[x86] Add lowering for @llvm.experimental.vector.compress

llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp

@@ -11582,11 +11582,12 @@ SDValue TargetLowering::expandVECTOR_COMPRESS(SDNode *Node,
     // ... if it is not a splat vector, we need to get the passthru value at
     // position = popcount(mask) and re-load it from the stack before it is
     // overwritten in the loop below.
+    EVT PopcountVT = ScalarVT.changeTypeToInteger();
     SDValue Popcount = DAG.getNode(
         ISD::TRUNCATE, DL, MaskVT.changeVectorElementType(MVT::i1), Mask);
     Popcount = DAG.getNode(ISD::ZERO_EXTEND, DL,
-                           MaskVT.changeVectorElementType(ScalarVT), Popcount);
-    Popcount = DAG.getNode(ISD::VECREDUCE_ADD, DL, ScalarVT, Popcount);
+                           MaskVT.changeVectorElementType(PopcountVT), Popcount);
+    Popcount = DAG.getNode(ISD::VECREDUCE_ADD, DL, PopcountVT, Popcount);
     SDValue LastElmtPtr =
         getVectorElementPointer(DAG, StackPtr, VecVT, Popcount);
     LastWriteVal = DAG.getLoad(
@@ -11625,8 +11626,10 @@ SDValue TargetLowering::expandVECTOR_COMPRESS(SDNode *Node,
 
       // Re-write the last ValI if all lanes were selected. Otherwise,
       // overwrite the last write it with the passthru value.
+      SDNodeFlags Flags{};
+      Flags.setUnpredictable(true);
       LastWriteVal =
-          DAG.getSelect(DL, ScalarVT, AllLanesSelected, ValI, LastWriteVal);
+          DAG.getSelect(DL, ScalarVT, AllLanesSelected, ValI, LastWriteVal, Flags);
       Chain = DAG.getStore(
           Chain, DL, LastWriteVal, OutPtr,
           MachinePointerInfo::getUnknownStack(DAG.getMachineFunction()));

llvm/lib/Target/X86/X86ISelLowering.cpp

@@ -2321,6 +2321,18 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
     }
   }
 
+  if (Subtarget.hasAVX512()) {
+    for (MVT VT : {MVT::v4i32, MVT::v4f32, MVT::v2i64, MVT::v2f64, MVT::v8i32,
+                   MVT::v8f32, MVT::v4i64, MVT::v4f64, MVT::v16i32, MVT::v16f32,
+                   MVT::v8i64, MVT::v8f64})
+      setOperationAction(ISD::VECTOR_COMPRESS, VT, Legal);
+
+    if (Subtarget.hasVBMI2())
+      for (MVT VT : {MVT::v16i8, MVT::v8i16, MVT::v32i8, MVT::v16i16,
+                     MVT::v64i8, MVT::v32i16})
+        setOperationAction(ISD::VECTOR_COMPRESS, VT, Legal);
+  }
+
   if (!Subtarget.useSoftFloat() &&
       (Subtarget.hasAVXNECONVERT() || Subtarget.hasBF16())) {
     addRegisterClass(MVT::v8bf16, Subtarget.hasAVX512() ? &X86::VR128XRegClass

llvm/lib/Target/X86/X86InstrAVX512.td

@@ -10543,6 +10543,12 @@ multiclass compress_by_vec_width_lowering<X86VectorVTInfo _, string Name> {
   def : Pat<(X86compress (_.VT _.RC:$src), _.ImmAllZerosV, _.KRCWM:$mask),
             (!cast<Instruction>(Name#_.ZSuffix#rrkz)
                             _.KRCWM:$mask, _.RC:$src)>;
+  def : Pat<(_.VT (vector_compress _.RC:$src, _.KRCWM:$mask, undef)),
+            (!cast<Instruction>(Name#_.ZSuffix#rrkz)
+                            _.KRCWM:$mask, _.RC:$src)>;
+  def : Pat<(_.VT (vector_compress _.RC:$src, _.KRCWM:$mask, _.RC:$passthru)),
+              (!cast<Instruction>(Name#_.ZSuffix#rrk)
+                            _.RC:$passthru, _.KRCWM:$mask, _.RC:$src)>;
 }
 
 multiclass compress_by_elt_width<bits<8> opc, string OpcodeStr,

llvm/test/CodeGen/X86/vector-compress-avx2.ll

@@ -0,0 +1,281 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -mtriple=x86_64 -mattr=avx2 < %s | FileCheck %s
+
+; The main logic for vpcompress is tested in the -avx512.ll version of this file.
+; This file only checks the fallback expand path.
+
+define <4 x i32> @test_compress_v4i32(<4 x i32> %vec, <4 x i1> %mask, <4 x i32> %passthru) {
+; CHECK-LABEL: test_compress_v4i32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpslld $31, %xmm1, %xmm1
+; CHECK-NEXT:    vpsrad $31, %xmm1, %xmm1
+; CHECK-NEXT:    vmovaps %xmm2, -{{[0-9]+}}(%rsp)
+; CHECK-NEXT:    vpextrd $1, %xmm1, %eax
+; CHECK-NEXT:    vmovd %xmm1, %esi
+; CHECK-NEXT:    andl $1, %esi
+; CHECK-NEXT:    movl %esi, %edi
+; CHECK-NEXT:    subl %eax, %edi
+; CHECK-NEXT:    vpextrd $2, %xmm1, %edx
+; CHECK-NEXT:    subl %edx, %edi
+; CHECK-NEXT:    vpextrd $3, %xmm1, %ecx
+; CHECK-NEXT:    subl %ecx, %edi
+; CHECK-NEXT:    andl $3, %edi
+; CHECK-NEXT:    andl $1, %eax
+; CHECK-NEXT:    addq %rsi, %rax
+; CHECK-NEXT:    andl $1, %edx
+; CHECK-NEXT:    addq %rax, %rdx
+; CHECK-NEXT:    andl $1, %ecx
+; CHECK-NEXT:    addq %rdx, %rcx
+; CHECK-NEXT:    vextractps $3, %xmm0, %r8d
+; CHECK-NEXT:    cmpq $4, %rcx
+; CHECK-NEXT:    cmovbl -24(%rsp,%rdi,4), %r8d
+; CHECK-NEXT:    vmovss %xmm0, -{{[0-9]+}}(%rsp)
+; CHECK-NEXT:    vextractps $1, %xmm0, -24(%rsp,%rsi,4)
+; CHECK-NEXT:    vextractps $2, %xmm0, -24(%rsp,%rax,4)
+; CHECK-NEXT:    andl $3, %edx
+; CHECK-NEXT:    vextractps $3, %xmm0, -24(%rsp,%rdx,4)
+; CHECK-NEXT:    cmpq $3, %rcx
+; CHECK-NEXT:    movl $3, %eax
+; CHECK-NEXT:    cmovbq %rcx, %rax
+; CHECK-NEXT:    movl %r8d, -24(%rsp,%rax,4)
+; CHECK-NEXT:    vmovaps -{{[0-9]+}}(%rsp), %xmm0
+; CHECK-NEXT:    retq
+    %out = call <4 x i32> @llvm.experimental.vector.compress(<4 x i32> %vec, <4 x i1> %mask, <4 x i32> %passthru)
+    ret <4 x i32> %out
+}
+
+define <4 x float> @test_compress_v4f32(<4 x float> %vec, <4 x i1> %mask, <4 x float> %passthru) {
+; CHECK-LABEL: test_compress_v4f32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpslld $31, %xmm1, %xmm1
+; CHECK-NEXT:    vpsrad $31, %xmm1, %xmm1
+; CHECK-NEXT:    vmovaps %xmm2, -{{[0-9]+}}(%rsp)
+; CHECK-NEXT:    vpextrd $1, %xmm1, %edx
+; CHECK-NEXT:    vmovd %xmm1, %esi
+; CHECK-NEXT:    andl $1, %esi
+; CHECK-NEXT:    movl %esi, %edi
+; CHECK-NEXT:    subl %edx, %edi
+; CHECK-NEXT:    vpextrd $2, %xmm1, %ecx
+; CHECK-NEXT:    subl %ecx, %edi
+; CHECK-NEXT:    vpextrd $3, %xmm1, %eax
+; CHECK-NEXT:    subl %eax, %edi
+; CHECK-NEXT:    andl $3, %edi
+; CHECK-NEXT:    vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; CHECK-NEXT:    vmovss %xmm0, -{{[0-9]+}}(%rsp)
+; CHECK-NEXT:    vextractps $1, %xmm0, -24(%rsp,%rsi,4)
+; CHECK-NEXT:    andl $1, %edx
+; CHECK-NEXT:    addq %rsi, %rdx
+; CHECK-NEXT:    vextractps $2, %xmm0, -24(%rsp,%rdx,4)
+; CHECK-NEXT:    andl $1, %ecx
+; CHECK-NEXT:    addq %rdx, %rcx
+; CHECK-NEXT:    andl $1, %eax
+; CHECK-NEXT:    addq %rcx, %rax
+; CHECK-NEXT:    # kill: def $ecx killed $ecx killed $rcx def $rcx
+; CHECK-NEXT:    andl $3, %ecx
+; CHECK-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[3,3,3,3]
+; CHECK-NEXT:    vmovss %xmm0, -24(%rsp,%rcx,4)
+; CHECK-NEXT:    cmpq $3, %rax
+; CHECK-NEXT:    movl $3, %ecx
+; CHECK-NEXT:    cmovbq %rax, %rcx
+; CHECK-NEXT:    ja .LBB1_2
+; CHECK-NEXT:  # %bb.1:
+; CHECK-NEXT:    vmovaps %xmm1, %xmm0
+; CHECK-NEXT:  .LBB1_2:
+; CHECK-NEXT:    vmovss %xmm0, -24(%rsp,%rcx,4)
+; CHECK-NEXT:    vmovaps -{{[0-9]+}}(%rsp), %xmm0
+; CHECK-NEXT:    retq
+    %out = call <4 x float> @llvm.experimental.vector.compress(<4 x float> %vec, <4 x i1> %mask, <4 x float> %passthru)
+    ret <4 x float> %out
+}
+
+define <2 x i64> @test_compress_v2i64(<2 x i64> %vec, <2 x i1> %mask, <2 x i64> %passthru) {
+; CHECK-LABEL: test_compress_v2i64:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpsllq $63, %xmm1, %xmm1
+; CHECK-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; CHECK-NEXT:    vpcmpgtq %xmm1, %xmm3, %xmm1
+; CHECK-NEXT:    vmovaps %xmm2, -{{[0-9]+}}(%rsp)
+; CHECK-NEXT:    vpextrq $1, %xmm1, %rax
+; CHECK-NEXT:    vmovq %xmm1, %rcx
+; CHECK-NEXT:    movl %ecx, %edx
+; CHECK-NEXT:    subl %eax, %edx
+; CHECK-NEXT:    andl $1, %edx
+; CHECK-NEXT:    andl $1, %eax
+; CHECK-NEXT:    andl $1, %ecx
+; CHECK-NEXT:    addq %rcx, %rax
+; CHECK-NEXT:    vpextrq $1, %xmm0, %rsi
+; CHECK-NEXT:    cmpq $2, %rax
+; CHECK-NEXT:    cmovbq -24(%rsp,%rdx,8), %rsi
+; CHECK-NEXT:    vmovq %xmm0, -{{[0-9]+}}(%rsp)
+; CHECK-NEXT:    movl %ecx, %ecx
+; CHECK-NEXT:    vpextrq $1, %xmm0, -24(%rsp,%rcx,8)
+; CHECK-NEXT:    cmpq $1, %rax
+; CHECK-NEXT:    movl $1, %ecx
+; CHECK-NEXT:    cmovbq %rax, %rcx
+; CHECK-NEXT:    movq %rsi, -24(%rsp,%rcx,8)
+; CHECK-NEXT:    vmovaps -{{[0-9]+}}(%rsp), %xmm0
+; CHECK-NEXT:    retq
+    %out = call <2 x i64> @llvm.experimental.vector.compress(<2 x i64> %vec, <2 x i1> %mask, <2 x i64> %passthru)
+    ret <2 x i64> %out
+}
+
+define <2 x double> @test_compress_v2f64(<2 x double> %vec, <2 x i1> %mask, <2 x double> %passthru) {
+; CHECK-LABEL: test_compress_v2f64:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpsllq $63, %xmm1, %xmm1
+; CHECK-NEXT:    vpxor %xmm3, %xmm3, %xmm3
+; CHECK-NEXT:    vpcmpgtq %xmm1, %xmm3, %xmm1
+; CHECK-NEXT:    vmovaps %xmm2, -{{[0-9]+}}(%rsp)
+; CHECK-NEXT:    vpextrq $1, %xmm1, %rax
+; CHECK-NEXT:    vmovq %xmm1, %rcx
+; CHECK-NEXT:    movl %ecx, %edx
+; CHECK-NEXT:    subl %eax, %edx
+; CHECK-NEXT:    andl $1, %edx
+; CHECK-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero
+; CHECK-NEXT:    vmovlpd %xmm0, -{{[0-9]+}}(%rsp)
+; CHECK-NEXT:    andl $1, %ecx
+; CHECK-NEXT:    movl %ecx, %edx
+; CHECK-NEXT:    vmovhpd %xmm0, -24(%rsp,%rdx,8)
+; CHECK-NEXT:    andl $1, %eax
+; CHECK-NEXT:    addq %rcx, %rax
+; CHECK-NEXT:    cmpq $2, %rax
+; CHECK-NEXT:    jb .LBB3_2
+; CHECK-NEXT:  # %bb.1:
+; CHECK-NEXT:    vshufpd {{.*#+}} xmm1 = xmm0[1,0]
+; CHECK-NEXT:  .LBB3_2:
+; CHECK-NEXT:    cmpq $1, %rax
+; CHECK-NEXT:    movl $1, %ecx
+; CHECK-NEXT:    cmovbq %rax, %rcx
+; CHECK-NEXT:    vmovsd %xmm1, -24(%rsp,%rcx,8)
+; CHECK-NEXT:    vmovaps -{{[0-9]+}}(%rsp), %xmm0
+; CHECK-NEXT:    retq
+    %out = call <2 x double> @llvm.experimental.vector.compress(<2 x double> %vec, <2 x i1> %mask, <2 x double> %passthru)
+    ret <2 x double> %out
+}
+
+define <8 x i32> @test_compress_v8i32(<8 x i32> %vec, <8 x i1> %mask, <8 x i32> %passthru) {
+; CHECK-LABEL: test_compress_v8i32:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    pushq %rbp
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:    .cfi_offset %rbp, -16
+; CHECK-NEXT:    movq %rsp, %rbp
+; CHECK-NEXT:    .cfi_def_cfa_register %rbp
+; CHECK-NEXT:    pushq %rbx
+; CHECK-NEXT:    andq $-32, %rsp
+; CHECK-NEXT:    subq $64, %rsp
+; CHECK-NEXT:    .cfi_offset %rbx, -24
+; CHECK-NEXT:    vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
+; CHECK-NEXT:    vpslld $31, %ymm1, %ymm1
+; CHECK-NEXT:    vpsrad $31, %ymm1, %ymm3
+; CHECK-NEXT:    vmovaps %ymm2, (%rsp)
+; CHECK-NEXT:    vextracti128 $1, %ymm3, %xmm1
+; CHECK-NEXT:    vpackssdw %xmm1, %xmm3, %xmm2
+; CHECK-NEXT:    vpmovzxwd {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero
+; CHECK-NEXT:    vpslld $31, %ymm2, %ymm2
+; CHECK-NEXT:    vpsrld $31, %ymm2, %ymm2
+; CHECK-NEXT:    vextracti128 $1, %ymm2, %xmm4
+; CHECK-NEXT:    vpaddd %xmm4, %xmm2, %xmm2
+; CHECK-NEXT:    vpextrd $1, %xmm2, %eax
+; CHECK-NEXT:    vmovd %xmm2, %ecx
+; CHECK-NEXT:    addl %eax, %ecx
+; CHECK-NEXT:    vpextrd $2, %xmm2, %edx
+; CHECK-NEXT:    vpextrd $3, %xmm2, %eax
+; CHECK-NEXT:    addl %edx, %eax
+; CHECK-NEXT:    addl %ecx, %eax
+; CHECK-NEXT:    andl $7, %eax
+; CHECK-NEXT:    vpextrd $1, %xmm3, %ecx
+; CHECK-NEXT:    andl $1, %ecx
+; CHECK-NEXT:    vmovd %xmm3, %edx
+; CHECK-NEXT:    andl $1, %edx
+; CHECK-NEXT:    addq %rdx, %rcx
+; CHECK-NEXT:    vpextrd $2, %xmm3, %esi
+; CHECK-NEXT:    andl $1, %esi
+; CHECK-NEXT:    addq %rcx, %rsi
+; CHECK-NEXT:    vpextrd $3, %xmm3, %edi
+; CHECK-NEXT:    andl $1, %edi
+; CHECK-NEXT:    addq %rsi, %rdi
+; CHECK-NEXT:    vmovd %xmm1, %r8d
+; CHECK-NEXT:    andl $1, %r8d
+; CHECK-NEXT:    addq %rdi, %r8
+; CHECK-NEXT:    vpextrd $1, %xmm1, %r9d
+; CHECK-NEXT:    andl $1, %r9d
+; CHECK-NEXT:    addq %r8, %r9
+; CHECK-NEXT:    vpextrd $2, %xmm1, %r10d
+; CHECK-NEXT:    andl $1, %r10d
+; CHECK-NEXT:    addq %r9, %r10
+; CHECK-NEXT:    vpextrd $3, %xmm1, %r11d
+; CHECK-NEXT:    andl $1, %r11d
+; CHECK-NEXT:    addq %r10, %r11
+; CHECK-NEXT:    vextractf128 $1, %ymm0, %xmm1
+; CHECK-NEXT:    vextractps $3, %xmm1, %ebx
+; CHECK-NEXT:    cmpq $8, %r11
+; CHECK-NEXT:    cmovbl (%rsp,%rax,4), %ebx
+; CHECK-NEXT:    vmovss %xmm0, (%rsp)
+; CHECK-NEXT:    vextractps $1, %xmm0, (%rsp,%rdx,4)
+; CHECK-NEXT:    vextractps $2, %xmm0, (%rsp,%rcx,4)
+; CHECK-NEXT:    vextractps $3, %xmm0, (%rsp,%rsi,4)
+; CHECK-NEXT:    andl $7, %edi
+; CHECK-NEXT:    vmovss %xmm1, (%rsp,%rdi,4)
+; CHECK-NEXT:    andl $7, %r8d
+; CHECK-NEXT:    vextractps $1, %xmm1, (%rsp,%r8,4)
+; CHECK-NEXT:    andl $7, %r9d
+; CHECK-NEXT:    vextractps $2, %xmm1, (%rsp,%r9,4)
+; CHECK-NEXT:    andl $7, %r10d
+; CHECK-NEXT:    vextractps $3, %xmm1, (%rsp,%r10,4)
+; CHECK-NEXT:    cmpq $7, %r11
+; CHECK-NEXT:    movl $7, %eax
+; CHECK-NEXT:    cmovbq %r11, %rax
+; CHECK-NEXT:    movl %eax, %eax
+; CHECK-NEXT:    movl %ebx, (%rsp,%rax,4)
+; CHECK-NEXT:    vmovaps (%rsp), %ymm0
+; CHECK-NEXT:    leaq -8(%rbp), %rsp
+; CHECK-NEXT:    popq %rbx
+; CHECK-NEXT:    popq %rbp
+; CHECK-NEXT:    .cfi_def_cfa %rsp, 8
+; CHECK-NEXT:    retq
+    %out = call <8 x i32> @llvm.experimental.vector.compress(<8 x i32> %vec, <8 x i1> %mask, <8 x i32> %passthru)
+    ret <8 x i32> %out
+}
+
+define <4 x i32> @test_compress_all_const() {
+; CHECK-LABEL: test_compress_all_const:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vmovsd {{.*#+}} xmm0 = [5,9,0,0]
+; CHECK-NEXT:    retq
+    %out = call <4 x i32> @llvm.experimental.vector.compress(<4 x i32> <i32 3, i32 5, i32 7, i32 9>,
+                                                <4 x i1>   <i1 0,  i1 1,  i1 0,  i1 1>,
+                                                <4 x i32> undef)
+    ret <4 x i32> %out
+}
+
+define <4 x i32> @test_compress_const_mask(<4 x i32> %vec) {
+; CHECK-LABEL: test_compress_const_mask:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,3,2,3]
+; CHECK-NEXT:    retq
+    %out = call <4 x i32> @llvm.experimental.vector.compress(<4 x i32> %vec, <4 x i1> <i1 1, i1 undef, i1 0, i1 1>, <4 x i32> undef)
+    ret <4 x i32> %out
+}
+
+define <4 x i32> @test_compress_const_mask_passthrough(<4 x i32> %vec, <4 x i32> %passthru) {
+; CHECK-LABEL: test_compress_const_mask_passthrough:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vshufps {{.*#+}} xmm0 = xmm0[0,3],xmm1[2,3]
+; CHECK-NEXT:    retq
+    %out = call <4 x i32> @llvm.experimental.vector.compress(<4 x i32> %vec, <4 x i1> <i1 1, i1 undef, i1 0, i1 1>, <4 x i32> %passthru)
+    ret <4 x i32> %out
+}
+
+define <4 x i32> @test_compress_const_mask_const_passthrough(<4 x i32> %vec) {
+; CHECK-LABEL: test_compress_const_mask_const_passthrough:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,3,2,3]
+; CHECK-NEXT:    movl $7, %eax
+; CHECK-NEXT:    vpinsrd $2, %eax, %xmm0, %xmm0
+; CHECK-NEXT:    movl $8, %eax
+; CHECK-NEXT:    vpinsrd $3, %eax, %xmm0, %xmm0
+; CHECK-NEXT:    retq
+    %out = call <4 x i32> @llvm.experimental.vector.compress(<4 x i32> %vec, <4 x i1> <i1 1, i1 0, i1 0, i1 1>, <4 x i32> <i32 5, i32 6, i32 7, i32 8>)
+    ret <4 x i32> %out
+}