Fix wrong mul type and tighted up dot() testing

Was using int dot for the float operands as it was originally an
int-only lowering function.

Author: Greg Roth

lib/HLSL/HLOperationLower.cpp

@@ -2517,12 +2517,16 @@ Value *ExpandDot(Value *arg0, Value *arg1, unsigned vecSize, hlsl::OP *hlslOP,
                  DXIL::OpCode MadOpCode = DXIL::OpCode::IMad) {
   Value *Elt0 = Builder.CreateExtractElement(arg0, (uint64_t)0);
   Value *Elt1 = Builder.CreateExtractElement(arg1, (uint64_t)0);
-  Value *Result = Builder.CreateMul(Elt0, Elt1);
+  Value *Result;
+  if (Elt0->getType()->isFloatingPointTy())
+    Result = Builder.CreateFMul(Elt0, Elt1);
+  else
+    Result = Builder.CreateMul(Elt0, Elt1);
   for (unsigned Elt = 1; Elt < vecSize; ++Elt) {
     Elt0 = Builder.CreateExtractElement(arg0, Elt);
     Elt1 = Builder.CreateExtractElement(arg1, Elt);
-    Result = TrivialDxilTrinaryOperation(MadOpCode, Elt0, Elt1, Result, hlslOP,
-                                         Builder);
+    Result =
+        TrivialDxilTrinaryOperation(MadOpCode, Elt0, Elt1, Result, OP, Builder);
   }
 
   return Result;

tools/clang/test/CodeGenDXIL/hlsl/types/longvec-scalarized-intrinsics.hlsl

@@ -104,14 +104,15 @@ float4 main(uint i : SV_PrimitiveID, bool b : B) : SV_Target {
   // CHECK: fsub fast <8 x float>
   vec1 = modf(vec1, vec2);
 
-  // CHECK: fmul fast float
-  // CHECK: call float @dx.op.tertiary.f32(i32 46, float %{{.*}}, float %{{.*}}, float %{{.*}}) ; FMad(a,b,c)
-  // CHECK: call float @dx.op.tertiary.f32(i32 46, float %{{.*}}, float %{{.*}}, float %{{.*}}) ; FMad(a,b,c)
-  // CHECK: call float @dx.op.tertiary.f32(i32 46, float %{{.*}}, float %{{.*}}, float %{{.*}}) ; FMad(a,b,c)
-  // CHECK: call float @dx.op.tertiary.f32(i32 46, float %{{.*}}, float %{{.*}}, float %{{.*}}) ; FMad(a,b,c)
-  // CHECK: call float @dx.op.tertiary.f32(i32 46, float %{{.*}}, float %{{.*}}, float %{{.*}}) ; FMad(a,b,c)
-  // CHECK: call float @dx.op.tertiary.f32(i32 46, float %{{.*}}, float %{{.*}}, float %{{.*}}) ; FMad(a,b,c)
-  // CHECK: call float @dx.op.tertiary.f32(i32 46, float %{{.*}}, float %{{.*}}, float %{{.*}}) ; FMad(a,b,c)
+  // CHECK: [[el:%.*]] = extractelement <8 x float>
+  // CHECK: [[mul:%.*]] = fmul fast float [[el]]
+  // CHECK: [[ping:%.*]] = call float @dx.op.tertiary.f32(i32 46, float %{{.*}}, float %{{.*}}, float [[mul]]) ; FMad(a,b,c)
+  // CHECK: [[pong:%.*]] = call float @dx.op.tertiary.f32(i32 46, float %{{.*}}, float %{{.*}}, float [[ping]]) ; FMad(a,b,c)
+  // CHECK: [[ping:%.*]] = call float @dx.op.tertiary.f32(i32 46, float %{{.*}}, float %{{.*}}, float [[pong]]) ; FMad(a,b,c)
+  // CHECK: [[pong:%.*]] = call float @dx.op.tertiary.f32(i32 46, float %{{.*}}, float %{{.*}}, float [[ping]]) ; FMad(a,b,c)
+  // CHECK: [[ping:%.*]] = call float @dx.op.tertiary.f32(i32 46, float %{{.*}}, float %{{.*}}, float [[pong]]) ; FMad(a,b,c)
+  // CHECK: [[pong:%.*]] = call float @dx.op.tertiary.f32(i32 46, float %{{.*}}, float %{{.*}}, float [[ping]]) ; FMad(a,b,c)
+  // CHECK: [[ping:%.*]] = call float @dx.op.tertiary.f32(i32 46, float %{{.*}}, float %{{.*}}, float [[pong]]) ; FMad(a,b,c)
   vec1 = dot(vec1, vec2);
 
   vector<bool, 8> bvec = b;