[CIR] Upstream CIR codegen for undef x86 builtins

clang/lib/CIR/CodeGen/CIRGenBuiltinX86.cpp

@@ -93,13 +93,20 @@ mlir::Value CIRGenFunction::emitX86BuiltinExpr(unsigned builtinID,
   case X86::BI__builtin_ia32_tzcnt_u16:
   case X86::BI__builtin_ia32_tzcnt_u32:
   case X86::BI__builtin_ia32_tzcnt_u64:
-  case X86::BI__builtin_ia32_undef128:
-  case X86::BI__builtin_ia32_undef256:
-  case X86::BI__builtin_ia32_undef512:
     cgm.errorNYI(e->getSourceRange(),
                  std::string("unimplemented X86 builtin call: ") +
                      getContext().BuiltinInfo.getName(builtinID));
     return {};
+  case X86::BI__builtin_ia32_undef128:
+  case X86::BI__builtin_ia32_undef256:
+  case X86::BI__builtin_ia32_undef512:
+    // The x86 definition of "undef" is not the same as the LLVM definition
+    // (PR32176). We leave optimizing away an unnecessary zero constant to the
+    // IR optimizer and backend.
+    // TODO: If we had a "freeze" IR instruction to generate a fixed undef
+    //  value, we should use that here instead of a zero.
+    return builder.getNullValue(convertType(e->getType()),
+                                getLoc(e->getExprLoc()));
   case X86::BI__builtin_ia32_vec_ext_v4hi:
   case X86::BI__builtin_ia32_vec_ext_v16qi:
   case X86::BI__builtin_ia32_vec_ext_v8hi:

clang/test/CIR/CodeGen/X86/sse2-builtins.c

@@ -16,6 +16,43 @@
 
 #include <immintrin.h>
 
+__m128d test_mm_undefined_pd(void) {
+  // CIR-LABEL: _mm_undefined_pd
+  // CIR: %{{.*}} = cir.const #cir.zero : !cir.vector<2 x !cir.double>
+  // CIR: cir.return %{{.*}} : !cir.vector<2 x !cir.double>
+
+  // CIR-LABEL: cir.func {{.*}}test_mm_undefined_pd
+  // CIR: call @_mm_undefined_pd
+
+  // LLVM-LABEL: test_mm_undefined_pd
+  // LLVM: store <2 x double> zeroinitializer, ptr %[[A:.*]], align 16
+  // LLVM: %{{.*}} = load <2 x double>, ptr %[[A]], align 16
+  // LLVM: ret <2 x double> %{{.*}}
+
+  // OGCG-LABEL: test_mm_undefined_pd
+  // OGCG: ret <2 x double> zeroinitializer
+  return _mm_undefined_pd();
+}
+
+__m128i test_mm_undefined_si128(void) {
+  // CIR-LABEL: _mm_undefined_si128
+  // CIR: %[[A:.*]] = cir.const #cir.zero : !cir.vector<2 x !cir.double>
+  // CIR: %{{.*}} = cir.cast bitcast %[[A]] : !cir.vector<2 x !cir.double> ->
+  // CIR: cir.return %{{.*}} :
+
+  // CIR-LABEL: cir.func {{.*}}test_mm_undefined_si128
+  // CIR: call @_mm_undefined_si128
+
+  // LLVM-LABEL: test_mm_undefined_si128
+  // LLVM: store <2 x i64> zeroinitializer, ptr %[[A:.*]], align 16
+  // LLVM: %{{.*}} = load <2 x i64>, ptr %[[A]], align 16
+  // LLVM: ret <2 x i64> %{{.*}}
+
+  // OGCG-LABEL: test_mm_undefined_si128
+  // OGCG: ret <2 x i64> zeroinitializer
+  return _mm_undefined_si128();
+}
+
 // Lowering to pextrw requires optimization.
 int test_mm_extract_epi16(__m128i A) {
   // CIR-LABEL: test_mm_extract_epi16