align 16
// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <16 x i8>, <16 x i8>, <16 x i8>, <16 x i8> } [[TMP1]], 1
// CHECK-NEXT:    [[TMP4:%.*]] = getelementptr inbounds nuw i8, ptr [[RESP]], i64 16
// CHECK-NEXT:    store <16 x i8> [[TMP3]], ptr [[TMP4]], align 16
// CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <16 x i8>, <16 x i8>, <16 x i8>, <16 x i8> } [[TMP1]], 2
// CHECK-NEXT:    [[TMP6:%.*]] = getelementptr inbounds nuw i8, ptr [[RESP]], i64 32
// CHECK-NEXT:    store <16 x i8> [[TMP5]], ptr [[TMP6]], align 16
// CHECK-NEXT:    [[TMP7:%.*]] = extractvalue { <16 x i8>, <16 x i8>, <16 x i8>, <16 x i8> } [[TMP1]], 3
// CHECK-NEXT:    [[TMP8:%.*]] = getelementptr inbounds nuw i8, ptr [[RESP]], i64 48
// CHECK-NEXT:    store <16 x i8> [[TMP7]], ptr [[TMP8]], align 16
// CHECK-NEXT:    ret void
//
void test2(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __builtin_mma_disassemble_acc(resp, (__vector_quad*)vqp);
}

// CHECK-LABEL: define dso_local void @test3(
// CHECK-SAME: ptr noundef readnone captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 32)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR0]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = tail call <256 x i1> @llvm.ppc.vsx.assemble.pair(<16 x i8> [[VC]], <16 x i8> [[VC]])
// CHECK-NEXT:    store <256 x i1> [[TMP0]], ptr [[RESP]], align 32, !tbaa [[__VECTOR_PAIR_TBAA8:![0-9]+]]
// CHECK-NEXT:    ret void
//
void test3(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __vector_pair res;
  __builtin_vsx_assemble_pair(&res, vc, vc);
  *((__vector_pair *)resp) = res;
}

// CHECK-LABEL: define dso_local void @test4(
// CHECK-SAME: ptr noundef readnone captures(none) [[VQP:%.*]], ptr noundef readonly captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 32)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <256 x i1>, ptr [[VPP]], align 32
// CHECK-NEXT:    [[TMP1:%.*]] = tail call { <16 x i8>, <16 x i8> } @llvm.ppc.vsx.disassemble.pair(<256 x i1> [[TMP0]])
// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <16 x i8>, <16 x i8> } [[TMP1]], 0
// CHECK-NEXT:    store <16 x i8> [[TMP2]], ptr [[RESP]], align 16
// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <16 x i8>, <16 x i8> } [[TMP1]], 1
// CHECK-NEXT:    [[TMP4:%.*]] = getelementptr inbounds nuw i8, ptr [[RESP]], i64 16
// CHECK-NEXT:    store <16 x i8> [[TMP3]], ptr [[TMP4]], align 16
// CHECK-NEXT:    ret void
//
void test4(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __builtin_vsx_disassemble_pair(resp, (__vector_pair*)vpp);
}

// CHECK-LABEL: define dso_local void @test5(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <512 x i1> @llvm.ppc.mma.xxmtacc(<512 x i1> [[TMP0]])
// CHECK-NEXT:    store <512 x i1> [[TMP1]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test5(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_xxmtacc(&vq);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test6(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <512 x i1> @llvm.ppc.mma.xxmfacc(<512 x i1> [[TMP0]])
// CHECK-NEXT:    store <512 x i1> [[TMP1]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test6(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_xxmfacc(&vq);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test7(
// CHECK-SAME: ptr noundef readnone captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR0]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = tail call <512 x i1> @llvm.ppc.mma.xxsetaccz()
// CHECK-NEXT:    store <512 x i1> [[TMP0]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test7(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_xxsetaccz(&vq);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test8(
// CHECK-SAME: ptr noundef readnone captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR0]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = tail call <512 x i1> @llvm.ppc.mma.xvi4ger8(<16 x i8> [[VC]], <16 x i8> [[VC]])
// CHECK-NEXT:    store <512 x i1> [[TMP0]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test8(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_xvi4ger8(&vq, vc, vc);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test9(
// CHECK-SAME: ptr noundef readnone captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR0]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = tail call <512 x i1> @llvm.ppc.mma.xvi8ger4(<16 x i8> [[VC]], <16 x i8> [[VC]])
// CHECK-NEXT:    store <512 x i1> [[TMP0]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test9(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_xvi8ger4(&vq, vc, vc);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test10(
// CHECK-SAME: ptr noundef readnone captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR0]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = tail call <512 x i1> @llvm.ppc.mma.xvi16ger2(<16 x i8> [[VC]], <16 x i8> [[VC]])
// CHECK-NEXT:    store <512 x i1> [[TMP0]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test10(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_xvi16ger2(&vq, vc, vc);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test11(
// CHECK-SAME: ptr noundef readnone captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR0]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = tail call <512 x i1> @llvm.ppc.mma.xvi16ger2s(<16 x i8> [[VC]], <16 x i8> [[VC]])
// CHECK-NEXT:    store <512 x i1> [[TMP0]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test11(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_xvi16ger2s(&vq, vc, vc);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test12(
// CHECK-SAME: ptr noundef readnone captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR0]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = tail call <512 x i1> @llvm.ppc.mma.xvf16ger2(<16 x i8> [[VC]], <16 x i8> [[VC]])
// CHECK-NEXT:    store <512 x i1> [[TMP0]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test12(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_xvf16ger2(&vq, vc, vc);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test13(
// CHECK-SAME: ptr noundef readnone captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR0]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = tail call <512 x i1> @llvm.ppc.mma.xvf32ger(<16 x i8> [[VC]], <16 x i8> [[VC]])
// CHECK-NEXT:    store <512 x i1> [[TMP0]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test13(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_xvf32ger(&vq, vc, vc);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test14(
// CHECK-SAME: ptr noundef readnone captures(none) [[VQP:%.*]], ptr noundef readonly captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <256 x i1>, ptr [[VPP]], align 32, !tbaa [[__VECTOR_PAIR_TBAA8]]
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <512 x i1> @llvm.ppc.mma.xvf64ger(<256 x i1> [[TMP0]], <16 x i8> [[VC]])
// CHECK-NEXT:    store <512 x i1> [[TMP1]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test14(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_xvf64ger(&vq, vp, vc);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test15(
// CHECK-SAME: ptr noundef readnone captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR0]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = tail call <512 x i1> @llvm.ppc.mma.pmxvi4ger8(<16 x i8> [[VC]], <16 x i8> [[VC]], i32 0, i32 0, i32 0)
// CHECK-NEXT:    store <512 x i1> [[TMP0]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test15(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_pmxvi4ger8(&vq, vc, vc, 0, 0, 0);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test16(
// CHECK-SAME: ptr noundef readnone captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR0]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = tail call <512 x i1> @llvm.ppc.mma.pmxvi8ger4(<16 x i8> [[VC]], <16 x i8> [[VC]], i32 0, i32 0, i32 0)
// CHECK-NEXT:    store <512 x i1> [[TMP0]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test16(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_pmxvi8ger4(&vq, vc, vc, 0, 0, 0);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test17(
// CHECK-SAME: ptr noundef readnone captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR0]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = tail call <512 x i1> @llvm.ppc.mma.pmxvi16ger2(<16 x i8> [[VC]], <16 x i8> [[VC]], i32 0, i32 0, i32 0)
// CHECK-NEXT:    store <512 x i1> [[TMP0]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test17(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_pmxvi16ger2(&vq, vc, vc, 0, 0, 0);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test18(
// CHECK-SAME: ptr noundef readnone captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR0]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = tail call <512 x i1> @llvm.ppc.mma.pmxvi16ger2s(<16 x i8> [[VC]], <16 x i8> [[VC]], i32 0, i32 0, i32 0)
// CHECK-NEXT:    store <512 x i1> [[TMP0]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test18(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_pmxvi16ger2s(&vq, vc, vc, 0, 0, 0);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test19(
// CHECK-SAME: ptr noundef readnone captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR0]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = tail call <512 x i1> @llvm.ppc.mma.pmxvf16ger2(<16 x i8> [[VC]], <16 x i8> [[VC]], i32 0, i32 0, i32 0)
// CHECK-NEXT:    store <512 x i1> [[TMP0]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test19(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_pmxvf16ger2(&vq, vc, vc, 0, 0, 0);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test20(
// CHECK-SAME: ptr noundef readnone captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR0]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = tail call <512 x i1> @llvm.ppc.mma.pmxvf32ger(<16 x i8> [[VC]], <16 x i8> [[VC]], i32 0, i32 0)
// CHECK-NEXT:    store <512 x i1> [[TMP0]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test20(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_pmxvf32ger(&vq, vc, vc, 0, 0);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test21(
// CHECK-SAME: ptr noundef readnone captures(none) [[VQP:%.*]], ptr noundef readonly captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <256 x i1>, ptr [[VPP]], align 32, !tbaa [[__VECTOR_PAIR_TBAA8]]
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <512 x i1> @llvm.ppc.mma.pmxvf64ger(<256 x i1> [[TMP0]], <16 x i8> [[VC]], i32 0, i32 0)
// CHECK-NEXT:    store <512 x i1> [[TMP1]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test21(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_pmxvf64ger(&vq, vp, vc, 0, 0);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test22(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <512 x i1> @llvm.ppc.mma.xvi4ger8pp(<512 x i1> [[TMP0]], <16 x i8> [[VC]], <16 x i8> [[VC]])
// CHECK-NEXT:    store <512 x i1> [[TMP1]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test22(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_xvi4ger8pp(&vq, vc, vc);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test23(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <512 x i1> @llvm.ppc.mma.xvi8ger4pp(<512 x i1> [[TMP0]], <16 x i8> [[VC]], <16 x i8> [[VC]])
// CHECK-NEXT:    store <512 x i1> [[TMP1]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test23(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_xvi8ger4pp(&vq, vc, vc);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test24(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <512 x i1> @llvm.ppc.mma.xvi8ger4spp(<512 x i1> [[TMP0]], <16 x i8> [[VC]], <16 x i8> [[VC]])
// CHECK-NEXT:    store <512 x i1> [[TMP1]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test24(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_xvi8ger4spp(&vq, vc, vc);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test25(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <512 x i1> @llvm.ppc.mma.xvi16ger2pp(<512 x i1> [[TMP0]], <16 x i8> [[VC]], <16 x i8> [[VC]])
// CHECK-NEXT:    store <512 x i1> [[TMP1]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test25(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_xvi16ger2pp(&vq, vc, vc);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test26(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <512 x i1> @llvm.ppc.mma.xvi16ger2spp(<512 x i1> [[TMP0]], <16 x i8> [[VC]], <16 x i8> [[VC]])
// CHECK-NEXT:    store <512 x i1> [[TMP1]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test26(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_xvi16ger2spp(&vq, vc, vc);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test27(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <512 x i1> @llvm.ppc.mma.pmxvi4ger8pp(<512 x i1> [[TMP0]], <16 x i8> [[VC]], <16 x i8> [[VC]], i32 0, i32 0, i32 0)
// CHECK-NEXT:    store <512 x i1> [[TMP1]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test27(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_pmxvi4ger8pp(&vq, vc, vc, 0, 0, 0);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test28(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <512 x i1> @llvm.ppc.mma.pmxvi8ger4pp(<512 x i1> [[TMP0]], <16 x i8> [[VC]], <16 x i8> [[VC]], i32 0, i32 0, i32 0)
// CHECK-NEXT:    store <512 x i1> [[TMP1]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test28(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_pmxvi8ger4pp(&vq, vc, vc, 0, 0, 0);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test29(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <512 x i1> @llvm.ppc.mma.pmxvi8ger4spp(<512 x i1> [[TMP0]], <16 x i8> [[VC]], <16 x i8> [[VC]], i32 0, i32 0, i32 0)
// CHECK-NEXT:    store <512 x i1> [[TMP1]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test29(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_pmxvi8ger4spp(&vq, vc, vc, 0, 0, 0);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test30(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <512 x i1> @llvm.ppc.mma.pmxvi16ger2pp(<512 x i1> [[TMP0]], <16 x i8> [[VC]], <16 x i8> [[VC]], i32 0, i32 0, i32 0)
// CHECK-NEXT:    store <512 x i1> [[TMP1]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test30(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_pmxvi16ger2pp(&vq, vc, vc, 0, 0, 0);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test31(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <512 x i1> @llvm.ppc.mma.pmxvi16ger2spp(<512 x i1> [[TMP0]], <16 x i8> [[VC]], <16 x i8> [[VC]], i32 0, i32 0, i32 0)
// CHECK-NEXT:    store <512 x i1> [[TMP1]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test31(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_pmxvi16ger2spp(&vq, vc, vc, 0, 0, 0);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test32(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <512 x i1> @llvm.ppc.mma.xvf16ger2pp(<512 x i1> [[TMP0]], <16 x i8> [[VC]], <16 x i8> [[VC]])
// CHECK-NEXT:    store <512 x i1> [[TMP1]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test32(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_xvf16ger2pp(&vq, vc, vc);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test33(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <512 x i1> @llvm.ppc.mma.xvf16ger2pn(<512 x i1> [[TMP0]], <16 x i8> [[VC]], <16 x i8> [[VC]])
// CHECK-NEXT:    store <512 x i1> [[TMP1]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test33(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_xvf16ger2pn(&vq, vc, vc);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test34(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <512 x i1> @llvm.ppc.mma.xvf16ger2np(<512 x i1> [[TMP0]], <16 x i8> [[VC]], <16 x i8> [[VC]])
// CHECK-NEXT:    store <512 x i1> [[TMP1]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test34(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_xvf16ger2np(&vq, vc, vc);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test35(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <512 x i1> @llvm.ppc.mma.xvf16ger2nn(<512 x i1> [[TMP0]], <16 x i8> [[VC]], <16 x i8> [[VC]])
// CHECK-NEXT:    store <512 x i1> [[TMP1]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test35(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_xvf16ger2nn(&vq, vc, vc);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test36(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <512 x i1> @llvm.ppc.mma.pmxvf16ger2pp(<512 x i1> [[TMP0]], <16 x i8> [[VC]], <16 x i8> [[VC]], i32 0, i32 0, i32 0)
// CHECK-NEXT:    store <512 x i1> [[TMP1]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test36(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_pmxvf16ger2pp(&vq, vc, vc, 0, 0, 0);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test37(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <512 x i1> @llvm.ppc.mma.pmxvf16ger2pn(<512 x i1> [[TMP0]], <16 x i8> [[VC]], <16 x i8> [[VC]], i32 0, i32 0, i32 0)
// CHECK-NEXT:    store <512 x i1> [[TMP1]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test37(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_pmxvf16ger2pn(&vq, vc, vc, 0, 0, 0);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test38(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <512 x i1> @llvm.ppc.mma.pmxvf16ger2np(<512 x i1> [[TMP0]], <16 x i8> [[VC]], <16 x i8> [[VC]], i32 0, i32 0, i32 0)
// CHECK-NEXT:    store <512 x i1> [[TMP1]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test38(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_pmxvf16ger2np(&vq, vc, vc, 0, 0, 0);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test39(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <512 x i1> @llvm.ppc.mma.pmxvf16ger2nn(<512 x i1> [[TMP0]], <16 x i8> [[VC]], <16 x i8> [[VC]], i32 0, i32 0, i32 0)
// CHECK-NEXT:    store <512 x i1> [[TMP1]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test39(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_pmxvf16ger2nn(&vq, vc, vc, 0, 0, 0);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test40(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <512 x i1> @llvm.ppc.mma.xvf32gerpp(<512 x i1> [[TMP0]], <16 x i8> [[VC]], <16 x i8> [[VC]])
// CHECK-NEXT:    store <512 x i1> [[TMP1]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test40(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_xvf32gerpp(&vq, vc, vc);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test41(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <512 x i1> @llvm.ppc.mma.xvf32gerpn(<512 x i1> [[TMP0]], <16 x i8> [[VC]], <16 x i8> [[VC]])
// CHECK-NEXT:    store <512 x i1> [[TMP1]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test41(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_xvf32gerpn(&vq, vc, vc);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test42(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <512 x i1> @llvm.ppc.mma.xvf32gernp(<512 x i1> [[TMP0]], <16 x i8> [[VC]], <16 x i8> [[VC]])
// CHECK-NEXT:    store <512 x i1> [[TMP1]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test42(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_xvf32gernp(&vq, vc, vc);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test43(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <512 x i1> @llvm.ppc.mma.xvf32gernn(<512 x i1> [[TMP0]], <16 x i8> [[VC]], <16 x i8> [[VC]])
// CHECK-NEXT:    store <512 x i1> [[TMP1]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test43(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_xvf32gernn(&vq, vc, vc);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test44(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <512 x i1> @llvm.ppc.mma.pmxvf32gerpp(<512 x i1> [[TMP0]], <16 x i8> [[VC]], <16 x i8> [[VC]], i32 0, i32 0)
// CHECK-NEXT:    store <512 x i1> [[TMP1]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test44(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_pmxvf32gerpp(&vq, vc, vc, 0, 0);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test45(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <512 x i1> @llvm.ppc.mma.pmxvf32gerpn(<512 x i1> [[TMP0]], <16 x i8> [[VC]], <16 x i8> [[VC]], i32 0, i32 0)
// CHECK-NEXT:    store <512 x i1> [[TMP1]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test45(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_pmxvf32gerpn(&vq, vc, vc, 0, 0);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test46(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <512 x i1> @llvm.ppc.mma.pmxvf32gernp(<512 x i1> [[TMP0]], <16 x i8> [[VC]], <16 x i8> [[VC]], i32 0, i32 0)
// CHECK-NEXT:    store <512 x i1> [[TMP1]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test46(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_pmxvf32gernp(&vq, vc, vc, 0, 0);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test47(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <512 x i1> @llvm.ppc.mma.pmxvf32gernn(<512 x i1> [[TMP0]], <16 x i8> [[VC]], <16 x i8> [[VC]], i32 0, i32 0)
// CHECK-NEXT:    store <512 x i1> [[TMP1]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test47(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_pmxvf32gernn(&vq, vc, vc, 0, 0);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test48(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readonly captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = load <256 x i1>, ptr [[VPP]], align 32, !tbaa [[__VECTOR_PAIR_TBAA8]]
// CHECK-NEXT:    [[TMP2:%.*]] = tail call <512 x i1> @llvm.ppc.mma.xvf64gerpp(<512 x i1> [[TMP0]], <256 x i1> [[TMP1]], <16 x i8> [[VC]])
// CHECK-NEXT:    store <512 x i1> [[TMP2]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test48(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_xvf64gerpp(&vq, vp, vc);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test49(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readonly captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = load <256 x i1>, ptr [[VPP]], align 32, !tbaa [[__VECTOR_PAIR_TBAA8]]
// CHECK-NEXT:    [[TMP2:%.*]] = tail call <512 x i1> @llvm.ppc.mma.xvf64gerpn(<512 x i1> [[TMP0]], <256 x i1> [[TMP1]], <16 x i8> [[VC]])
// CHECK-NEXT:    store <512 x i1> [[TMP2]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test49(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_xvf64gerpn(&vq, vp, vc);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test50(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readonly captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = load <256 x i1>, ptr [[VPP]], align 32, !tbaa [[__VECTOR_PAIR_TBAA8]]
// CHECK-NEXT:    [[TMP2:%.*]] = tail call <512 x i1> @llvm.ppc.mma.xvf64gernp(<512 x i1> [[TMP0]], <256 x i1> [[TMP1]], <16 x i8> [[VC]])
// CHECK-NEXT:    store <512 x i1> [[TMP2]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test50(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_xvf64gernp(&vq, vp, vc);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test51(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readonly captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = load <256 x i1>, ptr [[VPP]], align 32, !tbaa [[__VECTOR_PAIR_TBAA8]]
// CHECK-NEXT:    [[TMP2:%.*]] = tail call <512 x i1> @llvm.ppc.mma.xvf64gernn(<512 x i1> [[TMP0]], <256 x i1> [[TMP1]], <16 x i8> [[VC]])
// CHECK-NEXT:    store <512 x i1> [[TMP2]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test51(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_xvf64gernn(&vq, vp, vc);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test52(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readonly captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = load <256 x i1>, ptr [[VPP]], align 32, !tbaa [[__VECTOR_PAIR_TBAA8]]
// CHECK-NEXT:    [[TMP2:%.*]] = tail call <512 x i1> @llvm.ppc.mma.pmxvf64gerpp(<512 x i1> [[TMP0]], <256 x i1> [[TMP1]], <16 x i8> [[VC]], i32 0, i32 0)
// CHECK-NEXT:    store <512 x i1> [[TMP2]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test52(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_pmxvf64gerpp(&vq, vp, vc, 0, 0);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test53(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readonly captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = load <256 x i1>, ptr [[VPP]], align 32, !tbaa [[__VECTOR_PAIR_TBAA8]]
// CHECK-NEXT:    [[TMP2:%.*]] = tail call <512 x i1> @llvm.ppc.mma.pmxvf64gerpn(<512 x i1> [[TMP0]], <256 x i1> [[TMP1]], <16 x i8> [[VC]], i32 0, i32 0)
// CHECK-NEXT:    store <512 x i1> [[TMP2]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test53(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_pmxvf64gerpn(&vq, vp, vc, 0, 0);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test54(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readonly captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = load <256 x i1>, ptr [[VPP]], align 32, !tbaa [[__VECTOR_PAIR_TBAA8]]
// CHECK-NEXT:    [[TMP2:%.*]] = tail call <512 x i1> @llvm.ppc.mma.pmxvf64gernp(<512 x i1> [[TMP0]], <256 x i1> [[TMP1]], <16 x i8> [[VC]], i32 0, i32 0)
// CHECK-NEXT:    store <512 x i1> [[TMP2]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test54(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_pmxvf64gernp(&vq, vp, vc, 0, 0);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test55(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readonly captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = load <256 x i1>, ptr [[VPP]], align 32, !tbaa [[__VECTOR_PAIR_TBAA8]]
// CHECK-NEXT:    [[TMP2:%.*]] = tail call <512 x i1> @llvm.ppc.mma.pmxvf64gernn(<512 x i1> [[TMP0]], <256 x i1> [[TMP1]], <16 x i8> [[VC]], i32 0, i32 0)
// CHECK-NEXT:    store <512 x i1> [[TMP2]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test55(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_pmxvf64gernn(&vq, vp, vc, 0, 0);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test56(
// CHECK-SAME: ptr noundef readnone captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR0]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = tail call <512 x i1> @llvm.ppc.mma.xvbf16ger2(<16 x i8> [[VC]], <16 x i8> [[VC]])
// CHECK-NEXT:    store <512 x i1> [[TMP0]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test56(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_xvbf16ger2(&vq, vc, vc);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test57(
// CHECK-SAME: ptr noundef readnone captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR0]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = tail call <512 x i1> @llvm.ppc.mma.pmxvbf16ger2(<16 x i8> [[VC]], <16 x i8> [[VC]], i32 0, i32 0, i32 0)
// CHECK-NEXT:    store <512 x i1> [[TMP0]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test57(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_pmxvbf16ger2(&vq, vc, vc, 0, 0, 0);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test58(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <512 x i1> @llvm.ppc.mma.xvbf16ger2pp(<512 x i1> [[TMP0]], <16 x i8> [[VC]], <16 x i8> [[VC]])
// CHECK-NEXT:    store <512 x i1> [[TMP1]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test58(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_xvbf16ger2pp(&vq, vc, vc);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test59(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <512 x i1> @llvm.ppc.mma.xvbf16ger2pn(<512 x i1> [[TMP0]], <16 x i8> [[VC]], <16 x i8> [[VC]])
// CHECK-NEXT:    store <512 x i1> [[TMP1]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test59(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_xvbf16ger2pn(&vq, vc, vc);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test60(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <512 x i1> @llvm.ppc.mma.xvbf16ger2np(<512 x i1> [[TMP0]], <16 x i8> [[VC]], <16 x i8> [[VC]])
// CHECK-NEXT:    store <512 x i1> [[TMP1]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test60(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_xvbf16ger2np(&vq, vc, vc);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test61(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <512 x i1> @llvm.ppc.mma.xvbf16ger2nn(<512 x i1> [[TMP0]], <16 x i8> [[VC]], <16 x i8> [[VC]])
// CHECK-NEXT:    store <512 x i1> [[TMP1]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test61(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_xvbf16ger2nn(&vq, vc, vc);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test62(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <512 x i1> @llvm.ppc.mma.pmxvbf16ger2pp(<512 x i1> [[TMP0]], <16 x i8> [[VC]], <16 x i8> [[VC]], i32 0, i32 0, i32 0)
// CHECK-NEXT:    store <512 x i1> [[TMP1]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test62(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_pmxvbf16ger2pp(&vq, vc, vc, 0, 0, 0);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test63(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <512 x i1> @llvm.ppc.mma.pmxvbf16ger2pn(<512 x i1> [[TMP0]], <16 x i8> [[VC]], <16 x i8> [[VC]], i32 0, i32 0, i32 0)
// CHECK-NEXT:    store <512 x i1> [[TMP1]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test63(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_pmxvbf16ger2pn(&vq, vc, vc, 0, 0, 0);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test64(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <512 x i1> @llvm.ppc.mma.pmxvbf16ger2np(<512 x i1> [[TMP0]], <16 x i8> [[VC]], <16 x i8> [[VC]], i32 0, i32 0, i32 0)
// CHECK-NEXT:    store <512 x i1> [[TMP1]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test64(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_pmxvbf16ger2np(&vq, vc, vc, 0, 0, 0);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test65(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <512 x i1> @llvm.ppc.mma.pmxvbf16ger2nn(<512 x i1> [[TMP0]], <16 x i8> [[VC]], <16 x i8> [[VC]], i32 0, i32 0, i32 0)
// CHECK-NEXT:    store <512 x i1> [[TMP1]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test65(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __builtin_mma_pmxvbf16ger2nn(&vq, vc, vc, 0, 0, 0);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test66(
// CHECK-SAME: ptr noundef [[VPP:%.*]], ptr noundef [[VP2:%.*]]) local_unnamed_addr #[[ATTR3:[0-9]+]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = tail call <256 x i1> @llvm.ppc.vsx.lxvp(ptr [[VPP]])
// CHECK-NEXT:    tail call void @llvm.ppc.vsx.stxvp(<256 x i1> [[TMP0]], ptr [[VP2]])
// CHECK-NEXT:    ret void
//
void test66(const __vector_pair *vpp, __vector_pair *vp2) {
  __vector_pair vp = __builtin_vsx_lxvp(0L, vpp);
  __builtin_vsx_stxvp(vp, 0L, vp2);
}

// CHECK-LABEL: define dso_local void @test67(
// CHECK-SAME: ptr noundef [[VPP:%.*]], i64 noundef [[OFFSET:%.*]], ptr noundef [[VP2:%.*]]) local_unnamed_addr #[[ATTR3]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = getelementptr i8, ptr [[VPP]], i64 [[OFFSET]]
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <256 x i1> @llvm.ppc.vsx.lxvp(ptr [[TMP0]])
// CHECK-NEXT:    [[TMP2:%.*]] = getelementptr i8, ptr [[VP2]], i64 [[OFFSET]]
// CHECK-NEXT:    tail call void @llvm.ppc.vsx.stxvp(<256 x i1> [[TMP1]], ptr [[TMP2]])
// CHECK-NEXT:    ret void
//
void test67(const __vector_pair *vpp, signed long offset, __vector_pair *vp2) {
  __vector_pair vp = __builtin_vsx_lxvp(offset, vpp);
  __builtin_vsx_stxvp(vp, offset, vp2);
}

// CHECK-LABEL: define dso_local void @test68(
// CHECK-SAME: ptr noundef [[VPP:%.*]], ptr noundef [[VP2:%.*]]) local_unnamed_addr #[[ATTR3]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = getelementptr i8, ptr [[VPP]], i64 18
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <256 x i1> @llvm.ppc.vsx.lxvp(ptr [[TMP0]])
// CHECK-NEXT:    [[TMP2:%.*]] = getelementptr i8, ptr [[VP2]], i64 18
// CHECK-NEXT:    tail call void @llvm.ppc.vsx.stxvp(<256 x i1> [[TMP1]], ptr [[TMP2]])
// CHECK-NEXT:    ret void
//
void test68(const __vector_pair *vpp, __vector_pair *vp2) {
  __vector_pair vp = __builtin_vsx_lxvp(18L, vpp);
  __builtin_vsx_stxvp(vp, 18L, vp2);
}

// CHECK-LABEL: define dso_local void @test69(
// CHECK-SAME: ptr noundef [[VPP:%.*]], ptr noundef [[VP2:%.*]]) local_unnamed_addr #[[ATTR3]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = getelementptr i8, ptr [[VPP]], i64 1
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <256 x i1> @llvm.ppc.vsx.lxvp(ptr [[TMP0]])
// CHECK-NEXT:    [[TMP2:%.*]] = getelementptr i8, ptr [[VP2]], i64 1
// CHECK-NEXT:    tail call void @llvm.ppc.vsx.stxvp(<256 x i1> [[TMP1]], ptr [[TMP2]])
// CHECK-NEXT:    ret void
//
void test69(const __vector_pair *vpp, __vector_pair *vp2) {
  __vector_pair vp = __builtin_vsx_lxvp(1L, vpp);
  __builtin_vsx_stxvp(vp, 1L, vp2);
}

// CHECK-LABEL: define dso_local void @test70(
// CHECK-SAME: ptr noundef [[VPP:%.*]], ptr noundef [[VP2:%.*]]) local_unnamed_addr #[[ATTR3]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = getelementptr i8, ptr [[VPP]], i64 42
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <256 x i1> @llvm.ppc.vsx.lxvp(ptr [[TMP0]])
// CHECK-NEXT:    [[TMP2:%.*]] = getelementptr i8, ptr [[VP2]], i64 42
// CHECK-NEXT:    tail call void @llvm.ppc.vsx.stxvp(<256 x i1> [[TMP1]], ptr [[TMP2]])
// CHECK-NEXT:    ret void
//
void test70(const __vector_pair *vpp, __vector_pair *vp2) {
  __vector_pair vp = __builtin_vsx_lxvp(42L, vpp);
  __builtin_vsx_stxvp(vp, 42L, vp2);
}

// CHECK-LABEL: define dso_local void @test71(
// CHECK-SAME: ptr noundef [[VPP:%.*]], ptr noundef [[VP2:%.*]]) local_unnamed_addr #[[ATTR3]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = getelementptr i8, ptr [[VPP]], i64 32768
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <256 x i1> @llvm.ppc.vsx.lxvp(ptr [[TMP0]])
// CHECK-NEXT:    [[TMP2:%.*]] = getelementptr i8, ptr [[VP2]], i64 32768
// CHECK-NEXT:    tail call void @llvm.ppc.vsx.stxvp(<256 x i1> [[TMP1]], ptr [[TMP2]])
// CHECK-NEXT:    ret void
//
void test71(const __vector_pair *vpp, __vector_pair *vp2) {
  __vector_pair vp = __builtin_vsx_lxvp(32768L, vpp);
  __builtin_vsx_stxvp(vp, 32768L, vp2);
}

// CHECK-LABEL: define dso_local void @test72(
// CHECK-SAME: ptr noundef [[VPP:%.*]], ptr noundef [[VP2:%.*]]) local_unnamed_addr #[[ATTR3]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = getelementptr i8, ptr [[VPP]], i64 32799
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <256 x i1> @llvm.ppc.vsx.lxvp(ptr [[TMP0]])
// CHECK-NEXT:    [[TMP2:%.*]] = getelementptr i8, ptr [[VP2]], i64 32799
// CHECK-NEXT:    tail call void @llvm.ppc.vsx.stxvp(<256 x i1> [[TMP1]], ptr [[TMP2]])
// CHECK-NEXT:    ret void
//
void test72(const __vector_pair *vpp, __vector_pair *vp2) {
  __vector_pair vp = __builtin_vsx_lxvp(32799L, vpp);
  __builtin_vsx_stxvp(vp, 32799L, vp2);
}

// CHECK-LABEL: define dso_local void @test73(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = getelementptr i8, ptr [[VPP]], i64 8
// CHECK-NEXT:    [[TMP2:%.*]] = tail call <256 x i1> @llvm.ppc.vsx.lxvp(ptr [[TMP1]])
// CHECK-NEXT:    [[TMP3:%.*]] = tail call <512 x i1> @llvm.ppc.mma.pmxvf64gernn(<512 x i1> [[TMP0]], <256 x i1> [[TMP2]], <16 x i8> [[VC]], i32 0, i32 0)
// CHECK-NEXT:    store <512 x i1> [[TMP3]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test73(unsigned char *vqp, const __vector_pair *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = __builtin_vsx_lxvp(8L, vpp);
  __builtin_mma_pmxvf64gernn(&vq, vp, vc, 0, 0);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test74(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <256 x i1> @llvm.ppc.vsx.lxvp(ptr [[VPP]])
// CHECK-NEXT:    [[TMP2:%.*]] = tail call <512 x i1> @llvm.ppc.mma.xvf64gernp(<512 x i1> [[TMP0]], <256 x i1> [[TMP1]], <16 x i8> [[VC]])
// CHECK-NEXT:    store <512 x i1> [[TMP2]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test74(unsigned char *vqp, const __vector_pair *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = __builtin_vsx_lxvp(0L, vpp);
  __builtin_mma_xvf64gernp(&vq, vp, vc);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test75(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], i64 noundef [[OFFS:%.*]], ptr noundef [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = getelementptr i8, ptr [[VPP]], i64 [[OFFS]]
// CHECK-NEXT:    [[TMP2:%.*]] = tail call <256 x i1> @llvm.ppc.vsx.lxvp(ptr [[TMP1]])
// CHECK-NEXT:    [[TMP3:%.*]] = tail call <512 x i1> @llvm.ppc.mma.xvf64gernp(<512 x i1> [[TMP0]], <256 x i1> [[TMP2]], <16 x i8> [[VC]])
// CHECK-NEXT:    store <512 x i1> [[TMP3]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test75(unsigned char *vqp, signed long offs, const __vector_pair *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = __builtin_vsx_lxvp(offs, vpp);
  __builtin_mma_xvf64gernp(&vq, vp, vc);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test76(
// CHECK-SAME: ptr noundef readnone captures(none) [[VQP:%.*]], ptr noundef readnone captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 32)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR0]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = tail call <256 x i1> @llvm.ppc.vsx.assemble.pair(<16 x i8> [[VC]], <16 x i8> [[VC]])
// CHECK-NEXT:    store <256 x i1> [[TMP0]], ptr [[RESP]], align 32, !tbaa [[__VECTOR_PAIR_TBAA8]]
// CHECK-NEXT:    ret void
//
void test76(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = *((__vector_pair *)vpp);
  __vector_pair res;
  __builtin_mma_assemble_pair(&res, vc, vc);
  *((__vector_pair *)resp) = res;
}

// CHECK-LABEL: define dso_local void @test77(
// CHECK-SAME: ptr noundef readnone captures(none) [[VQP:%.*]], ptr noundef readonly captures(none) [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 32)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <256 x i1>, ptr [[VPP]], align 32
// CHECK-NEXT:    [[TMP1:%.*]] = tail call { <16 x i8>, <16 x i8> } @llvm.ppc.vsx.disassemble.pair(<256 x i1> [[TMP0]])
// CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <16 x i8>, <16 x i8> } [[TMP1]], 0
// CHECK-NEXT:    store <16 x i8> [[TMP2]], ptr [[RESP]], align 16
// CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <16 x i8>, <16 x i8> } [[TMP1]], 1
// CHECK-NEXT:    [[TMP4:%.*]] = getelementptr inbounds nuw i8, ptr [[RESP]], i64 16
// CHECK-NEXT:    store <16 x i8> [[TMP3]], ptr [[TMP4]], align 16
// CHECK-NEXT:    ret void
//
void test77(unsigned char *vqp, unsigned char *vpp, vector unsigned char vc, unsigned char *resp) {
  __builtin_mma_disassemble_pair(resp, (__vector_pair*)vpp);
}

// CHECK-LABEL: define dso_local void @test78(
// CHECK-SAME: ptr noundef [[VPP:%.*]], ptr noundef [[VP2:%.*]]) local_unnamed_addr #[[ATTR3]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = tail call <256 x i1> @llvm.ppc.vsx.lxvp(ptr [[VPP]])
// CHECK-NEXT:    tail call void @llvm.ppc.vsx.stxvp(<256 x i1> [[TMP0]], ptr [[VP2]])
// CHECK-NEXT:    ret void
//
void test78(const __vector_pair *vpp, __vector_pair *vp2) {
  __vector_pair vp = __builtin_mma_lxvp(0L, vpp);
  __builtin_mma_stxvp(vp, 0L, vp2);
}

// CHECK-LABEL: define dso_local void @test79(
// CHECK-SAME: ptr noundef [[VPP:%.*]], i64 noundef [[OFFSET:%.*]], ptr noundef [[VP2:%.*]]) local_unnamed_addr #[[ATTR3]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = getelementptr i8, ptr [[VPP]], i64 [[OFFSET]]
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <256 x i1> @llvm.ppc.vsx.lxvp(ptr [[TMP0]])
// CHECK-NEXT:    [[TMP2:%.*]] = getelementptr i8, ptr [[VP2]], i64 [[OFFSET]]
// CHECK-NEXT:    tail call void @llvm.ppc.vsx.stxvp(<256 x i1> [[TMP1]], ptr [[TMP2]])
// CHECK-NEXT:    ret void
//
void test79(const __vector_pair *vpp, signed long offset, __vector_pair *vp2) {
  __vector_pair vp = __builtin_mma_lxvp(offset, vpp);
  __builtin_mma_stxvp(vp, offset, vp2);
}

// CHECK-LABEL: define dso_local void @test80(
// CHECK-SAME: ptr noundef [[VPP:%.*]], ptr noundef [[VP2:%.*]]) local_unnamed_addr #[[ATTR3]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = getelementptr i8, ptr [[VPP]], i64 18
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <256 x i1> @llvm.ppc.vsx.lxvp(ptr [[TMP0]])
// CHECK-NEXT:    [[TMP2:%.*]] = getelementptr i8, ptr [[VP2]], i64 18
// CHECK-NEXT:    tail call void @llvm.ppc.vsx.stxvp(<256 x i1> [[TMP1]], ptr [[TMP2]])
// CHECK-NEXT:    ret void
//
void test80(const __vector_pair *vpp, __vector_pair *vp2) {
  __vector_pair vp = __builtin_mma_lxvp(18L, vpp);
  __builtin_mma_stxvp(vp, 18L, vp2);
}

// CHECK-LABEL: define dso_local void @test81(
// CHECK-SAME: ptr noundef [[VPP:%.*]], ptr noundef [[VP2:%.*]]) local_unnamed_addr #[[ATTR3]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = getelementptr i8, ptr [[VPP]], i64 1
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <256 x i1> @llvm.ppc.vsx.lxvp(ptr [[TMP0]])
// CHECK-NEXT:    [[TMP2:%.*]] = getelementptr i8, ptr [[VP2]], i64 1
// CHECK-NEXT:    tail call void @llvm.ppc.vsx.stxvp(<256 x i1> [[TMP1]], ptr [[TMP2]])
// CHECK-NEXT:    ret void
//
void test81(const __vector_pair *vpp, __vector_pair *vp2) {
  __vector_pair vp = __builtin_mma_lxvp(1L, vpp);
  __builtin_mma_stxvp(vp, 1L, vp2);
}

// CHECK-LABEL: define dso_local void @test82(
// CHECK-SAME: ptr noundef [[VPP:%.*]], ptr noundef [[VP2:%.*]]) local_unnamed_addr #[[ATTR3]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = getelementptr i8, ptr [[VPP]], i64 42
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <256 x i1> @llvm.ppc.vsx.lxvp(ptr [[TMP0]])
// CHECK-NEXT:    [[TMP2:%.*]] = getelementptr i8, ptr [[VP2]], i64 42
// CHECK-NEXT:    tail call void @llvm.ppc.vsx.stxvp(<256 x i1> [[TMP1]], ptr [[TMP2]])
// CHECK-NEXT:    ret void
//
void test82(const __vector_pair *vpp, __vector_pair *vp2) {
  __vector_pair vp = __builtin_mma_lxvp(42L, vpp);
  __builtin_mma_stxvp(vp, 42L, vp2);
}

// CHECK-LABEL: define dso_local void @test83(
// CHECK-SAME: ptr noundef [[VPP:%.*]], ptr noundef [[VP2:%.*]]) local_unnamed_addr #[[ATTR3]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = getelementptr i8, ptr [[VPP]], i64 32768
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <256 x i1> @llvm.ppc.vsx.lxvp(ptr [[TMP0]])
// CHECK-NEXT:    [[TMP2:%.*]] = getelementptr i8, ptr [[VP2]], i64 32768
// CHECK-NEXT:    tail call void @llvm.ppc.vsx.stxvp(<256 x i1> [[TMP1]], ptr [[TMP2]])
// CHECK-NEXT:    ret void
//
void test83(const __vector_pair *vpp, __vector_pair *vp2) {
  __vector_pair vp = __builtin_mma_lxvp(32768L, vpp);
  __builtin_mma_stxvp(vp, 32768L, vp2);
}

// CHECK-LABEL: define dso_local void @test84(
// CHECK-SAME: ptr noundef [[VPP:%.*]], ptr noundef [[VP2:%.*]]) local_unnamed_addr #[[ATTR3]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = getelementptr i8, ptr [[VPP]], i64 32799
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <256 x i1> @llvm.ppc.vsx.lxvp(ptr [[TMP0]])
// CHECK-NEXT:    [[TMP2:%.*]] = getelementptr i8, ptr [[VP2]], i64 32799
// CHECK-NEXT:    tail call void @llvm.ppc.vsx.stxvp(<256 x i1> [[TMP1]], ptr [[TMP2]])
// CHECK-NEXT:    ret void
//
void test84(const __vector_pair *vpp, __vector_pair *vp2) {
  __vector_pair vp = __builtin_mma_lxvp(32799L, vpp);
  __builtin_mma_stxvp(vp, 32799L, vp2);
}

// CHECK-LABEL: define dso_local void @test85(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = getelementptr i8, ptr [[VPP]], i64 8
// CHECK-NEXT:    [[TMP2:%.*]] = tail call <256 x i1> @llvm.ppc.vsx.lxvp(ptr [[TMP1]])
// CHECK-NEXT:    [[TMP3:%.*]] = tail call <512 x i1> @llvm.ppc.mma.pmxvf64gernn(<512 x i1> [[TMP0]], <256 x i1> [[TMP2]], <16 x i8> [[VC]], i32 0, i32 0)
// CHECK-NEXT:    store <512 x i1> [[TMP3]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test85(unsigned char *vqp, const __vector_pair *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = __builtin_mma_lxvp(8L, vpp);
  __builtin_mma_pmxvf64gernn(&vq, vp, vc, 0, 0);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test86(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], ptr noundef [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = tail call <256 x i1> @llvm.ppc.vsx.lxvp(ptr [[VPP]])
// CHECK-NEXT:    [[TMP2:%.*]] = tail call <512 x i1> @llvm.ppc.mma.xvf64gernp(<512 x i1> [[TMP0]], <256 x i1> [[TMP1]], <16 x i8> [[VC]])
// CHECK-NEXT:    store <512 x i1> [[TMP2]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test86(unsigned char *vqp, const __vector_pair *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = __builtin_mma_lxvp(0L, vpp);
  __builtin_mma_xvf64gernp(&vq, vp, vc);
  *((__vector_quad *)resp) = vq;
}

// CHECK-LABEL: define dso_local void @test87(
// CHECK-SAME: ptr noundef readonly captures(none) [[VQP:%.*]], i64 noundef [[OFFS:%.*]], ptr noundef [[VPP:%.*]], <16 x i8> noundef [[VC:%.*]], ptr noundef writeonly captures(none) initializes((0, 64)) [[RESP:%.*]]) local_unnamed_addr #[[ATTR2]] {
// CHECK-NEXT:  [[ENTRY:.*:]]
// CHECK-NEXT:    [[TMP0:%.*]] = load <512 x i1>, ptr [[VQP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    [[TMP1:%.*]] = getelementptr i8, ptr [[VPP]], i64 [[OFFS]]
// CHECK-NEXT:    [[TMP2:%.*]] = tail call <256 x i1> @llvm.ppc.vsx.lxvp(ptr [[TMP1]])
// CHECK-NEXT:    [[TMP3:%.*]] = tail call <512 x i1> @llvm.ppc.mma.xvf64gernp(<512 x i1> [[TMP0]], <256 x i1> [[TMP2]], <16 x i8> [[VC]])
// CHECK-NEXT:    store <512 x i1> [[TMP3]], ptr [[RESP]], align 64, !tbaa [[__VECTOR_QUAD_TBAA6]]
// CHECK-NEXT:    ret void
//
void test87(unsigned char *vqp, signed long offs, const __vector_pair *vpp, vector unsigned char vc, unsigned char *resp) {
  __vector_quad vq = *((__vector_quad *)vqp);
  __vector_pair vp = __builtin_mma_lxvp(offs, vpp);
  __builtin_mma_xvf64gernp(&vq, vp, vc);
  *((__vector_quad *)resp) = vq;
}
//.
// CHECK: [[META4:![0-9]+]] = !{!"omnipotent char", [[META5:![0-9]+]], i64 0}
// CHECK: [[META5]] = !{!"Simple C/C++ TBAA"}
// CHECK: [[__VECTOR_QUAD_TBAA6]] = !{[[META7:![0-9]+]], [[META7]], i64 0}
// CHECK: [[META7]] = !{!"__vector_quad", [[META4]], i64 0}
// CHECK: [[__VECTOR_PAIR_TBAA8]] = !{[[META9:![0-9]+]], [[META9]], i64 0}
// CHECK: [[META9]] = !{!"__vector_pair", [[META4]], i64 0}
//.