[AVX-512] Add unmasked intrinsics for variable shifts of dwords and qwords.

These will be used to replace the masked intrinsics so that InstCombineCalls can optimize the AVX-512 variable shifts the same way it does for AVX2.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@286754 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/include/llvm/IR/IntrinsicsX86.td b/include/llvm/IR/IntrinsicsX86.td
index d3d607431177690b605b8f85d46bcca0a7bfa11d..d52f8dc3007451f91ee8a2739b50ba3914bf64cc 100644 (file)
--- a/include/llvm/IR/IntrinsicsX86.td
+++ b/include/llvm/IR/IntrinsicsX86.td
@@ -2311,6 +2311,13 @@ let TargetPrefix = "x86" in {  // All intrinsics start with "llvm.x86.".
               Intrinsic<[llvm_v4i64_ty], [llvm_v4i64_ty, llvm_v4i64_ty],
                         [IntrNoMem]>;
 
+  def int_x86_avx512_psllv_d_512 : GCCBuiltin<"__builtin_ia32_psllv16si">,
+              Intrinsic<[llvm_v16i32_ty], [llvm_v16i32_ty, llvm_v16i32_ty],
+                        [IntrNoMem]>;
+  def int_x86_avx512_psllv_q_512 : GCCBuiltin<"__builtin_ia32_psllv8di">,
+              Intrinsic<[llvm_v8i64_ty], [llvm_v8i64_ty, llvm_v8i64_ty],
+                        [IntrNoMem]>;
+
   def int_x86_avx2_psrlv_d : GCCBuiltin<"__builtin_ia32_psrlv4si">,
               Intrinsic<[llvm_v4i32_ty], [llvm_v4i32_ty, llvm_v4i32_ty],
                         [IntrNoMem]>;
@@ -2324,6 +2331,13 @@ let TargetPrefix = "x86" in {  // All intrinsics start with "llvm.x86.".
               Intrinsic<[llvm_v4i64_ty], [llvm_v4i64_ty, llvm_v4i64_ty],
                         [IntrNoMem]>;
 
+  def int_x86_avx512_psrlv_d_512 : GCCBuiltin<"__builtin_ia32_psrlv16si">,
+              Intrinsic<[llvm_v16i32_ty], [llvm_v16i32_ty, llvm_v16i32_ty],
+                        [IntrNoMem]>;
+  def int_x86_avx512_psrlv_q_512 : GCCBuiltin<"__builtin_ia32_psrlv8di">,
+              Intrinsic<[llvm_v8i64_ty], [llvm_v8i64_ty, llvm_v8i64_ty],
+                        [IntrNoMem]>;
+
   def int_x86_avx2_psrav_d : GCCBuiltin<"__builtin_ia32_psrav4si">,
               Intrinsic<[llvm_v4i32_ty], [llvm_v4i32_ty, llvm_v4i32_ty],
                         [IntrNoMem]>;
@@ -2331,6 +2345,19 @@ let TargetPrefix = "x86" in {  // All intrinsics start with "llvm.x86.".
               Intrinsic<[llvm_v8i32_ty], [llvm_v8i32_ty, llvm_v8i32_ty],
                         [IntrNoMem]>;
 
+  def int_x86_avx512_psrav_d_512 : GCCBuiltin<"__builtin_ia32_psrav16si">,
+              Intrinsic<[llvm_v16i32_ty], [llvm_v16i32_ty, llvm_v16i32_ty],
+                        [IntrNoMem]>;
+  def int_x86_avx512_psrav_q_128 : GCCBuiltin<"__builtin_ia32_psravq128">,
+              Intrinsic<[llvm_v2i64_ty], [llvm_v2i64_ty, llvm_v2i64_ty],
+                        [IntrNoMem]>;
+  def int_x86_avx512_psrav_q_256 : GCCBuiltin<"__builtin_ia32_psravq256">,
+              Intrinsic<[llvm_v4i64_ty], [llvm_v4i64_ty, llvm_v4i64_ty],
+                        [IntrNoMem]>;
+  def int_x86_avx512_psrav_q_512 : GCCBuiltin<"__builtin_ia32_psrav8di">,
+              Intrinsic<[llvm_v8i64_ty], [llvm_v8i64_ty, llvm_v8i64_ty],
+                        [IntrNoMem]>;
+
   def int_x86_avx512_mask_psllv_d : GCCBuiltin<"__builtin_ia32_psllv16si_mask">,
               Intrinsic<[llvm_v16i32_ty], [llvm_v16i32_ty,
                          llvm_v16i32_ty, llvm_v16i32_ty, llvm_i16_ty],

diff --git a/lib/Target/X86/X86IntrinsicsInfo.h b/lib/Target/X86/X86IntrinsicsInfo.h
index b450dc3fdafb753b80b4ec997fc5a7022e19db94..4e52d7fcdcbb309b1b12d88bc15531ee88223b4a 100644 (file)
--- a/lib/Target/X86/X86IntrinsicsInfo.h
+++ b/lib/Target/X86/X86IntrinsicsInfo.h
@@ -1515,6 +1515,8 @@ static const IntrinsicData  IntrinsicsWithoutChain[] = {
   X86_INTRINSIC_DATA(avx512_pslli_d_512, VSHIFT, X86ISD::VSHLI, 0),
   X86_INTRINSIC_DATA(avx512_pslli_q_512, VSHIFT, X86ISD::VSHLI, 0),
   X86_INTRINSIC_DATA(avx512_pslli_w_512, VSHIFT, X86ISD::VSHLI, 0),
+  X86_INTRINSIC_DATA(avx512_psllv_d_512, INTR_TYPE_2OP, ISD::SHL, 0),
+  X86_INTRINSIC_DATA(avx512_psllv_q_512, INTR_TYPE_2OP, ISD::SHL, 0),
   X86_INTRINSIC_DATA(avx512_psra_d_512, INTR_TYPE_2OP, X86ISD::VSRA, 0),
   X86_INTRINSIC_DATA(avx512_psra_q_128, INTR_TYPE_2OP, X86ISD::VSRA, 0),
   X86_INTRINSIC_DATA(avx512_psra_q_256, INTR_TYPE_2OP, X86ISD::VSRA, 0),
@@ -1525,12 +1527,18 @@ static const IntrinsicData  IntrinsicsWithoutChain[] = {
   X86_INTRINSIC_DATA(avx512_psrai_q_256, VSHIFT, X86ISD::VSRAI, 0),
   X86_INTRINSIC_DATA(avx512_psrai_q_512, VSHIFT, X86ISD::VSRAI, 0),
   X86_INTRINSIC_DATA(avx512_psrai_w_512, VSHIFT, X86ISD::VSRAI, 0),
+  X86_INTRINSIC_DATA(avx512_psrav_d_512, INTR_TYPE_2OP, X86ISD::VSRAV, 0),
+  X86_INTRINSIC_DATA(avx512_psrav_q_128, INTR_TYPE_2OP, X86ISD::VSRAV, 0),
+  X86_INTRINSIC_DATA(avx512_psrav_q_256, INTR_TYPE_2OP, X86ISD::VSRAV, 0),
+  X86_INTRINSIC_DATA(avx512_psrav_q_512, INTR_TYPE_2OP, X86ISD::VSRAV, 0),
   X86_INTRINSIC_DATA(avx512_psrl_d_512, INTR_TYPE_2OP, X86ISD::VSRL, 0),
   X86_INTRINSIC_DATA(avx512_psrl_q_512, INTR_TYPE_2OP, X86ISD::VSRL, 0),
   X86_INTRINSIC_DATA(avx512_psrl_w_512, INTR_TYPE_2OP, X86ISD::VSRL, 0),
   X86_INTRINSIC_DATA(avx512_psrli_d_512, VSHIFT, X86ISD::VSRLI, 0),
   X86_INTRINSIC_DATA(avx512_psrli_q_512, VSHIFT, X86ISD::VSRLI, 0),
   X86_INTRINSIC_DATA(avx512_psrli_w_512, VSHIFT, X86ISD::VSRLI, 0),
+  X86_INTRINSIC_DATA(avx512_psrlv_d_512, INTR_TYPE_2OP, ISD::SRL, 0),
+  X86_INTRINSIC_DATA(avx512_psrlv_q_512, INTR_TYPE_2OP, ISD::SRL, 0),
   X86_INTRINSIC_DATA(avx512_ptestm_b_128, CMP_MASK, X86ISD::TESTM, 0),
   X86_INTRINSIC_DATA(avx512_ptestm_b_256, CMP_MASK, X86ISD::TESTM, 0),
   X86_INTRINSIC_DATA(avx512_ptestm_b_512, CMP_MASK, X86ISD::TESTM, 0),

diff --git a/test/CodeGen/X86/avx512-intrinsics.ll b/test/CodeGen/X86/avx512-intrinsics.ll
index 4828158be900d3ec3038e6bc9b575eff56b4cd04..0a787986d93e7e1f3abea5fdbcf78091bc37a273 100644 (file)
--- a/test/CodeGen/X86/avx512-intrinsics.ll
+++ b/test/CodeGen/X86/avx512-intrinsics.ll
@@ -5742,3 +5742,219 @@ define <8 x i64> @test_x86_avx512_maskz_psrli_q_512(<8 x i64> %a0, i8 %mask) {
 }
 declare <8 x i64> @llvm.x86.avx512.psrli.q.512(<8 x i64>, i32) nounwind readnone
 
+define <16 x i32> @test_x86_avx512_psllv_d_512(<16 x i32> %a0, <16 x i32> %a1) {
+; CHECK-LABEL: test_x86_avx512_psllv_d_512:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    vpsllvd %zmm1, %zmm0, %zmm0
+; CHECK-NEXT:    retq
+  %res = call <16 x i32> @llvm.x86.avx512.psllv.d.512(<16 x i32> %a0, <16 x i32> %a1)
+  ret <16 x i32> %res
+}
+
+define <16 x i32> @test_x86_avx512_mask_psllv_d_512(<16 x i32> %a0, <16 x i32> %a1, <16 x i32> %a2, i16 %mask) {
+; CHECK-LABEL: test_x86_avx512_mask_psllv_d_512:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    kmovw %edi, %k1
+; CHECK-NEXT:    vpsllvd %zmm1, %zmm0, %zmm2 {%k1}
+; CHECK-NEXT:    vmovdqa64 %zmm2, %zmm0
+; CHECK-NEXT:    retq
+  %res = call <16 x i32> @llvm.x86.avx512.psllv.d.512(<16 x i32> %a0, <16 x i32> %a1)
+  %mask.cast = bitcast i16 %mask to <16 x i1>
+  %res2 = select <16 x i1> %mask.cast, <16 x i32> %res, <16 x i32> %a2
+  ret <16 x i32> %res2
+}
+
+define <16 x i32> @test_x86_avx512_maskz_psllv_d_512(<16 x i32> %a0, <16 x i32> %a1, i16 %mask) {
+; CHECK-LABEL: test_x86_avx512_maskz_psllv_d_512:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    kmovw %edi, %k1
+; CHECK-NEXT:    vpsllvd %zmm1, %zmm0, %zmm0 {%k1} {z}
+; CHECK-NEXT:    retq
+  %res = call <16 x i32> @llvm.x86.avx512.psllv.d.512(<16 x i32> %a0, <16 x i32> %a1)
+  %mask.cast = bitcast i16 %mask to <16 x i1>
+  %res2 = select <16 x i1> %mask.cast, <16 x i32> %res, <16 x i32> zeroinitializer
+  ret <16 x i32> %res2
+}
+
+declare <16 x i32> @llvm.x86.avx512.psllv.d.512(<16 x i32>, <16 x i32>) nounwind readnone
+
+define <8 x i64> @test_x86_avx512_psllv_q_512(<8 x i64> %a0, <8 x i64> %a1) {
+; CHECK-LABEL: test_x86_avx512_psllv_q_512:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    vpsllvq %zmm1, %zmm0, %zmm0
+; CHECK-NEXT:    retq
+  %res = call <8 x i64> @llvm.x86.avx512.psllv.q.512(<8 x i64> %a0, <8 x i64> %a1)
+  ret <8 x i64> %res
+}
+
+define <8 x i64> @test_x86_avx512_mask_psllv_q_512(<8 x i64> %a0, <8 x i64> %a1, <8 x i64> %a2, i8 %mask) {
+; CHECK-LABEL: test_x86_avx512_mask_psllv_q_512:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    kmovw %edi, %k1
+; CHECK-NEXT:    vpsllvq %zmm1, %zmm0, %zmm2 {%k1}
+; CHECK-NEXT:    vmovdqa64 %zmm2, %zmm0
+; CHECK-NEXT:    retq
+  %res = call <8 x i64> @llvm.x86.avx512.psllv.q.512(<8 x i64> %a0, <8 x i64> %a1)
+  %mask.cast = bitcast i8 %mask to <8 x i1>
+  %res2 = select <8 x i1> %mask.cast, <8 x i64> %res, <8 x i64> %a2
+  ret <8 x i64> %res2
+}
+
+define <8 x i64> @test_x86_avx512_maskz_psllv_q_512(<8 x i64> %a0, <8 x i64> %a1, i8 %mask) {
+; CHECK-LABEL: test_x86_avx512_maskz_psllv_q_512:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    kmovw %edi, %k1
+; CHECK-NEXT:    vpsllvq %zmm1, %zmm0, %zmm0 {%k1} {z}
+; CHECK-NEXT:    retq
+  %res = call <8 x i64> @llvm.x86.avx512.psllv.q.512(<8 x i64> %a0, <8 x i64> %a1)
+  %mask.cast = bitcast i8 %mask to <8 x i1>
+  %res2 = select <8 x i1> %mask.cast, <8 x i64> %res, <8 x i64> zeroinitializer
+  ret <8 x i64> %res2
+}
+
+declare <8 x i64> @llvm.x86.avx512.psllv.q.512(<8 x i64>, <8 x i64>) nounwind readnone
+
+define <16 x i32> @test_x86_avx512_psrav_d_512(<16 x i32> %a0, <16 x i32> %a1) {
+; CHECK-LABEL: test_x86_avx512_psrav_d_512:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    vpsravd %zmm1, %zmm0, %zmm0
+; CHECK-NEXT:    retq
+  %res = call <16 x i32> @llvm.x86.avx512.psrav.d.512(<16 x i32> %a0, <16 x i32> %a1)
+  ret <16 x i32> %res
+}
+
+define <16 x i32> @test_x86_avx512_mask_psrav_d_512(<16 x i32> %a0, <16 x i32> %a1, <16 x i32> %a2, i16 %mask) {
+; CHECK-LABEL: test_x86_avx512_mask_psrav_d_512:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    kmovw %edi, %k1
+; CHECK-NEXT:    vpsravd %zmm1, %zmm0, %zmm2 {%k1}
+; CHECK-NEXT:    vmovdqa64 %zmm2, %zmm0
+; CHECK-NEXT:    retq
+  %res = call <16 x i32> @llvm.x86.avx512.psrav.d.512(<16 x i32> %a0, <16 x i32> %a1)
+  %mask.cast = bitcast i16 %mask to <16 x i1>
+  %res2 = select <16 x i1> %mask.cast, <16 x i32> %res, <16 x i32> %a2
+  ret <16 x i32> %res2
+}
+
+define <16 x i32> @test_x86_avx512_maskz_psrav_d_512(<16 x i32> %a0, <16 x i32> %a1, i16 %mask) {
+; CHECK-LABEL: test_x86_avx512_maskz_psrav_d_512:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    kmovw %edi, %k1
+; CHECK-NEXT:    vpsravd %zmm1, %zmm0, %zmm0 {%k1} {z}
+; CHECK-NEXT:    retq
+  %res = call <16 x i32> @llvm.x86.avx512.psrav.d.512(<16 x i32> %a0, <16 x i32> %a1)
+  %mask.cast = bitcast i16 %mask to <16 x i1>
+  %res2 = select <16 x i1> %mask.cast, <16 x i32> %res, <16 x i32> zeroinitializer
+  ret <16 x i32> %res2
+}
+
+declare <16 x i32> @llvm.x86.avx512.psrav.d.512(<16 x i32>, <16 x i32>) nounwind readnone
+
+define <8 x i64> @test_x86_avx512_psrav_q_512(<8 x i64> %a0, <8 x i64> %a1) {
+; CHECK-LABEL: test_x86_avx512_psrav_q_512:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    vpsravq %zmm1, %zmm0, %zmm0
+; CHECK-NEXT:    retq
+  %res = call <8 x i64> @llvm.x86.avx512.psrav.q.512(<8 x i64> %a0, <8 x i64> %a1)
+  ret <8 x i64> %res
+}
+
+define <8 x i64> @test_x86_avx512_mask_psrav_q_512(<8 x i64> %a0, <8 x i64> %a1, <8 x i64> %a2, i8 %mask) {
+; CHECK-LABEL: test_x86_avx512_mask_psrav_q_512:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    kmovw %edi, %k1
+; CHECK-NEXT:    vpsravq %zmm1, %zmm0, %zmm2 {%k1}
+; CHECK-NEXT:    vmovdqa64 %zmm2, %zmm0
+; CHECK-NEXT:    retq
+  %res = call <8 x i64> @llvm.x86.avx512.psrav.q.512(<8 x i64> %a0, <8 x i64> %a1)
+  %mask.cast = bitcast i8 %mask to <8 x i1>
+  %res2 = select <8 x i1> %mask.cast, <8 x i64> %res, <8 x i64> %a2
+  ret <8 x i64> %res2
+}
+
+define <8 x i64> @test_x86_avx512_maskz_psrav_q_512(<8 x i64> %a0, <8 x i64> %a1, i8 %mask) {
+; CHECK-LABEL: test_x86_avx512_maskz_psrav_q_512:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    kmovw %edi, %k1
+; CHECK-NEXT:    vpsravq %zmm1, %zmm0, %zmm0 {%k1} {z}
+; CHECK-NEXT:    retq
+  %res = call <8 x i64> @llvm.x86.avx512.psrav.q.512(<8 x i64> %a0, <8 x i64> %a1)
+  %mask.cast = bitcast i8 %mask to <8 x i1>
+  %res2 = select <8 x i1> %mask.cast, <8 x i64> %res, <8 x i64> zeroinitializer
+  ret <8 x i64> %res2
+}
+
+declare <8 x i64> @llvm.x86.avx512.psrav.q.512(<8 x i64>, <8 x i64>) nounwind readnone
+
+define <16 x i32> @test_x86_avx512_psrlv_d_512(<16 x i32> %a0, <16 x i32> %a1) {
+; CHECK-LABEL: test_x86_avx512_psrlv_d_512:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    vpsrlvd %zmm1, %zmm0, %zmm0
+; CHECK-NEXT:    retq
+  %res = call <16 x i32> @llvm.x86.avx512.psrlv.d.512(<16 x i32> %a0, <16 x i32> %a1)
+  ret <16 x i32> %res
+}
+
+define <16 x i32> @test_x86_avx512_mask_psrlv_d_512(<16 x i32> %a0, <16 x i32> %a1, <16 x i32> %a2, i16 %mask) {
+; CHECK-LABEL: test_x86_avx512_mask_psrlv_d_512:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    kmovw %edi, %k1
+; CHECK-NEXT:    vpsrlvd %zmm1, %zmm0, %zmm2 {%k1}
+; CHECK-NEXT:    vmovdqa64 %zmm2, %zmm0
+; CHECK-NEXT:    retq
+  %res = call <16 x i32> @llvm.x86.avx512.psrlv.d.512(<16 x i32> %a0, <16 x i32> %a1)
+  %mask.cast = bitcast i16 %mask to <16 x i1>
+  %res2 = select <16 x i1> %mask.cast, <16 x i32> %res, <16 x i32> %a2
+  ret <16 x i32> %res2
+}
+
+define <16 x i32> @test_x86_avx512_maskz_psrlv_d_512(<16 x i32> %a0, <16 x i32> %a1, i16 %mask) {
+; CHECK-LABEL: test_x86_avx512_maskz_psrlv_d_512:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    kmovw %edi, %k1
+; CHECK-NEXT:    vpsrlvd %zmm1, %zmm0, %zmm0 {%k1} {z}
+; CHECK-NEXT:    retq
+  %res = call <16 x i32> @llvm.x86.avx512.psrlv.d.512(<16 x i32> %a0, <16 x i32> %a1)
+  %mask.cast = bitcast i16 %mask to <16 x i1>
+  %res2 = select <16 x i1> %mask.cast, <16 x i32> %res, <16 x i32> zeroinitializer
+  ret <16 x i32> %res2
+}
+
+declare <16 x i32> @llvm.x86.avx512.psrlv.d.512(<16 x i32>, <16 x i32>) nounwind readnone
+
+define <8 x i64> @test_x86_avx512_psrlv_q_512(<8 x i64> %a0, <8 x i64> %a1) {
+; CHECK-LABEL: test_x86_avx512_psrlv_q_512:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    vpsrlvq %zmm1, %zmm0, %zmm0
+; CHECK-NEXT:    retq
+  %res = call <8 x i64> @llvm.x86.avx512.psrlv.q.512(<8 x i64> %a0, <8 x i64> %a1)
+  ret <8 x i64> %res
+}
+
+define <8 x i64> @test_x86_avx512_mask_psrlv_q_512(<8 x i64> %a0, <8 x i64> %a1, <8 x i64> %a2, i8 %mask) {
+; CHECK-LABEL: test_x86_avx512_mask_psrlv_q_512:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    kmovw %edi, %k1
+; CHECK-NEXT:    vpsrlvq %zmm1, %zmm0, %zmm2 {%k1}
+; CHECK-NEXT:    vmovdqa64 %zmm2, %zmm0
+; CHECK-NEXT:    retq
+  %res = call <8 x i64> @llvm.x86.avx512.psrlv.q.512(<8 x i64> %a0, <8 x i64> %a1)
+  %mask.cast = bitcast i8 %mask to <8 x i1>
+  %res2 = select <8 x i1> %mask.cast, <8 x i64> %res, <8 x i64> %a2
+  ret <8 x i64> %res2
+}
+
+define <8 x i64> @test_x86_avx512_maskz_psrlv_q_512(<8 x i64> %a0, <8 x i64> %a1, i8 %mask) {
+; CHECK-LABEL: test_x86_avx512_maskz_psrlv_q_512:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    kmovw %edi, %k1
+; CHECK-NEXT:    vpsrlvq %zmm1, %zmm0, %zmm0 {%k1} {z}
+; CHECK-NEXT:    retq
+  %res = call <8 x i64> @llvm.x86.avx512.psrlv.q.512(<8 x i64> %a0, <8 x i64> %a1)
+  %mask.cast = bitcast i8 %mask to <8 x i1>
+  %res2 = select <8 x i1> %mask.cast, <8 x i64> %res, <8 x i64> zeroinitializer
+  ret <8 x i64> %res2
+}
+
+declare <8 x i64> @llvm.x86.avx512.psrlv.q.512(<8 x i64>, <8 x i64>) nounwind readnone
+

diff --git a/test/CodeGen/X86/avx512vl-intrinsics.ll b/test/CodeGen/X86/avx512vl-intrinsics.ll
index ae2f8dbdae4986b1010c0e978fa8184118b81572..d882b7bc7d2a8be6bc02ac50c414e7a5ac104216 100644 (file)
--- a/test/CodeGen/X86/avx512vl-intrinsics.ll
+++ b/test/CodeGen/X86/avx512vl-intrinsics.ll
@@ -5421,3 +5421,79 @@ define <4 x i64> @test_x86_avx512_maskz_psrai_q_256(<4 x i64> %a0, i8 %mask) {
   ret <4 x i64> %res2
 }
 declare <4 x i64> @llvm.x86.avx512.psrai.q.256(<4 x i64>, i32) nounwind readnone
+
+define <2 x i64> @test_x86_avx512_psrav_q_128(<2 x i64> %a0, <2 x i64> %a1) {
+; CHECK-LABEL: test_x86_avx512_psrav_q_128:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    vpsravq %xmm1, %xmm0, %xmm0 ## encoding: [0x62,0xf2,0xfd,0x08,0x46,0xc1]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %res = call <2 x i64> @llvm.x86.avx512.psrav.q.128(<2 x i64> %a0, <2 x i64> %a1)
+  ret <2 x i64> %res
+}
+
+define <2 x i64> @test_x86_avx512_mask_psrav_q_128(<2 x i64> %a0, <2 x i64> %a1, <2 x i64> %a2, i8 %mask) {
+; CHECK-LABEL: test_x86_avx512_mask_psrav_q_128:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vpsravq %xmm1, %xmm0, %xmm2 {%k1} ## encoding: [0x62,0xf2,0xfd,0x09,0x46,0xd1]
+; CHECK-NEXT:    vmovdqa64 %xmm2, %xmm0 ## encoding: [0x62,0xf1,0xfd,0x08,0x6f,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %res = call <2 x i64> @llvm.x86.avx512.psrav.q.128(<2 x i64> %a0, <2 x i64> %a1)
+  %mask.cast = bitcast i8 %mask to <8 x i1>
+  %mask.extract = shufflevector <8 x i1> %mask.cast, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %res2 = select <2 x i1> %mask.extract, <2 x i64> %res, <2 x i64> %a2
+  ret <2 x i64> %res2
+}
+
+define <2 x i64> @test_x86_avx512_maskz_psrav_q_128(<2 x i64> %a0, <2 x i64> %a1, i8 %mask) {
+; CHECK-LABEL: test_x86_avx512_maskz_psrav_q_128:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vpsravq %xmm1, %xmm0, %xmm0 {%k1} {z} ## encoding: [0x62,0xf2,0xfd,0x89,0x46,0xc1]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %res = call <2 x i64> @llvm.x86.avx512.psrav.q.128(<2 x i64> %a0, <2 x i64> %a1)
+  %mask.cast = bitcast i8 %mask to <8 x i1>
+  %mask.extract = shufflevector <8 x i1> %mask.cast, <8 x i1> undef, <2 x i32> <i32 0, i32 1>
+  %res2 = select <2 x i1> %mask.extract, <2 x i64> %res, <2 x i64> zeroinitializer
+  ret <2 x i64> %res2
+}
+
+declare <2 x i64> @llvm.x86.avx512.psrav.q.128(<2 x i64>, <2 x i64>) nounwind readnone
+
+define <4 x i64> @test_x86_avx512_psrav_q_256(<4 x i64> %a0, <4 x i64> %a1) {
+; CHECK-LABEL: test_x86_avx512_psrav_q_256:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    vpsravq %ymm1, %ymm0, %ymm0 ## encoding: [0x62,0xf2,0xfd,0x28,0x46,0xc1]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %res = call <4 x i64> @llvm.x86.avx512.psrav.q.256(<4 x i64> %a0, <4 x i64> %a1)
+  ret <4 x i64> %res
+}
+
+define <4 x i64> @test_x86_avx512_mask_psrav_q_256(<4 x i64> %a0, <4 x i64> %a1, <4 x i64> %a2, i8 %mask) {
+; CHECK-LABEL: test_x86_avx512_mask_psrav_q_256:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vpsravq %ymm1, %ymm0, %ymm2 {%k1} ## encoding: [0x62,0xf2,0xfd,0x29,0x46,0xd1]
+; CHECK-NEXT:    vmovdqa64 %ymm2, %ymm0 ## encoding: [0x62,0xf1,0xfd,0x28,0x6f,0xc2]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %res = call <4 x i64> @llvm.x86.avx512.psrav.q.256(<4 x i64> %a0, <4 x i64> %a1)
+  %mask.cast = bitcast i8 %mask to <8 x i1>
+  %mask.extract = shufflevector <8 x i1> %mask.cast, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %res2 = select <4 x i1> %mask.extract, <4 x i64> %res, <4 x i64> %a2
+  ret <4 x i64> %res2
+}
+
+define <4 x i64> @test_x86_avx512_maskz_psrav_q_256(<4 x i64> %a0, <4 x i64> %a1, i8 %mask) {
+; CHECK-LABEL: test_x86_avx512_maskz_psrav_q_256:
+; CHECK:       ## BB#0:
+; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
+; CHECK-NEXT:    vpsravq %ymm1, %ymm0, %ymm0 {%k1} {z} ## encoding: [0x62,0xf2,0xfd,0xa9,0x46,0xc1]
+; CHECK-NEXT:    retq ## encoding: [0xc3]
+  %res = call <4 x i64> @llvm.x86.avx512.psrav.q.256(<4 x i64> %a0, <4 x i64> %a1)
+  %mask.cast = bitcast i8 %mask to <8 x i1>
+  %mask.extract = shufflevector <8 x i1> %mask.cast, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %res2 = select <4 x i1> %mask.extract, <4 x i64> %res, <4 x i64> zeroinitializer
+  ret <4 x i64> %res2
+}
+
+declare <4 x i64> @llvm.x86.avx512.psrav.q.256(<4 x i64>, <4 x i64>) nounwind readnone