[AArch64][FastISel] Handle CRC32 intrinsics

With a similar reason as D148023; some applications make heavy use of
the CRC32 intrinsic (e.g., as part of a hash function) and therefore
benefit from avoiding frequent SelectionDAG fallbacks. In our
application, we get a 2% compile-time improvement.

Reviewed By: efriedma

Differential Revision: https://reviews.llvm.org/D148917

Author: aengelke

llvm/lib/Target/AArch64/AArch64FastISel.cpp

@@ -53,6 +53,7 @@
 #include "llvm/IR/Instructions.h"
 #include "llvm/IR/IntrinsicInst.h"
 #include "llvm/IR/Intrinsics.h"
+#include "llvm/IR/IntrinsicsAArch64.h"
 #include "llvm/IR/Operator.h"
 #include "llvm/IR/Type.h"
 #include "llvm/IR/User.h"
@@ -3776,6 +3777,57 @@ bool AArch64FastISel::fastLowerIntrinsicCall(const IntrinsicInst *II) {
     updateValueMap(II, ResultReg1, 2);
     return true;
   }
+  case Intrinsic::aarch64_crc32b:
+  case Intrinsic::aarch64_crc32h:
+  case Intrinsic::aarch64_crc32w:
+  case Intrinsic::aarch64_crc32x:
+  case Intrinsic::aarch64_crc32cb:
+  case Intrinsic::aarch64_crc32ch:
+  case Intrinsic::aarch64_crc32cw:
+  case Intrinsic::aarch64_crc32cx: {
+    if (!Subtarget->hasCRC())
+      return false;
+
+    unsigned Opc;
+    switch (II->getIntrinsicID()) {
+    default:
+      llvm_unreachable("Unexpected intrinsic!");
+    case Intrinsic::aarch64_crc32b:
+      Opc = AArch64::CRC32Brr;
+      break;
+    case Intrinsic::aarch64_crc32h:
+      Opc = AArch64::CRC32Hrr;
+      break;
+    case Intrinsic::aarch64_crc32w:
+      Opc = AArch64::CRC32Wrr;
+      break;
+    case Intrinsic::aarch64_crc32x:
+      Opc = AArch64::CRC32Xrr;
+      break;
+    case Intrinsic::aarch64_crc32cb:
+      Opc = AArch64::CRC32CBrr;
+      break;
+    case Intrinsic::aarch64_crc32ch:
+      Opc = AArch64::CRC32CHrr;
+      break;
+    case Intrinsic::aarch64_crc32cw:
+      Opc = AArch64::CRC32CWrr;
+      break;
+    case Intrinsic::aarch64_crc32cx:
+      Opc = AArch64::CRC32CXrr;
+      break;
+    }
+
+    Register LHSReg = getRegForValue(II->getArgOperand(0));
+    Register RHSReg = getRegForValue(II->getArgOperand(1));
+    if (!LHSReg || !RHSReg)
+      return false;
+
+    Register ResultReg =
+        fastEmitInst_rr(Opc, &AArch64::GPR32RegClass, LHSReg, RHSReg);
+    updateValueMap(II, ResultReg);
+    return true;
+  }
   }
   return false;
 }

llvm/test/CodeGen/AArch64/arm64-fast-isel-crc32.ll

@@ -0,0 +1,71 @@
+; RUN: llc -mtriple=arm64-eabi -fast-isel -fast-isel-abort=3 -mattr=+crc -o - %s | FileCheck %s
+; RUN: llc -mtriple=arm64-eabi -fast-isel -fast-isel-abort=3 -mattr=+v8r -o - %s | FileCheck %s
+; RUN: llc -mtriple=arm64-eabi -fast-isel -fast-isel-abort=3 -mcpu=cortex-a53 -mattr=+crc -o - %s | FileCheck %s
+
+; Note: tests are a copy of arm64-crc32.ll
+
+define i32 @test_crc32b(i32 %cur, i32 %next) {
+; CHECK-LABEL: test_crc32b:
+; CHECK: crc32b w0, w0, w1
+  %val = call i32 @llvm.aarch64.crc32b(i32 %cur, i32 %next)
+  ret i32 %val
+}
+
+define i32 @test_crc32h(i32 %cur, i32 %next) {
+; CHECK-LABEL: test_crc32h:
+; CHECK: crc32h w0, w0, w1
+  %val = call i32 @llvm.aarch64.crc32h(i32 %cur, i32 %next)
+  ret i32 %val
+}
+
+define i32 @test_crc32w(i32 %cur, i32 %next) {
+; CHECK-LABEL: test_crc32w:
+; CHECK: crc32w w0, w0, w1
+  %val = call i32 @llvm.aarch64.crc32w(i32 %cur, i32 %next)
+  ret i32 %val
+}
+
+define i32 @test_crc32x(i32 %cur, i64 %next) {
+; CHECK-LABEL: test_crc32x:
+; CHECK: crc32x w0, w0, x1
+  %val = call i32 @llvm.aarch64.crc32x(i32 %cur, i64 %next)
+  ret i32 %val
+}
+
+define i32 @test_crc32cb(i32 %cur, i32 %next) {
+; CHECK-LABEL: test_crc32cb:
+; CHECK: crc32cb w0, w0, w1
+  %val = call i32 @llvm.aarch64.crc32cb(i32 %cur, i32 %next)
+  ret i32 %val
+}
+
+define i32 @test_crc32ch(i32 %cur, i32 %next) {
+; CHECK-LABEL: test_crc32ch:
+; CHECK: crc32ch w0, w0, w1
+  %val = call i32 @llvm.aarch64.crc32ch(i32 %cur, i32 %next)
+  ret i32 %val
+}
+
+define i32 @test_crc32cw(i32 %cur, i32 %next) {
+; CHECK-LABEL: test_crc32cw:
+; CHECK: crc32cw w0, w0, w1
+  %val = call i32 @llvm.aarch64.crc32cw(i32 %cur, i32 %next)
+  ret i32 %val
+}
+
+define i32 @test_crc32cx(i32 %cur, i64 %next) {
+; CHECK-LABEL: test_crc32cx:
+; CHECK: crc32cx w0, w0, x1
+  %val = call i32 @llvm.aarch64.crc32cx(i32 %cur, i64 %next)
+  ret i32 %val
+}
+
+declare i32 @llvm.aarch64.crc32b(i32, i32)
+declare i32 @llvm.aarch64.crc32h(i32, i32)
+declare i32 @llvm.aarch64.crc32w(i32, i32)
+declare i32 @llvm.aarch64.crc32x(i32, i64)
+
+declare i32 @llvm.aarch64.crc32cb(i32, i32)
+declare i32 @llvm.aarch64.crc32ch(i32, i32)
+declare i32 @llvm.aarch64.crc32cw(i32, i32)
+declare i32 @llvm.aarch64.crc32cx(i32, i64)