[X86] Custom codegen 512-bit cvt(u)qq2tops, cvt(u)qqtopd, and cvt(u)dqtops intrinsics.

Summary:
The 512-bit cvt(u)qq2tops, cvt(u)qqtopd, and cvt(u)dqtops intrinsics all have the possibility of taking an explicit rounding mode argument. If the rounding mode is CUR_DIRECTION we'd like to emit a sitofp/uitofp instruction and a select like we do for 256-bit intrinsics.

For cvt(u)qqtopd and cvt(u)dqtops we do this when the form of the software intrinsics that doesn't take a rounding mode argument is used. This is done by using convertvector in the header with the select builtin. But if the explicit rounding mode form of the intrinsic is used and CUR_DIRECTION is passed, we don't do this. We shouldn't have this inconsistency.

For cvt(u)qqtops nothing is done because we can't use the select builtin in the header without avx512vl. So we need to use custom codegen for this.

Even when the rounding mode isn't CUR_DIRECTION we should also use select in IR for consistency. And it will remove another scalar integer mask from our intrinsics.

To accomplish all of these goals I've taken a slightly unusual approach. I've added two new X86 specific intrinsics for sitofp/uitofp with rounding. These intrinsics are variadic on the input and output type so we only need 2 instead of 6. This avoids the need for a switch to map them in CGBuiltin.cpp. We just need to check signed vs unsigned. I believe other targets also use variadic intrinsics like this.

So if the rounding mode is CUR_DIRECTION we'll use an sitofp/uitofp instruction. Otherwise we'll use one of the new intrinsics. After that we'll emit a select instruction if needed.

Reviewers: RKSimon, spatel

Reviewed By: RKSimon

Subscribers: cfe-commits

Differential Revision: https://reviews.llvm.org/D56998

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@352267 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/CodeGen/CGBuiltin.cpp b/lib/CodeGen/CGBuiltin.cpp
index 390d28fd08da121b81e24dd4314352e6ec403a76..2aa61ffc25a5ff474d847a0a624b6a0723046892 100644 (file)
--- a/lib/CodeGen/CGBuiltin.cpp
+++ b/lib/CodeGen/CGBuiltin.cpp
@@ -9328,6 +9328,25 @@ static Value *EmitX86ConvertToMask(CodeGenFunction &CGF, Value *In) {
   return EmitX86MaskedCompare(CGF, 1, true, { In, Zero });
 }
 
+static Value *EmitX86ConvertIntToFp(CodeGenFunction &CGF,
+                                    ArrayRef<Value *> Ops, bool IsSigned) {
+  unsigned Rnd = cast<llvm::ConstantInt>(Ops[3])->getZExtValue();
+  llvm::Type *Ty = Ops[1]->getType();
+
+  Value *Res;
+  if (Rnd != 4) {
+    Intrinsic::ID IID = IsSigned ? Intrinsic::x86_avx512_sitofp_round
+                                 : Intrinsic::x86_avx512_uitofp_round;
+    Function *F = CGF.CGM.getIntrinsic(IID, { Ty, Ops[0]->getType() });
+    Res = CGF.Builder.CreateCall(F, { Ops[0], Ops[3] });
+  } else {
+    Res = IsSigned ? CGF.Builder.CreateSIToFP(Ops[0], Ty)
+                   : CGF.Builder.CreateUIToFP(Ops[0], Ty);
+  }
+
+  return EmitX86Select(CGF, Ops[2], Res, Ops[1]);
+}
+
 static Value *EmitX86Abs(CodeGenFunction &CGF, ArrayRef<Value *> Ops) {
 
   llvm::Type *Ty = Ops[0]->getType();
@@ -9989,6 +10008,15 @@ Value *CodeGenFunction::EmitX86BuiltinExpr(unsigned BuiltinID,
   case X86::BI__builtin_ia32_cvtq2mask512:
     return EmitX86ConvertToMask(*this, Ops[0]);
 
+  case X86::BI__builtin_ia32_cvtdq2ps512_mask:
+  case X86::BI__builtin_ia32_cvtqq2ps512_mask:
+  case X86::BI__builtin_ia32_cvtqq2pd512_mask:
+    return EmitX86ConvertIntToFp(*this, Ops, /*IsSigned*/true);
+  case X86::BI__builtin_ia32_cvtudq2ps512_mask:
+  case X86::BI__builtin_ia32_cvtuqq2ps512_mask:
+  case X86::BI__builtin_ia32_cvtuqq2pd512_mask:
+    return EmitX86ConvertIntToFp(*this, Ops, /*IsSigned*/false);
+
   case X86::BI__builtin_ia32_vfmaddss3:
   case X86::BI__builtin_ia32_vfmaddsd3:
   case X86::BI__builtin_ia32_vfmaddss3_mask:

diff --git a/test/CodeGen/avx512dq-builtins.c b/test/CodeGen/avx512dq-builtins.c
index 6227a83b55d430124c68ad50e1984533458a6783..a85e173432d1a5f38a311bf4254376ff89f7c88c 100644 (file)
--- a/test/CodeGen/avx512dq-builtins.c
+++ b/test/CodeGen/avx512dq-builtins.c
@@ -613,55 +613,61 @@ __m512d test_mm512_maskz_cvtepi64_pd(__mmask8 __U, __m512i __A) {
 
 __m512d test_mm512_cvt_roundepi64_pd(__m512i __A) {
   // CHECK-LABEL: @test_mm512_cvt_roundepi64_pd
-  // CHECK: @llvm.x86.avx512.mask.cvtqq2pd.512
+  // CHECK: @llvm.x86.avx512.sitofp.round.v8f64.v8i64
   return _mm512_cvt_roundepi64_pd(__A, _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC); 
 }
 
 __m512d test_mm512_mask_cvt_roundepi64_pd(__m512d __W, __mmask8 __U, __m512i __A) {
   // CHECK-LABEL: @test_mm512_mask_cvt_roundepi64_pd
-  // CHECK: @llvm.x86.avx512.mask.cvtqq2pd.512
+  // CHECK: @llvm.x86.avx512.sitofp.round.v8f64.v8i64
+  // CHECK: select <8 x i1> %{{.*}}, <8 x double> %{{.*}}, <8 x double> %{{.*}}
   return _mm512_mask_cvt_roundepi64_pd(__W, __U, __A, _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC); 
 }
 
 __m512d test_mm512_maskz_cvt_roundepi64_pd(__mmask8 __U, __m512i __A) {
   // CHECK-LABEL: @test_mm512_maskz_cvt_roundepi64_pd
-  // CHECK: @llvm.x86.avx512.mask.cvtqq2pd.512
+  // CHECK: @llvm.x86.avx512.sitofp.round.v8f64.v8i64
+  // CHECK: select <8 x i1> %{{.*}}, <8 x double> %{{.*}}, <8 x double> %{{.*}}
   return _mm512_maskz_cvt_roundepi64_pd(__U, __A, _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC); 
 }
 
 __m256 test_mm512_cvtepi64_ps(__m512i __A) {
   // CHECK-LABEL: @test_mm512_cvtepi64_ps
-  // CHECK: @llvm.x86.avx512.mask.cvtqq2ps.512
+  // CHECK: sitofp <8 x i64> %{{.*}} to <8 x float>
   return _mm512_cvtepi64_ps(__A); 
 }
 
 __m256 test_mm512_mask_cvtepi64_ps(__m256 __W, __mmask8 __U, __m512i __A) {
   // CHECK-LABEL: @test_mm512_mask_cvtepi64_ps
-  // CHECK: @llvm.x86.avx512.mask.cvtqq2ps.512
+  // CHECK: sitofp <8 x i64> %{{.*}} to <8 x float>
+  // CHECK: select <8 x i1> %{{.*}}, <8 x float> %{{.*}}, <8 x float> %{{.*}}
   return _mm512_mask_cvtepi64_ps(__W, __U, __A); 
 }
 
 __m256 test_mm512_maskz_cvtepi64_ps(__mmask8 __U, __m512i __A) {
   // CHECK-LABEL: @test_mm512_maskz_cvtepi64_ps
-  // CHECK: @llvm.x86.avx512.mask.cvtqq2ps.512
+  // CHECK: sitofp <8 x i64> %{{.*}} to <8 x float>
+  // CHECK: select <8 x i1> %{{.*}}, <8 x float> %{{.*}}, <8 x float> %{{.*}}
   return _mm512_maskz_cvtepi64_ps(__U, __A); 
 }
 
 __m256 test_mm512_cvt_roundepi64_ps(__m512i __A) {
   // CHECK-LABEL: @test_mm512_cvt_roundepi64_ps
-  // CHECK: @llvm.x86.avx512.mask.cvtqq2ps.512
+  // CHECK: @llvm.x86.avx512.sitofp.round.v8f32.v8i64
   return _mm512_cvt_roundepi64_ps(__A, _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC); 
 }
 
 __m256 test_mm512_mask_cvt_roundepi64_ps(__m256 __W, __mmask8 __U, __m512i __A) {
   // CHECK-LABEL: @test_mm512_mask_cvt_roundepi64_ps
-  // CHECK: @llvm.x86.avx512.mask.cvtqq2ps.512
+  // CHECK: @llvm.x86.avx512.sitofp.round.v8f32.v8i64
+  // CHECK: select <8 x i1> %{{.*}}, <8 x float> %{{.*}}, <8 x float> %{{.*}}
   return _mm512_mask_cvt_roundepi64_ps(__W, __U, __A, _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC); 
 }
 
 __m256 test_mm512_maskz_cvt_roundepi64_ps(__mmask8 __U, __m512i __A) {
   // CHECK-LABEL: @test_mm512_maskz_cvt_roundepi64_ps
-  // CHECK: @llvm.x86.avx512.mask.cvtqq2ps.512
+  // CHECK: @llvm.x86.avx512.sitofp.round.v8f32.v8i64
+  // CHECK: select <8 x i1> %{{.*}}, <8 x float> %{{.*}}, <8 x float> %{{.*}}
   return _mm512_maskz_cvt_roundepi64_ps(__U, __A, _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC); 
 }
 
@@ -831,55 +837,61 @@ __m512d test_mm512_maskz_cvtepu64_pd(__mmask8 __U, __m512i __A) {
 
 __m512d test_mm512_cvt_roundepu64_pd(__m512i __A) {
   // CHECK-LABEL: @test_mm512_cvt_roundepu64_pd
-  // CHECK: @llvm.x86.avx512.mask.cvtuqq2pd.512
+  // CHECK: @llvm.x86.avx512.uitofp.round.v8f64.v8i64
   return _mm512_cvt_roundepu64_pd(__A, _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC); 
 }
 
 __m512d test_mm512_mask_cvt_roundepu64_pd(__m512d __W, __mmask8 __U, __m512i __A) {
   // CHECK-LABEL: @test_mm512_mask_cvt_roundepu64_pd
-  // CHECK: @llvm.x86.avx512.mask.cvtuqq2pd.512
+  // CHECK: @llvm.x86.avx512.uitofp.round.v8f64.v8i64
+  // CHECK: select <8 x i1> %{{.*}}, <8 x double> %{{.*}}, <8 x double> %{{.*}}
   return _mm512_mask_cvt_roundepu64_pd(__W, __U, __A, _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC); 
 }
 
 __m512d test_mm512_maskz_cvt_roundepu64_pd(__mmask8 __U, __m512i __A) {
   // CHECK-LABEL: @test_mm512_maskz_cvt_roundepu64_pd
-  // CHECK: @llvm.x86.avx512.mask.cvtuqq2pd.512
+  // CHECK: @llvm.x86.avx512.uitofp.round.v8f64.v8i64
+  // CHECK: select <8 x i1> %{{.*}}, <8 x double> %{{.*}}, <8 x double> %{{.*}}
   return _mm512_maskz_cvt_roundepu64_pd(__U, __A, _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC); 
 }
 
 __m256 test_mm512_cvtepu64_ps(__m512i __A) {
   // CHECK-LABEL: @test_mm512_cvtepu64_ps
-  // CHECK: @llvm.x86.avx512.mask.cvtuqq2ps.512
+  // CHECK: uitofp <8 x i64> %{{.*}} to <8 x float>
   return _mm512_cvtepu64_ps(__A); 
 }
 
 __m256 test_mm512_mask_cvtepu64_ps(__m256 __W, __mmask8 __U, __m512i __A) {
   // CHECK-LABEL: @test_mm512_mask_cvtepu64_ps
-  // CHECK: @llvm.x86.avx512.mask.cvtuqq2ps.512
+  // CHECK: uitofp <8 x i64> %{{.*}} to <8 x float>
+  // CHECK: select <8 x i1> %{{.*}}, <8 x float> %{{.*}}, <8 x float> %{{.*}}
   return _mm512_mask_cvtepu64_ps(__W, __U, __A); 
 }
 
 __m256 test_mm512_maskz_cvtepu64_ps(__mmask8 __U, __m512i __A) {
   // CHECK-LABEL: @test_mm512_maskz_cvtepu64_ps
-  // CHECK: @llvm.x86.avx512.mask.cvtuqq2ps.512
+  // CHECK: uitofp <8 x i64> %{{.*}} to <8 x float>
+  // CHECK: select <8 x i1> %{{.*}}, <8 x float> %{{.*}}, <8 x float> %{{.*}}
   return _mm512_maskz_cvtepu64_ps(__U, __A); 
 }
 
 __m256 test_mm512_cvt_roundepu64_ps(__m512i __A) {
   // CHECK-LABEL: @test_mm512_cvt_roundepu64_ps
-  // CHECK: @llvm.x86.avx512.mask.cvtuqq2ps.512
+  // CHECK: @llvm.x86.avx512.uitofp.round.v8f32.v8i64
   return _mm512_cvt_roundepu64_ps(__A, _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC); 
 }
 
 __m256 test_mm512_mask_cvt_roundepu64_ps(__m256 __W, __mmask8 __U, __m512i __A) {
   // CHECK-LABEL: @test_mm512_mask_cvt_roundepu64_ps
-  // CHECK: @llvm.x86.avx512.mask.cvtuqq2ps.512
+  // CHECK: @llvm.x86.avx512.uitofp.round.v8f32.v8i64
+  // CHECK: select <8 x i1> %{{.*}}, <8 x float> %{{.*}}, <8 x float> %{{.*}}
   return _mm512_mask_cvt_roundepu64_ps(__W, __U, __A, _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC); 
 }
 
 __m256 test_mm512_maskz_cvt_roundepu64_ps(__mmask8 __U, __m512i __A) {
   // CHECK-LABEL: @test_mm512_maskz_cvt_roundepu64_ps
-  // CHECK: @llvm.x86.avx512.mask.cvtuqq2ps.512
+  // CHECK: @llvm.x86.avx512.uitofp.round.v8f32.v8i64
+  // CHECK: select <8 x i1> %{{.*}}, <8 x float> %{{.*}}, <8 x float> %{{.*}}
   return _mm512_maskz_cvt_roundepu64_ps(__U, __A, _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC); 
 }
 

diff --git a/test/CodeGen/avx512f-builtins.c b/test/CodeGen/avx512f-builtins.c
index 6b041cea71cb616d2ca54bd6c1d2f0287148b089..55bdf4f5fc4ef1b5f2af230ee6aac1d87387537b 100644 (file)
--- a/test/CodeGen/avx512f-builtins.c
+++ b/test/CodeGen/avx512f-builtins.c
@@ -5022,42 +5022,46 @@ __m512 test_mm512_maskz_cvt_roundph_ps(__mmask16 __U, __m256i __A)
 __m512 test_mm512_cvt_roundepi32_ps( __m512i __A)
 {
   // CHECK-LABEL: @test_mm512_cvt_roundepi32_ps
-  // CHECK: @llvm.x86.avx512.mask.cvtdq2ps.512
+  // CHECK: @llvm.x86.avx512.sitofp.round.v16f32.v16i32
   return _mm512_cvt_roundepi32_ps(__A, _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
 }
 
 __m512 test_mm512_mask_cvt_roundepi32_ps(__m512 __W, __mmask16 __U, __m512i __A)
 {
   // CHECK-LABEL: @test_mm512_mask_cvt_roundepi32_ps
-  // CHECK: @llvm.x86.avx512.mask.cvtdq2ps.512
+  // CHECK: @llvm.x86.avx512.sitofp.round.v16f32.v16i32
+  // CHECK: select <16 x i1> %{{.*}}, <16 x float> %{{.*}}, <16 x float> %{{.*}}
   return _mm512_mask_cvt_roundepi32_ps(__W,__U,__A, _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
 }
 
 __m512 test_mm512_maskz_cvt_roundepi32_ps(__mmask16 __U, __m512i __A)
 {
   // CHECK-LABEL: @test_mm512_maskz_cvt_roundepi32_ps
-  // CHECK: @llvm.x86.avx512.mask.cvtdq2ps.512
+  // CHECK: @llvm.x86.avx512.sitofp.round.v16f32.v16i32
+  // CHECK: select <16 x i1> %{{.*}}, <16 x float> %{{.*}}, <16 x float> %{{.*}}
   return _mm512_maskz_cvt_roundepi32_ps(__U,__A, _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
 }
 
 __m512 test_mm512_cvt_roundepu32_ps(__m512i __A)
 {
   // CHECK-LABEL: @test_mm512_cvt_roundepu32_ps
-  // CHECK: @llvm.x86.avx512.mask.cvtudq2ps.512
+  // CHECK: @llvm.x86.avx512.uitofp.round.v16f32.v16i32
   return _mm512_cvt_roundepu32_ps(__A, _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
 }
 
 __m512 test_mm512_mask_cvt_roundepu32_ps(__m512 __W, __mmask16 __U,__m512i __A)
 {
   // CHECK-LABEL: @test_mm512_mask_cvt_roundepu32_ps
-  // CHECK: @llvm.x86.avx512.mask.cvtudq2ps.512
+  // CHECK: @llvm.x86.avx512.uitofp.round.v16f32.v16i32
+  // CHECK: select <16 x i1> %{{.*}}, <16 x float> %{{.*}}, <16 x float> %{{.*}}
   return _mm512_mask_cvt_roundepu32_ps(__W,__U,__A, _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
 }
 
 __m512 test_mm512_maskz_cvt_roundepu32_ps(__mmask16 __U,__m512i __A)
 {
   // CHECK-LABEL: @test_mm512_maskz_cvt_roundepu32_ps
-  // CHECK: @llvm.x86.avx512.mask.cvtudq2ps.512
+  // CHECK: @llvm.x86.avx512.uitofp.round.v16f32.v16i32
+  // CHECK: select <16 x i1> %{{.*}}, <16 x float> %{{.*}}, <16 x float> %{{.*}}
   return _mm512_maskz_cvt_roundepu32_ps(__U,__A, _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
 }