Lower neon_vmaxvq_u8,neon_vmaxvq_s8, neon_vmaxv_u8 and neon_vmaxvq_s8 (…

…#1265)

[Neon
definiton](https://developer.arm.com/architectures/instruction-sets/intrinsics/#f:@navigationhierarchiessimdisa=[Neon]&q=vmaxv_s8)
[OG
implementation](https://github.com/llvm/clangir/blob/04d7dcfb2582753f3eccbf01ec900d60297cbf4b/clang/lib/CodeGen/CGBuiltin.cpp#L13202)
Implementation in this PR is different from OG as 
1. avoided code duplication by extracting out the common pattern
2. avoided using i32 as return type of the intrinsic call, so eliminated
the need for casting result of the intrinsic call. This way of OG's
implementation is quite unnecessary IMHO, this is MAX, not ADD or MUL.
After all, using the expected type as return type of intrinsic call
produces [the same ASM code](https://godbolt.org/z/3nKG7fxPb).

clang/lib/CIR/CodeGen/CIRGenBuiltinAArch64.cpp

@@ -2351,6 +2351,26 @@ emitCommonNeonCallPattern0(CIRGenFunction &cgf, llvm::StringRef intrincsName,
   return builder.createBitcast(res, resultType);
 }
 
+/// The function `emitCommonNeonVecAcrossCall` implements a common way
+/// to implement neon intrinsic which has the following pattern:
+///  1. There is only one argument which is of vector type
+///  2. The result of the neon intrinsic is the element type of the input.
+/// This type of intrinsic usually is for across operations of the input vector.
+
+static mlir::Value emitCommonNeonVecAcrossCall(CIRGenFunction &cgf,
+                                               llvm::StringRef intrincsName,
+                                               mlir::Type eltTy,
+                                               unsigned vecLen,
+                                               const clang::CallExpr *e) {
+  CIRGenBuilderTy &builder = cgf.getBuilder();
+  mlir::Value op = cgf.emitScalarExpr(e->getArg(0));
+  cir::VectorType vTy =
+      cir::VectorType::get(&cgf.getMLIRContext(), eltTy, vecLen);
+  llvm::SmallVector<mlir::Value, 1> args{op};
+  return emitNeonCall(builder, {vTy}, args, intrincsName, eltTy,
+                      cgf.getLoc(e->getExprLoc()));
+}
+
 mlir::Value CIRGenFunction::emitCommonNeonBuiltinExpr(
     unsigned builtinID, unsigned llvmIntrinsic, unsigned altLLVMIntrinsic,
     const char *nameHint, unsigned modifier, const CallExpr *e,
@@ -4269,25 +4289,29 @@ CIRGenFunction::emitAArch64BuiltinExpr(unsigned BuiltinID, const CallExpr *E,
     llvm_unreachable("NEON::BI__builtin_neon_vaddvq_s16 NYI");
   }
   case NEON::BI__builtin_neon_vmaxv_u8: {
-    llvm_unreachable("NEON::BI__builtin_neon_vmaxv_u8 NYI");
+    return emitCommonNeonVecAcrossCall(*this, "aarch64.neon.umaxv", UInt8Ty, 8,
+                                       E);
   }
   case NEON::BI__builtin_neon_vmaxv_u16: {
     llvm_unreachable("NEON::BI__builtin_neon_vmaxv_u16 NYI");
   }
   case NEON::BI__builtin_neon_vmaxvq_u8: {
-    llvm_unreachable("NEON::BI__builtin_neon_vmaxvq_u8 NYI");
+    return emitCommonNeonVecAcrossCall(*this, "aarch64.neon.umaxv", UInt8Ty, 16,
+                                       E);
   }
   case NEON::BI__builtin_neon_vmaxvq_u16: {
     llvm_unreachable("NEON::BI__builtin_neon_vmaxvq_u16 NYI");
   }
   case NEON::BI__builtin_neon_vmaxv_s8: {
-    llvm_unreachable("NEON::BI__builtin_neon_vmaxv_s8 NYI");
+    return emitCommonNeonVecAcrossCall(*this, "aarch64.neon.smaxv", SInt8Ty, 8,
+                                       E);
   }
   case NEON::BI__builtin_neon_vmaxv_s16: {
     llvm_unreachable("NEON::BI__builtin_neon_vmaxv_s16 NYI");
   }
   case NEON::BI__builtin_neon_vmaxvq_s8: {
-    llvm_unreachable("NEON::BI__builtin_neon_vmaxvq_s8 NYI");
+    return emitCommonNeonVecAcrossCall(*this, "aarch64.neon.smaxv", SInt8Ty, 16,
+                                       E);
   }
   case NEON::BI__builtin_neon_vmaxvq_s16: {
     llvm_unreachable("NEON::BI__builtin_neon_vmaxvq_s16 NYI");

clang/test/CIR/CodeGen/AArch64/neon-misc.c

@@ -1751,3 +1751,51 @@ uint64_t test_vaddlvq_u32(uint32x4_t a) {
   // LLVM-NEXT:    [[VADDLVQ_U32_I:%.*]] = call i64 @llvm.aarch64.neon.uaddlv.i64.v4i32(<4 x i32> [[A]])
   // LLVM-NEXT:    ret i64 [[VADDLVQ_U32_I]]
 }
+
+int8_t test_vmaxv_s8(int8x8_t a) {
+  return vmaxv_s8(a);
+
+  // CIR-LABEL: vmaxv_s8
+  // CIR: cir.llvm.intrinsic "aarch64.neon.smaxv" {{%.*}} : (!cir.vector<!s8i x 8>) -> !s8i
+
+  // LLVM-LABEL: @test_vmaxv_s8
+  // LLVM-SAME: (<8 x i8> [[a:%.*]])
+  // LLVM: [[res:%.*]] = call i8 @llvm.aarch64.neon.smaxv.i8.v8i8(<8 x i8> [[a]])
+  // LLVM: ret i8 [[res]]
+}
+
+int8_t test_vmaxv_u8(uint8x8_t a) {
+  return vmaxv_u8(a);
+
+  // CIR-LABEL: vmaxv_u8
+  // CIR: cir.llvm.intrinsic "aarch64.neon.umaxv" {{%.*}} : (!cir.vector<!u8i x 8>) -> !u8i
+
+  // LLVM-LABEL: @test_vmaxv_u8
+  // LLVM-SAME: (<8 x i8> [[a:%.*]])
+  // LLVM: [[res:%.*]] = call i8 @llvm.aarch64.neon.umaxv.i8.v8i8(<8 x i8> [[a]])
+  // LLVM: ret i8 [[res]]
+}
+
+int8_t test_vmaxvq_s8(int8x16_t a) {
+  return vmaxvq_s8(a);
+
+  // CIR-LABEL: vmaxvq_s8
+  // CIR: cir.llvm.intrinsic "aarch64.neon.smaxv" {{%.*}} : (!cir.vector<!s8i x 16>) -> !s8i
+
+  // LLVM-LABEL: @test_vmaxvq_s8
+  // LLVM-SAME: (<16 x i8> [[a:%.*]])
+  // LLVM: [[res:%.*]] = call i8 @llvm.aarch64.neon.smaxv.i8.v16i8(<16 x i8> [[a]])
+  // LLVM: ret i8 [[res]]
+}
+
+int8_t test_vmaxvq_u8(uint8x16_t a) {
+  return vmaxvq_u8(a);
+
+  // CIR-LABEL: vmaxvq_u8
+  // CIR: cir.llvm.intrinsic "aarch64.neon.umaxv" {{%.*}} : (!cir.vector<!u8i x 16>) -> !u8i
+
+  // LLVM-LABEL: @test_vmaxvq_u8
+  // LLVM-SAME: (<16 x i8> [[a:%.*]])
+  // LLVM: [[res:%.*]] = call i8 @llvm.aarch64.neon.umaxv.i8.v16i8(<16 x i8> [[a]])
+  // LLVM: ret i8 [[res]]
+}