Upstream CIR codegen for undef x86 builtins

Author: Thibault-Monnier

clang/lib/CIR/CodeGen/CIRGenBuiltinX86.cpp

@@ -16,7 +16,6 @@
 #include "clang/Basic/Builtins.h"
 #include "clang/Basic/TargetBuiltins.h"
 #include "clang/CIR/MissingFeatures.h"
-#include "llvm/IR/IntrinsicsX86.h"
 
 using namespace clang;
 using namespace clang::CIRGen;
@@ -60,9 +59,20 @@ 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:
+    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

@@ -0,0 +1,39 @@
+// RUN: %clang_cc1 -x c -flax-vector-conversions=none -ffreestanding %s -triple=x86_64-unknown-linux -target-feature +sse2 -fclangir -emit-cir -o %t.cir -Wall -Werror
+// RUN: FileCheck --check-prefixes=CIR-CHECK,CIR-X64 --input-file=%t.cir %s
+// RUN: %clang_cc1 -x c -flax-vector-conversions=none -ffreestanding %s -triple=x86_64-unknown-linux -target-feature +sse2 -fno-signed-char -fclangir -emit-cir -o %t.cir -Wall -Werror
+// RUN: FileCheck --check-prefixes=CIR-CHECK,CIR-X64 --input-file=%t.cir %s
+
+// RUN: %clang_cc1 -x c++ -flax-vector-conversions=none -ffreestanding %s -triple=x86_64-unknown-linux -target-feature +sse2 -fclangir -emit-llvm -o %t.ll -Wall -Werror
+// RUN: FileCheck --check-prefixes=LLVM-CHECK,LLVM-X64 --input-file=%t.ll %s
+// RUN: %clang_cc1 -x c++ -flax-vector-conversions=none -ffreestanding %s -triple=x86_64-unknown-linux -target-feature +sse2 -fno-signed-char -fclangir -emit-llvm -o %t.ll -Wall -Werror
+// RUN: FileCheck --check-prefixes=LLVM-CHECK,LLVM-X64 --input-file=%t.ll %s
+
+// This test mimics clang/test/CodeGen/X86/sse2-builtins.c, which eventually
+// CIR shall be able to support fully.
+
+#include <immintrin.h>
+
+__m128d test_mm_undefined_pd(void) {
+  // CIR-X64-LABEL: _mm_undefined_pd
+  // CIR-X64: %{{.*}} = cir.const #cir.zero : !cir.vector<2 x !cir.double>
+  // CIR-X64: cir.return %{{.*}} : !cir.vector<2 x !cir.double>
+
+  // LLVM-X64-LABEL: test_mm_undefined_pd
+  // LLVM-X64: store <2 x double> zeroinitializer, ptr %[[A:.*]], align 16
+  // LLVM-X64: %{{.*}} = load <2 x double>, ptr %[[A]], align 16
+  // LLVM-X64: ret <2 x double> %{{.*}}
+  return _mm_undefined_pd();
+}
+
+__m128i test_mm_undefined_si128(void) {
+  // CIR-LABEL: _mm_undefined_si128
+  // CIR-CHECK: %[[A:.*]] = cir.const #cir.zero : !cir.vector<2 x !cir.double>
+  // CIR-CHECK: %{{.*}} = cir.cast bitcast %[[A]] : !cir.vector<2 x !cir.double> -> !cir.vector<2 x !s64i>
+  // CIR-CHECK: cir.return %{{.*}} : !cir.vector<2 x !s64i>
+
+  // LLVM-CHECK-LABEL: test_mm_undefined_si128
+  // LLVM-CHECK: store <2 x i64> zeroinitializer, ptr %[[A:.*]], align 16
+  // LLVM-CHECK: %{{.*}} = load <2 x i64>, ptr %[[A]], align 16
+  // LLVM-CHECK: ret <2 x i64> %{{.*}}
+  return _mm_undefined_si128();
+}