[AMDGPU] Fix codegen for inline assembly

[yaxunl]

Summary

Need to override convertConstraint to recognise amdgpu specific register names.

[arsenm]

lib/Basic/Targets/AMDGPU.h

288

parm?

[yaxunl]

Fix the comment.

[yaxunl]

Ping

[arsenm]

LGTM

lib/Basic/Targets/AMDGPU.h

bool validateAsmConstraint(const char *&Name,
     if (!S.empty())
       return false;
     // Found {vn}, {sn}, {v[n]}, {s[n]}, {v[n:m]}, or {s[n:m]}.
     Info.setAllowsRegister();
     Name = S.data() - 1;
     return true;
   }
 
+  // \p Constraint will be left pointing at the last character of
+  // the constraint.  In practice, it won't be changed unless the
+  // constraint is longer than one character.
+  std::string convertConstraint(const char *&Constraint) const override {
+    const char *Begin = Constraint;
+    TargetInfo::ConstraintInfo Info("", "");
+    if (validateAsmConstraint(Constraint, Info))
+      return std::string(Begin).substr(0, Constraint - Begin + 1);
+
+    Constraint = Begin;
+    return std::string(1, *Constraint);
+  }
+
   bool
   initFeatureMap(llvm::StringMap<bool> &Features, DiagnosticsEngine &Diags,
                  StringRef CPU,
                  const std::vector<std::string> &FeatureVec) const override;
 
   void adjustTargetOptions(const CodeGenOptions &CGOpts,
                            TargetOptions &TargetOpts) const override;
 

lib/CodeGen/CGStmt.cpp

void CodeGenFunction::EmitAsmStmt(const AsmStmt &S) {
   bool ReadOnly = true, ReadNone = true;
 
   for (unsigned i = 0, e = S.getNumOutputs(); i != e; i++) {
     TargetInfo::ConstraintInfo &Info = OutputConstraintInfos[i];
 
     // Simplify the output constraint.
     std::string OutputConstraint(S.getOutputConstraint(i));
     OutputConstraint = SimplifyConstraint(OutputConstraint.c_str() + 1,
-                                          getTarget());
+                                          getTarget(), &OutputConstraintInfos);
 
     const Expr *OutExpr = S.getOutputExpr(i);
     OutExpr = OutExpr->IgnoreParenNoopCasts(getContext());
 
     OutputConstraint = AddVariableConstraints(OutputConstraint, *OutExpr,
                                               getTarget(), CGM, S,
                                               Info.earlyClobber());
 

test/CodeGenOpenCL/inline-asm-amdgcn.cl

+// REQUIRES: amdgpu-registered-target
+// RUN: %clang_cc1 -emit-llvm -o - -triple amdgcn %s | FileCheck %s
+
+kernel void test_long(int arg0) {
+  long v15_16;
+  // CHECK: tail call i64 asm sideeffect "v_lshlrev_b64 v[15:16], 0, $0", "={v[15:16]},v"(i32 %arg0)
+  __asm volatile("v_lshlrev_b64 v[15:16], 0, %0" : "={v[15:16]}"(v15_16) : "v"(arg0));
+}

test/Sema/inline-asm-validate-amdgpu.cl

test_double(const __global double *a, const __global double *b, __global double *c, unsigned i)
     __asm("v_add_f64_e64 v[1:2], v[3:4], v[5:6]" : "={v[1:2]"(ci) : "{v[3:4]}"(ai), "{v[5:6]}"(bi) : ); //expected-error {{invalid output constraint '={v[1:2]' in asm}}
     __asm("v_add_f64_e64 v[1:2], v[3:4], v[5:6]" : "={v[1:2}"(ci) : "{v[3:4]}"(ai), "{v[5:6]}"(bi) : ); //expected-error {{invalid output constraint '={v[1:2}' in asm}}
     __asm("v_add_f64_e64 v[1:2], v[3:4], v[5:6]" : "={v[2:1]}"(ci) : "{v[3:4]}"(ai), "{v[5:6]}"(bi) : ); //expected-error {{invalid output constraint '={v[2:1]}' in asm}}
 
     __asm("v_add_f64_e64 v[1:2], v[3:4], v[5:6]" : "=v[1:2]"(ci) : "v[3:4]"(ai), "v[5:6]"(bi) : ); //expected-error {{invalid output constraint '=v[1:2]' in asm}}
 
     c[i] = ci;
 }
+
+void test_long(int arg0) {
+  long v15_16;
+  __asm volatile("v_lshlrev_b64 v[15:16], 0, %0" : "={v[15:16]}"(v15_16) : "v"(arg0));
+}