[AVX-512][InstCombine] Teach InstCombine to turn scalar add/sub/mul/div with rounding intrinsics into normal IR operations if the rounding mode is CUR_DIRECTION.

[craig.topper]

Summary

I only do this for unmasked cases for now because isel is failing to fold the mask. I'll try to fix that soon.

I'll do the same thing for packed add/sub/mul/div in a future patch.

[zvi]

Other than a minor comment, LGTM.

lib/Transforms/InstCombine/InstCombineCalls.cpp

1783

is if(dync_cast<>) really needed here? Can the cast ever fail?

[craig.topper]

lib/Transforms/InstCombine/InstCombineCalls.cpp

1783

You're right the cast can never fail for well formed use of the intrinsic. I'll change it before I commit.

[RKSimon]

lib/Transforms/InstCombine/InstCombineCalls.cpp

1783

Will something have caught badly formed uses of the intrinsic before this?

[craig.topper]

I think Simon is right. Nothing will catch the badly formed intrinsic before this. So we should keep a dyn_cast here.

[craig.topper]

Accepting based on Zvi's LGTM and determining that the dyn_cast helps keep the optimization robust against badly formed uses of the intrinsics.

lib/Transforms/InstCombine/InstCombineCalls.cpp

if (Value *V = SimplifyDemandedVectorEltsLow(Arg1, VWidth, 1)) {
       II->setArgOperand(1, V);
       MadeChange = true;
     }
     if (MadeChange)
       return II;
     break;
   }
 
-  // X86 scalar intrinsics simplified with SimplifyDemandedVectorElts.
   case Intrinsic::x86_avx512_mask_add_ss_round:
   case Intrinsic::x86_avx512_mask_div_ss_round:
   case Intrinsic::x86_avx512_mask_mul_ss_round:
   case Intrinsic::x86_avx512_mask_sub_ss_round:
-  case Intrinsic::x86_avx512_mask_max_ss_round:
-  case Intrinsic::x86_avx512_mask_min_ss_round:
   case Intrinsic::x86_avx512_mask_add_sd_round:
   case Intrinsic::x86_avx512_mask_div_sd_round:
   case Intrinsic::x86_avx512_mask_mul_sd_round:
   case Intrinsic::x86_avx512_mask_sub_sd_round:
+    // If the rounding mode is CUR_DIRECTION(4) we can turn these into regular
+    // IR operations.
+    if (auto *R = dyn_cast<ConstantInt>(II->getArgOperand(4))) {
+      if (R->getValue() == 4) {
+        // Only do this if the mask bit is 1 so that we don't need a select.
+        // TODO: Improve this to handle masking cases. Isel doesn't fold
+        // the mask correctly right now.
+        if (auto *M = dyn_cast<ConstantInt>(II->getArgOperand(3))) {
+          if (M->getValue()[0]) {
+            // Extract the element as scalars.
+            Value *Arg0 = II->getArgOperand(0);
+            Value *Arg1 = II->getArgOperand(1);
+            Value *LHS = Builder->CreateExtractElement(Arg0, (uint64_t)0);
+            Value *RHS = Builder->CreateExtractElement(Arg1, (uint64_t)0);
+
+            Value *V;
+            switch (II->getIntrinsicID()) {
+            default: llvm_unreachable("Case stmts out of sync!");
+            case Intrinsic::x86_avx512_mask_add_ss_round:
+            case Intrinsic::x86_avx512_mask_add_sd_round:
+              V = Builder->CreateFAdd(LHS, RHS);
+              break;
+            case Intrinsic::x86_avx512_mask_sub_ss_round:
+            case Intrinsic::x86_avx512_mask_sub_sd_round:
+              V = Builder->CreateFSub(LHS, RHS);
+              break;
+            case Intrinsic::x86_avx512_mask_mul_ss_round:
+            case Intrinsic::x86_avx512_mask_mul_sd_round:
+              V = Builder->CreateFMul(LHS, RHS);
+              break;
+            case Intrinsic::x86_avx512_mask_div_ss_round:
+            case Intrinsic::x86_avx512_mask_div_sd_round:
+              V = Builder->CreateFDiv(LHS, RHS);
+              break;
+            }
+
+            // Insert the result back into the original argument 0.
+            V = Builder->CreateInsertElement(Arg0, V, (uint64_t)0);
+
+            return replaceInstUsesWith(*II, V);
+          }
+        }
+      }
+    }
+    LLVM_FALLTHROUGH;
+
+  // X86 scalar intrinsics simplified with SimplifyDemandedVectorElts.
+  case Intrinsic::x86_avx512_mask_max_ss_round:
+  case Intrinsic::x86_avx512_mask_min_ss_round:
   case Intrinsic::x86_avx512_mask_max_sd_round:
   case Intrinsic::x86_avx512_mask_min_sd_round:
   case Intrinsic::x86_avx512_mask_vfmadd_ss:
   case Intrinsic::x86_avx512_mask_vfmadd_sd:
   case Intrinsic::x86_avx512_maskz_vfmadd_ss:
   case Intrinsic::x86_avx512_maskz_vfmadd_sd:
   case Intrinsic::x86_avx512_mask3_vfmadd_ss:
   case Intrinsic::x86_avx512_mask3_vfmadd_sd:

test/Transforms/InstCombine/x86-avx512.ll

 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
 ; RUN: opt < %s -instcombine -S | FileCheck %s
 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
 
 declare <4 x float> @llvm.x86.avx512.mask.add.ss.round(<4 x float>, <4 x float>, <4 x float>, i8, i32)
 
 define <4 x float> @test_add_ss(<4 x float> %a, <4 x float> %b) {
 ; CHECK-LABEL: @test_add_ss(
-; CHECK-NEXT:    [[TMP1:%.*]] = tail call <4 x float> @llvm.x86.avx512.mask.add.ss.round(<4 x float> %a, <4 x float> %b, <4 x float> undef, i8 -1, i32 4)
-; CHECK-NEXT:    ret <4 x float> [[TMP1]]
+; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <4 x float> %a, i64 0
+; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <4 x float> %b, i32 0
+; CHECK-NEXT:    [[TMP3:%.*]] = fadd float [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <4 x float> %a, float [[TMP3]], i64 0
+; CHECK-NEXT:    ret <4 x float> [[TMP4]]
 ;
   %1 = insertelement <4 x float> %b, float 1.000000e+00, i32 1
   %2 = insertelement <4 x float> %1, float 2.000000e+00, i32 2
   %3 = insertelement <4 x float> %2, float 3.000000e+00, i32 3
   %4 = tail call <4 x float> @llvm.x86.avx512.mask.add.ss.round(<4 x float> %a, <4 x float> %3, <4 x float> undef, i8 -1, i32 4)
   ret <4 x float> %4
 }
 
+define <4 x float> @test_add_ss_round(<4 x float> %a, <4 x float> %b) {
+; CHECK-LABEL: @test_add_ss_round(
+; CHECK-NEXT:    [[TMP1:%.*]] = tail call <4 x float> @llvm.x86.avx512.mask.add.ss.round(<4 x float> %a, <4 x float> %b, <4 x float> undef, i8 -1, i32 8)
+; CHECK-NEXT:    ret <4 x float> [[TMP1]]
+;
+  %1 = insertelement <4 x float> %b, float 1.000000e+00, i32 1
+  %2 = insertelement <4 x float> %1, float 2.000000e+00, i32 2
+  %3 = insertelement <4 x float> %2, float 3.000000e+00, i32 3
+  %4 = tail call <4 x float> @llvm.x86.avx512.mask.add.ss.round(<4 x float> %a, <4 x float> %3, <4 x float> undef, i8 -1, i32 8)
+  ret <4 x float> %4
+}
+
 define <4 x float> @test_add_ss_mask(<4 x float> %a, <4 x float> %b, <4 x float> %c, i8 %mask) {
 ; CHECK-LABEL: @test_add_ss_mask(
 ; CHECK-NEXT:    [[TMP1:%.*]] = tail call <4 x float> @llvm.x86.avx512.mask.add.ss.round(<4 x float> %a, <4 x float> %b, <4 x float> %c, i8 %mask, i32 4)
 ; CHECK-NEXT:    ret <4 x float> [[TMP1]]
 ;
   %1 = insertelement <4 x float> %c, float 1.000000e+00, i32 1
   %2 = insertelement <4 x float> %1, float 2.000000e+00, i32 2
   %3 = insertelement <4 x float> %2, float 3.000000e+00, i32 3
;
   %10 = extractelement <4 x float> %9, i32 1
   ret float %10
 }
 
 declare <2 x double> @llvm.x86.avx512.mask.add.sd.round(<2 x double>, <2 x double>, <2 x double>, i8, i32)
 
 define <2 x double> @test_add_sd(<2 x double> %a, <2 x double> %b) {
 ; CHECK-LABEL: @test_add_sd(
-; CHECK-NEXT:    [[TMP1:%.*]] = tail call <2 x double> @llvm.x86.avx512.mask.add.sd.round(<2 x double> %a, <2 x double> %b, <2 x double> undef, i8 -1, i32 4)
-; CHECK-NEXT:    ret <2 x double> [[TMP1]]
+; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <2 x double> %a, i64 0
+; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <2 x double> %b, i64 0
+; CHECK-NEXT:    [[TMP3:%.*]] = fadd double [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <2 x double> %a, double [[TMP3]], i64 0
+; CHECK-NEXT:    ret <2 x double> [[TMP4]]
 ;
   %1 = insertelement <2 x double> %b, double 1.000000e+00, i32 1
   %2 = tail call <2 x double> @llvm.x86.avx512.mask.add.sd.round(<2 x double> %a, <2 x double> %1, <2 x double> undef, i8 -1, i32 4)
   ret <2 x double> %2
 }
 
+define <2 x double> @test_add_sd_round(<2 x double> %a, <2 x double> %b) {
+; CHECK-LABEL: @test_add_sd_round(
+; CHECK-NEXT:    [[TMP1:%.*]] = tail call <2 x double> @llvm.x86.avx512.mask.add.sd.round(<2 x double> %a, <2 x double> %b, <2 x double> undef, i8 -1, i32 8)
+; CHECK-NEXT:    ret <2 x double> [[TMP1]]
+;
+  %1 = insertelement <2 x double> %b, double 1.000000e+00, i32 1
+  %2 = tail call <2 x double> @llvm.x86.avx512.mask.add.sd.round(<2 x double> %a, <2 x double> %1, <2 x double> undef, i8 -1, i32 8)
+  ret <2 x double> %2
+}
+
 define <2 x double> @test_add_sd_mask(<2 x double> %a, <2 x double> %b, <2 x double> %c, i8 %mask) {
 ; CHECK-LABEL: @test_add_sd_mask(
 ; CHECK-NEXT:    [[TMP1:%.*]] = tail call <2 x double> @llvm.x86.avx512.mask.add.sd.round(<2 x double> %a, <2 x double> %b, <2 x double> %c, i8 %mask, i32 4)
 ; CHECK-NEXT:    ret <2 x double> [[TMP1]]
 ;
   %1 = insertelement <2 x double> %c, double 1.000000e+00, i32 1
   %2 = tail call <2 x double> @llvm.x86.avx512.mask.add.sd.round(<2 x double> %a, <2 x double> %b, <2 x double> %1, i8 %mask, i32 4)
   ret <2 x double> %2
;
   %6 = extractelement <2 x double> %5, i32 1
   ret double %6
 }
 
 declare <4 x float> @llvm.x86.avx512.mask.sub.ss.round(<4 x float>, <4 x float>, <4 x float>, i8, i32)
 
 define <4 x float> @test_sub_ss(<4 x float> %a, <4 x float> %b) {
 ; CHECK-LABEL: @test_sub_ss(
-; CHECK-NEXT:    [[TMP1:%.*]] = tail call <4 x float> @llvm.x86.avx512.mask.sub.ss.round(<4 x float> %a, <4 x float> %b, <4 x float> undef, i8 -1, i32 4)
-; CHECK-NEXT:    ret <4 x float> [[TMP1]]
+; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <4 x float> %a, i64 0
+; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <4 x float> %b, i32 0
+; CHECK-NEXT:    [[TMP3:%.*]] = fsub float [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <4 x float> %a, float [[TMP3]], i64 0
+; CHECK-NEXT:    ret <4 x float> [[TMP4]]
 ;
   %1 = insertelement <4 x float> %b, float 1.000000e+00, i32 1
   %2 = insertelement <4 x float> %1, float 2.000000e+00, i32 2
   %3 = insertelement <4 x float> %2, float 3.000000e+00, i32 3
   %4 = tail call <4 x float> @llvm.x86.avx512.mask.sub.ss.round(<4 x float> %a, <4 x float> %3, <4 x float> undef, i8 -1, i32 4)
   ret <4 x float> %4
 }
 
+define <4 x float> @test_sub_ss_round(<4 x float> %a, <4 x float> %b) {
+; CHECK-LABEL: @test_sub_ss_round(
+; CHECK-NEXT:    [[TMP1:%.*]] = tail call <4 x float> @llvm.x86.avx512.mask.sub.ss.round(<4 x float> %a, <4 x float> %b, <4 x float> undef, i8 -1, i32 8)
+; CHECK-NEXT:    ret <4 x float> [[TMP1]]
+;
+  %1 = insertelement <4 x float> %b, float 1.000000e+00, i32 1
+  %2 = insertelement <4 x float> %1, float 2.000000e+00, i32 2
+  %3 = insertelement <4 x float> %2, float 3.000000e+00, i32 3
+  %4 = tail call <4 x float> @llvm.x86.avx512.mask.sub.ss.round(<4 x float> %a, <4 x float> %3, <4 x float> undef, i8 -1, i32 8)
+  ret <4 x float> %4
+}
+
 define <4 x float> @test_sub_ss_mask(<4 x float> %a, <4 x float> %b, <4 x float> %c, i8 %mask) {
 ; CHECK-LABEL: @test_sub_ss_mask(
 ; CHECK-NEXT:    [[TMP1:%.*]] = tail call <4 x float> @llvm.x86.avx512.mask.sub.ss.round(<4 x float> %a, <4 x float> %b, <4 x float> %c, i8 %mask, i32 4)
 ; CHECK-NEXT:    ret <4 x float> [[TMP1]]
 ;
   %1 = insertelement <4 x float> %c, float 1.000000e+00, i32 1
   %2 = insertelement <4 x float> %1, float 2.000000e+00, i32 2
   %3 = insertelement <4 x float> %2, float 3.000000e+00, i32 3
;
   %10 = extractelement <4 x float> %9, i32 1
   ret float %10
 }
 
 declare <2 x double> @llvm.x86.avx512.mask.sub.sd.round(<2 x double>, <2 x double>, <2 x double>, i8, i32)
 
 define <2 x double> @test_sub_sd(<2 x double> %a, <2 x double> %b) {
 ; CHECK-LABEL: @test_sub_sd(
-; CHECK-NEXT:    [[TMP1:%.*]] = tail call <2 x double> @llvm.x86.avx512.mask.sub.sd.round(<2 x double> %a, <2 x double> %b, <2 x double> undef, i8 -1, i32 4)
-; CHECK-NEXT:    ret <2 x double> [[TMP1]]
+; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <2 x double> %a, i64 0
+; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <2 x double> %b, i64 0
+; CHECK-NEXT:    [[TMP3:%.*]] = fsub double [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <2 x double> %a, double [[TMP3]], i64 0
+; CHECK-NEXT:    ret <2 x double> [[TMP4]]
 ;
   %1 = insertelement <2 x double> %b, double 1.000000e+00, i32 1
   %2 = tail call <2 x double> @llvm.x86.avx512.mask.sub.sd.round(<2 x double> %a, <2 x double> %1, <2 x double> undef, i8 -1, i32 4)
   ret <2 x double> %2
 }
 
+define <2 x double> @test_sub_sd_round(<2 x double> %a, <2 x double> %b) {
+; CHECK-LABEL: @test_sub_sd_round(
+; CHECK-NEXT:    [[TMP1:%.*]] = tail call <2 x double> @llvm.x86.avx512.mask.sub.sd.round(<2 x double> %a, <2 x double> %b, <2 x double> undef, i8 -1, i32 8)
+; CHECK-NEXT:    ret <2 x double> [[TMP1]]
+;
+  %1 = insertelement <2 x double> %b, double 1.000000e+00, i32 1
+  %2 = tail call <2 x double> @llvm.x86.avx512.mask.sub.sd.round(<2 x double> %a, <2 x double> %1, <2 x double> undef, i8 -1, i32 8)
+  ret <2 x double> %2
+}
+
 define <2 x double> @test_sub_sd_mask(<2 x double> %a, <2 x double> %b, <2 x double> %c, i8 %mask) {
 ; CHECK-LABEL: @test_sub_sd_mask(
 ; CHECK-NEXT:    [[TMP1:%.*]] = tail call <2 x double> @llvm.x86.avx512.mask.sub.sd.round(<2 x double> %a, <2 x double> %b, <2 x double> %c, i8 %mask, i32 4)
 ; CHECK-NEXT:    ret <2 x double> [[TMP1]]
 ;
   %1 = insertelement <2 x double> %c, double 1.000000e+00, i32 1
   %2 = tail call <2 x double> @llvm.x86.avx512.mask.sub.sd.round(<2 x double> %a, <2 x double> %b, <2 x double> %1, i8 %mask, i32 4)
   ret <2 x double> %2
;
   %6 = extractelement <2 x double> %5, i32 1
   ret double %6
 }
 
 declare <4 x float> @llvm.x86.avx512.mask.mul.ss.round(<4 x float>, <4 x float>, <4 x float>, i8, i32)
 
 define <4 x float> @test_mul_ss(<4 x float> %a, <4 x float> %b) {
 ; CHECK-LABEL: @test_mul_ss(
-; CHECK-NEXT:    [[TMP1:%.*]] = tail call <4 x float> @llvm.x86.avx512.mask.mul.ss.round(<4 x float> %a, <4 x float> %b, <4 x float> undef, i8 -1, i32 4)
-; CHECK-NEXT:    ret <4 x float> [[TMP1]]
+; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <4 x float> %a, i64 0
+; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <4 x float> %b, i32 0
+; CHECK-NEXT:    [[TMP3:%.*]] = fmul float [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <4 x float> %a, float [[TMP3]], i64 0
+; CHECK-NEXT:    ret <4 x float> [[TMP4]]
 ;
   %1 = insertelement <4 x float> %b, float 1.000000e+00, i32 1
   %2 = insertelement <4 x float> %1, float 2.000000e+00, i32 2
   %3 = insertelement <4 x float> %2, float 3.000000e+00, i32 3
   %4 = tail call <4 x float> @llvm.x86.avx512.mask.mul.ss.round(<4 x float> %a, <4 x float> %3, <4 x float> undef, i8 -1, i32 4)
   ret <4 x float> %4
 }
 
+define <4 x float> @test_mul_ss_round(<4 x float> %a, <4 x float> %b) {
+; CHECK-LABEL: @test_mul_ss_round(
+; CHECK-NEXT:    [[TMP1:%.*]] = tail call <4 x float> @llvm.x86.avx512.mask.mul.ss.round(<4 x float> %a, <4 x float> %b, <4 x float> undef, i8 -1, i32 8)
+; CHECK-NEXT:    ret <4 x float> [[TMP1]]
+;
+  %1 = insertelement <4 x float> %b, float 1.000000e+00, i32 1
+  %2 = insertelement <4 x float> %1, float 2.000000e+00, i32 2
+  %3 = insertelement <4 x float> %2, float 3.000000e+00, i32 3
+  %4 = tail call <4 x float> @llvm.x86.avx512.mask.mul.ss.round(<4 x float> %a, <4 x float> %3, <4 x float> undef, i8 -1, i32 8)
+  ret <4 x float> %4
+}
+
 define <4 x float> @test_mul_ss_mask(<4 x float> %a, <4 x float> %b, <4 x float> %c, i8 %mask) {
 ; CHECK-LABEL: @test_mul_ss_mask(
 ; CHECK-NEXT:    [[TMP1:%.*]] = tail call <4 x float> @llvm.x86.avx512.mask.mul.ss.round(<4 x float> %a, <4 x float> %b, <4 x float> %c, i8 %mask, i32 4)
 ; CHECK-NEXT:    ret <4 x float> [[TMP1]]
 ;
   %1 = insertelement <4 x float> %c, float 1.000000e+00, i32 1
   %2 = insertelement <4 x float> %1, float 2.000000e+00, i32 2
   %3 = insertelement <4 x float> %2, float 3.000000e+00, i32 3
;
   %10 = extractelement <4 x float> %9, i32 1
   ret float %10
 }
 
 declare <2 x double> @llvm.x86.avx512.mask.mul.sd.round(<2 x double>, <2 x double>, <2 x double>, i8, i32)
 
 define <2 x double> @test_mul_sd(<2 x double> %a, <2 x double> %b) {
 ; CHECK-LABEL: @test_mul_sd(
-; CHECK-NEXT:    [[TMP1:%.*]] = tail call <2 x double> @llvm.x86.avx512.mask.mul.sd.round(<2 x double> %a, <2 x double> %b, <2 x double> undef, i8 -1, i32 4)
-; CHECK-NEXT:    ret <2 x double> [[TMP1]]
+; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <2 x double> %a, i64 0
+; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <2 x double> %b, i64 0
+; CHECK-NEXT:    [[TMP3:%.*]] = fmul double [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <2 x double> %a, double [[TMP3]], i64 0
+; CHECK-NEXT:    ret <2 x double> [[TMP4]]
 ;
   %1 = insertelement <2 x double> %b, double 1.000000e+00, i32 1
   %2 = tail call <2 x double> @llvm.x86.avx512.mask.mul.sd.round(<2 x double> %a, <2 x double> %1, <2 x double> undef, i8 -1, i32 4)
   ret <2 x double> %2
 }
 
+define <2 x double> @test_mul_sd_round(<2 x double> %a, <2 x double> %b) {
+; CHECK-LABEL: @test_mul_sd_round(
+; CHECK-NEXT:    [[TMP1:%.*]] = tail call <2 x double> @llvm.x86.avx512.mask.mul.sd.round(<2 x double> %a, <2 x double> %b, <2 x double> undef, i8 -1, i32 8)
+; CHECK-NEXT:    ret <2 x double> [[TMP1]]
+;
+  %1 = insertelement <2 x double> %b, double 1.000000e+00, i32 1
+  %2 = tail call <2 x double> @llvm.x86.avx512.mask.mul.sd.round(<2 x double> %a, <2 x double> %1, <2 x double> undef, i8 -1, i32 8)
+  ret <2 x double> %2
+}
+
 define <2 x double> @test_mul_sd_mask(<2 x double> %a, <2 x double> %b, <2 x double> %c, i8 %mask) {
 ; CHECK-LABEL: @test_mul_sd_mask(
 ; CHECK-NEXT:    [[TMP1:%.*]] = tail call <2 x double> @llvm.x86.avx512.mask.mul.sd.round(<2 x double> %a, <2 x double> %b, <2 x double> %c, i8 %mask, i32 4)
 ; CHECK-NEXT:    ret <2 x double> [[TMP1]]
 ;
   %1 = insertelement <2 x double> %c, double 1.000000e+00, i32 1
   %2 = tail call <2 x double> @llvm.x86.avx512.mask.mul.sd.round(<2 x double> %a, <2 x double> %b, <2 x double> %1, i8 %mask, i32 4)
   ret <2 x double> %2
;
   %6 = extractelement <2 x double> %5, i32 1
   ret double %6
 }
 
 declare <4 x float> @llvm.x86.avx512.mask.div.ss.round(<4 x float>, <4 x float>, <4 x float>, i8, i32)
 
 define <4 x float> @test_div_ss(<4 x float> %a, <4 x float> %b) {
 ; CHECK-LABEL: @test_div_ss(
-; CHECK-NEXT:    [[TMP1:%.*]] = tail call <4 x float> @llvm.x86.avx512.mask.div.ss.round(<4 x float> %a, <4 x float> %b, <4 x float> undef, i8 -1, i32 4)
-; CHECK-NEXT:    ret <4 x float> [[TMP1]]
+; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <4 x float> %a, i64 0
+; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <4 x float> %b, i32 0
+; CHECK-NEXT:    [[TMP3:%.*]] = fdiv float [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <4 x float> %a, float [[TMP3]], i64 0
+; CHECK-NEXT:    ret <4 x float> [[TMP4]]
 ;
   %1 = insertelement <4 x float> %b, float 1.000000e+00, i32 1
   %2 = insertelement <4 x float> %1, float 2.000000e+00, i32 2
   %3 = insertelement <4 x float> %2, float 3.000000e+00, i32 3
   %4 = tail call <4 x float> @llvm.x86.avx512.mask.div.ss.round(<4 x float> %a, <4 x float> %3, <4 x float> undef, i8 -1, i32 4)
   ret <4 x float> %4
 }
 
+define <4 x float> @test_div_ss_round(<4 x float> %a, <4 x float> %b) {
+; CHECK-LABEL: @test_div_ss_round(
+; CHECK-NEXT:    [[TMP1:%.*]] = tail call <4 x float> @llvm.x86.avx512.mask.div.ss.round(<4 x float> %a, <4 x float> %b, <4 x float> undef, i8 -1, i32 8)
+; CHECK-NEXT:    ret <4 x float> [[TMP1]]
+;
+  %1 = insertelement <4 x float> %b, float 1.000000e+00, i32 1
+  %2 = insertelement <4 x float> %1, float 2.000000e+00, i32 2
+  %3 = insertelement <4 x float> %2, float 3.000000e+00, i32 3
+  %4 = tail call <4 x float> @llvm.x86.avx512.mask.div.ss.round(<4 x float> %a, <4 x float> %3, <4 x float> undef, i8 -1, i32 8)
+  ret <4 x float> %4
+}
+
 define <4 x float> @test_div_ss_mask(<4 x float> %a, <4 x float> %b, <4 x float> %c, i8 %mask) {
 ; CHECK-LABEL: @test_div_ss_mask(
 ; CHECK-NEXT:    [[TMP1:%.*]] = tail call <4 x float> @llvm.x86.avx512.mask.div.ss.round(<4 x float> %a, <4 x float> %b, <4 x float> %c, i8 %mask, i32 4)
 ; CHECK-NEXT:    ret <4 x float> [[TMP1]]
 ;
   %1 = insertelement <4 x float> %c, float 1.000000e+00, i32 1
   %2 = insertelement <4 x float> %1, float 2.000000e+00, i32 2
   %3 = insertelement <4 x float> %2, float 3.000000e+00, i32 3
;
   %10 = extractelement <4 x float> %9, i32 1
   ret float %10
 }
 
 declare <2 x double> @llvm.x86.avx512.mask.div.sd.round(<2 x double>, <2 x double>, <2 x double>, i8, i32)
 
 define <2 x double> @test_div_sd(<2 x double> %a, <2 x double> %b) {
 ; CHECK-LABEL: @test_div_sd(
-; CHECK-NEXT:    [[TMP1:%.*]] = tail call <2 x double> @llvm.x86.avx512.mask.div.sd.round(<2 x double> %a, <2 x double> %b, <2 x double> undef, i8 -1, i32 4)
-; CHECK-NEXT:    ret <2 x double> [[TMP1]]
+; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <2 x double> %a, i64 0
+; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <2 x double> %b, i64 0
+; CHECK-NEXT:    [[TMP3:%.*]] = fdiv double [[TMP1]], [[TMP2]]
+; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <2 x double> %a, double [[TMP3]], i64 0
+; CHECK-NEXT:    ret <2 x double> [[TMP4]]
 ;
   %1 = insertelement <2 x double> %b, double 1.000000e+00, i32 1
   %2 = tail call <2 x double> @llvm.x86.avx512.mask.div.sd.round(<2 x double> %a, <2 x double> %1, <2 x double> undef, i8 -1, i32 4)
   ret <2 x double> %2
 }
 
+define <2 x double> @test_div_sd_round(<2 x double> %a, <2 x double> %b) {
+; CHECK-LABEL: @test_div_sd_round(
+; CHECK-NEXT:    [[TMP1:%.*]] = tail call <2 x double> @llvm.x86.avx512.mask.div.sd.round(<2 x double> %a, <2 x double> %b, <2 x double> undef, i8 -1, i32 8)
+; CHECK-NEXT:    ret <2 x double> [[TMP1]]
+;
+  %1 = insertelement <2 x double> %b, double 1.000000e+00, i32 1
+  %2 = tail call <2 x double> @llvm.x86.avx512.mask.div.sd.round(<2 x double> %a, <2 x double> %1, <2 x double> undef, i8 -1, i32 8)
+  ret <2 x double> %2
+}
+
 define <2 x double> @test_div_sd_mask(<2 x double> %a, <2 x double> %b, <2 x double> %c, i8 %mask) {
 ; CHECK-LABEL: @test_div_sd_mask(
 ; CHECK-NEXT:    [[TMP1:%.*]] = tail call <2 x double> @llvm.x86.avx512.mask.div.sd.round(<2 x double> %a, <2 x double> %b, <2 x double> %c, i8 %mask, i32 4)
 ; CHECK-NEXT:    ret <2 x double> [[TMP1]]
 ;
   %1 = insertelement <2 x double> %c, double 1.000000e+00, i32 1
   %2 = tail call <2 x double> @llvm.x86.avx512.mask.div.sd.round(<2 x double> %a, <2 x double> %b, <2 x double> %1, i8 %mask, i32 4)
   ret <2 x double> %2