[clang] 696becb - [Matrix] Remove bitcast when casting between matrices of the same size

[Florian Hahn via cfe-commits]

Author: Saurabh Jha
Date: 2021-05-03T15:31:43+01:00
New Revision: 696becbd139e4bbbe304ce2231de0fb50f307b68

URL: https://github.com/llvm/llvm-project/commit/696becbd139e4bbbe304ce2231de0fb50f307b68
DIFF: https://github.com/llvm/llvm-project/commit/696becbd139e4bbbe304ce2231de0fb50f307b68.diff

LOG: [Matrix] Remove bitcast when casting between matrices of the same size

In matrix type casts, we were doing bitcast when the matrices had the same size. This was incorrect and this patch fixes that.
Also added some new CodeGen tests for signed <-> usigned conversions

Reviewed By: fhahn

Differential Revision: https://reviews.llvm.org/D101754

Added: 
    

Modified: 
    clang/lib/CodeGen/CGExprScalar.cpp
    clang/test/CodeGen/matrix-cast.c

Removed: 
    


################################################################################
diff  --git a/clang/lib/CodeGen/CGExprScalar.cpp b/clang/lib/CodeGen/CGExprScalar.cpp
index b469a4454f158..89ab9abc7dadb 100644
--- a/clang/lib/CodeGen/CGExprScalar.cpp
+++ b/clang/lib/CodeGen/CGExprScalar.cpp
@@ -1205,10 +1205,6 @@ Value *ScalarExprEmitter::EmitScalarCast(Value *Src, QualType SrcType,
   QualType SrcElementType;
   QualType DstElementType;
   if (SrcType->isMatrixType() && DstType->isMatrixType()) {
-    // Allow bitcast between matrixes of the same size.
-    if (SrcTy->getPrimitiveSizeInBits() == DstTy->getPrimitiveSizeInBits())
-      return Builder.CreateBitCast(Src, DstTy, "conv");
-
     SrcElementTy = cast<llvm::VectorType>(SrcTy)->getElementType();
     DstElementTy = cast<llvm::VectorType>(DstTy)->getElementType();
     SrcElementType = SrcType->castAs<MatrixType>()->getElementType();

diff  --git a/clang/test/CodeGen/matrix-cast.c b/clang/test/CodeGen/matrix-cast.c
index 8871232f76b11..ad8f67acacdfb 100644
--- a/clang/test/CodeGen/matrix-cast.c
+++ b/clang/test/CodeGen/matrix-cast.c
@@ -52,13 +52,23 @@ void cast_int_matrix_to_short(ix5x5 i, sx5x5 s) {
 void cast_int_matrix_to_float(ix5x5 i, fx5x5 f) {
   // CHECK-LABEL: define{{.*}} void @cast_int_matrix_to_float(<25 x i32> %i, <25 x float> %f)
   // CHECK:       [[I:%.*]] = load <25 x i32>, <25 x i32>* {{.*}}, align 4
-  // CHECK-NEXT:  [[CONV:%.*]] = bitcast <25 x i32> [[I]] to <25 x float>
+  // CHECK-NEXT:  [[CONV:%.*]] = sitofp <25 x i32> [[I]] to <25 x float>
   // CHECK-NEXT:  store <25 x float> [[CONV]], <25 x float>* {{.*}}, align 4
   // CHECK-NEXT:  ret void
 
   f = (fx5x5)i;
 }
 
+void cast_unsigned_int_matrix_to_float(unsigned_short_int_5x5 u, fx5x5 f) {
+  // CHECK-LABEL: define{{.*}} void @cast_unsigned_int_matrix_to_float(<25 x i16> %u, <25 x float> %f)
+  // CHECK:       [[U:%.*]] = load <25 x i16>, <25 x i16>* {{.*}}, align 2
+  // CHECK-NEXT:  [[CONV:%.*]] = uitofp <25 x i16> [[U]] to <25 x float>
+  // CHECK-NEXT:  store <25 x float> [[CONV]], <25 x float>* {{.*}}, align 4
+  // CHECK-NEXT:  ret void
+
+  f = (fx5x5)u;
+}
+
 void cast_double_matrix_to_int(dx5x5 d, ix5x5 i) {
   // CHECK-LABEL: define{{.*}} void @cast_double_matrix_to_int(<25 x double> %d, <25 x i32> %i)
   // CHECK:       [[D:%.*]] = load <25 x double>, <25 x double>* {{.*}}, align 8
@@ -108,3 +118,23 @@ void cast_unsigned_long_int_to_unsigned_short_int(unsigned_long_int_5x5 l, unsig
 
   s = (unsigned_short_int_5x5)l;
 }
+
+void cast_unsigned_short_int_to_int(unsigned_short_int_5x5 u, ix5x5 i) {
+  // CHECK-LABEL: define{{.*}} void @cast_unsigned_short_int_to_int(<25 x i16> %u, <25 x i32> %i)
+  // CHECK:       [[U:%.*]] = load <25 x i16>, <25 x i16>* %0, align 2
+  // CHECK-NEXT:  [[CONV:%.*]] = zext <25 x i16> [[U]] to <25 x i32>
+  // CHECK-NEXT:  store <25 x i32> [[CONV]], <25 x i32>* {{.*}}, align 4
+  // CHECK-NEXT:  ret void
+
+  i = (ix5x5)u;
+}
+
+void cast_int_to_unsigned_long_int(ix5x5 i, unsigned_long_int_5x5 u) {
+  // CHECK-LABEL: define{{.*}} void @cast_int_to_unsigned_long_int(<25 x i32> %i, <25 x i64> %u)
+  // CHECK:       [[I:%.*]] = load <25 x i32>, <25 x i32>* %0, align 4
+  // CHECK-NEXT:  [[CONV:%.*]] = sext <25 x i32> [[I]] to <25 x i64>
+  // CHECK-NEXT:  store <25 x i64> [[CONV]], <25 x i64>* {{.*}}, align 8
+  // CHECK-NEXT:  ret void
+
+  u = (unsigned_long_int_5x5)i;
+}