[CodeGen] Allow mixed scalar type constraints for inline asm (#65465)

GCC supports code like "asm volatile ("" : "=r" (i) : "0" (f))" where i
is integer type and f is floating point type. Currently this code
produces an error with Clang. The change allows mixed scalar types
between input and output constraints.

Co-authored-by: Matt Arsenault <[email protected]>

Author: dfszabo

llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp

@@ -9591,9 +9591,11 @@ static void patchMatchingInput(const SDISelAsmOperandInfo &OpInfo,
   std::pair<unsigned, const TargetRegisterClass *> InputRC =
       TLI.getRegForInlineAsmConstraint(TRI, MatchingOpInfo.ConstraintCode,
                                        MatchingOpInfo.ConstraintVT);
-  if ((OpInfo.ConstraintVT.isInteger() !=
-       MatchingOpInfo.ConstraintVT.isInteger()) ||
-      (MatchRC.second != InputRC.second)) {
+  const bool OutOpIsIntOrFP =
+      OpInfo.ConstraintVT.isInteger() || OpInfo.ConstraintVT.isFloatingPoint();
+  const bool InOpIsIntOrFP = MatchingOpInfo.ConstraintVT.isInteger() ||
+                             MatchingOpInfo.ConstraintVT.isFloatingPoint();
+  if ((OutOpIsIntOrFP != InOpIsIntOrFP) || (MatchRC.second != InputRC.second)) {
     // FIXME: error out in a more elegant fashion
     report_fatal_error("Unsupported asm: input constraint"
                        " with a matching output constraint of"

llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp

@@ -5856,8 +5856,11 @@ TargetLowering::ParseConstraints(const DataLayout &DL,
         std::pair<unsigned, const TargetRegisterClass *> InputRC =
             getRegForInlineAsmConstraint(TRI, Input.ConstraintCode,
                                          Input.ConstraintVT);
-        if ((OpInfo.ConstraintVT.isInteger() !=
-             Input.ConstraintVT.isInteger()) ||
+        const bool OutOpIsIntOrFP = OpInfo.ConstraintVT.isInteger() ||
+                                    OpInfo.ConstraintVT.isFloatingPoint();
+        const bool InOpIsIntOrFP = Input.ConstraintVT.isInteger() ||
+                                   Input.ConstraintVT.isFloatingPoint();
+        if ((OutOpIsIntOrFP != InOpIsIntOrFP) ||
             (MatchRC.second != InputRC.second)) {
           report_fatal_error("Unsupported asm: input constraint"
                              " with a matching output constraint of"

llvm/test/CodeGen/X86/inline-asm-int-to-fp.ll

@@ -0,0 +1,61 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4
+; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr +avx < %s | FileCheck %s
+
+; The C source used as a base for generating this test:.
+
+; unsigned test(float f)
+; {
+;    unsigned i;
+;    // Copies f into the output operand i
+;    asm volatile ("" : "=r" (i) : "0" (f));
+;    return i;
+; }
+
+
+define i32 @test_int_float(float %f) {
+; CHECK-LABEL: test_int_float:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vmovd %xmm0, %eax
+; CHECK-NEXT:    #APP
+; CHECK-NEXT:    #NO_APP
+; CHECK-NEXT:    retq
+entry:
+  %asm_call = call i32 asm sideeffect "", "=r,0,~{dirflag},~{fpsr},~{flags}"(float %f)
+  ret i32 %asm_call
+}
+
+define i32 @test_int_ptr(ptr %f) {
+; CHECK-LABEL: test_int_ptr:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    movq %rdi, %rax
+; CHECK-NEXT:    #APP
+; CHECK-NEXT:    #NO_APP
+; CHECK-NEXT:    # kill: def $eax killed $eax killed $rax
+; CHECK-NEXT:    retq
+entry:
+  %asm_call = call i32 asm sideeffect "", "=r,0,~{dirflag},~{fpsr},~{flags}"(ptr %f)
+  ret i32 %asm_call
+}
+
+define i64 @test_int_vec(<4 x i16> %v) {
+; CHECK-LABEL: test_int_vec:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    #APP
+; CHECK-NEXT:    #NO_APP
+; CHECK-NEXT:    vmovq %xmm0, %rax
+; CHECK-NEXT:    retq
+entry:
+  %asm_call = call i64 asm sideeffect "", "=v,0,~{dirflag},~{fpsr},~{flags}"(<4 x i16> %v)
+  ret i64 %asm_call
+}
+
+define <4 x i32> @test_int_vec_float_vec(<4 x float> %f) {
+; CHECK-LABEL: test_int_vec_float_vec:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    #APP
+; CHECK-NEXT:    #NO_APP
+; CHECK-NEXT:    retq
+entry:
+  %asm_call = call <4 x i32> asm sideeffect "", "=v,0,~{dirflag},~{fpsr},~{flags}"(<4 x float> %f)
+  ret <4 x i32> %asm_call
+}