From: Sanjay Patel <spatel@rotateright.com>
Date: Thu, 10 Aug 2017 15:37:26 +0000 (+0000)
Subject: [InstCombine] add memcpy expansion tests with potential DL dependency; NFC
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=0fd98222cefa4258dd40d3981a2f8a4dc8108cb5;p=llvm

[InstCombine] add memcpy expansion tests with potential DL dependency; NFC

Current behavior is to transform these independently of the datalayout.

There's a proposal to change this in D35035:
https://reviews.llvm.org/D35035


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@310611 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/test/Transforms/InstCombine/memcpy-to-load.ll b/test/Transforms/InstCombine/memcpy-to-load.ll
index 7638df68b90..2b66552e4f7 100644
--- a/test/Transforms/InstCombine/memcpy-to-load.ll
+++ b/test/Transforms/InstCombine/memcpy-to-load.ll
@@ -1,20 +1,87 @@
-; RUN: opt < %s -instcombine -S | FileCheck %s
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt < %s -instcombine -S                         | FileCheck %s --check-prefix=ALL --check-prefix=NODL
+; RUN: opt < %s -instcombine -S -data-layout=n32        | FileCheck %s --check-prefix=ALL --check-prefix=I32
+; RUN: opt < %s -instcombine -S -data-layout=n32:64     | FileCheck %s --check-prefix=ALL --check-prefix=I64
+; RUN: opt < %s -instcombine -S -data-layout=n32:64:128 | FileCheck %s --check-prefix=ALL --check-prefix=I128
 
-target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128"
-target triple = "i686-apple-darwin8"
+declare void @llvm.memcpy.p0i8.p0i8.i32(i8* nocapture, i8* nocapture, i32, i32, i1) nounwind
+
+; memcpy can be expanded inline with load/store (based on the datalayout?)
+
+define void @copy_1_byte(i8* %d, i8* %s) {
+; ALL-LABEL: @copy_1_byte(
+; ALL-NEXT:    [[TMP1:%.*]] = load i8, i8* [[S:%.*]], align 1
+; ALL-NEXT:    store i8 [[TMP1]], i8* [[D:%.*]], align 1
+; ALL-NEXT:    ret void
+;
+  call void @llvm.memcpy.p0i8.p0i8.i32(i8* %d, i8* %s, i32 1, i32 1, i1 false)
+  ret void
+}
+
+define void @copy_2_bytes(i8* %d, i8* %s) {
+; ALL-LABEL: @copy_2_bytes(
+; ALL-NEXT:    [[TMP1:%.*]] = bitcast i8* [[S:%.*]] to i16*
+; ALL-NEXT:    [[TMP2:%.*]] = bitcast i8* [[D:%.*]] to i16*
+; ALL-NEXT:    [[TMP3:%.*]] = load i16, i16* [[TMP1]], align 1
+; ALL-NEXT:    store i16 [[TMP3]], i16* [[TMP2]], align 1
+; ALL-NEXT:    ret void
+;
+  call void @llvm.memcpy.p0i8.p0i8.i32(i8* %d, i8* %s, i32 2, i32 1, i1 false)
+  ret void
+}
+
+; We don't expand small non-power-of-2. Should we? Might be a target-dependent choice.
+
+define void @copy_3_bytes(i8* %d, i8* %s) {
+; ALL-LABEL: @copy_3_bytes(
+; ALL-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* [[D:%.*]], i8* [[S:%.*]], i32 3, i32 1, i1 false)
+; ALL-NEXT:    ret void
+;
+  call void @llvm.memcpy.p0i8.p0i8.i32(i8* %d, i8* %s, i32 3, i32 1, i1 false)
+  ret void
+}
 
-; Make sure that the memcpy has been replaced with a load/store of i64.
+define void @copy_4_bytes(i8* %d, i8* %s) {
+; ALL-LABEL: @copy_4_bytes(
+; ALL-NEXT:    [[TMP1:%.*]] = bitcast i8* [[S:%.*]] to i32*
+; ALL-NEXT:    [[TMP2:%.*]] = bitcast i8* [[D:%.*]] to i32*
+; ALL-NEXT:    [[TMP3:%.*]] = load i32, i32* [[TMP1]], align 1
+; ALL-NEXT:    store i32 [[TMP3]], i32* [[TMP2]], align 1
+; ALL-NEXT:    ret void
+;
+  call void @llvm.memcpy.p0i8.p0i8.i32(i8* %d, i8* %s, i32 4, i32 1, i1 false)
+  ret void
+}
+
+; We don't expand small non-power-of-2. Should we? Might be a target-dependent choice.
+
+define void @copy_5_bytes(i8* %d, i8* %s) {
+; ALL-LABEL: @copy_5_bytes(
+; ALL-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* [[D:%.*]], i8* [[S:%.*]], i32 5, i32 1, i1 false)
+; ALL-NEXT:    ret void
+;
+  call void @llvm.memcpy.p0i8.p0i8.i32(i8* %d, i8* %s, i32 5, i32 1, i1 false)
+  ret void
+}
 
-define void @foo(i8* %d, i8* %s) {
-; CHECK-LABEL: @foo(
-; CHECK-NEXT:    [[TMP1:%.*]] = bitcast i8* %s to i64*
-; CHECK-NEXT:    [[TMP2:%.*]] = bitcast i8* %d to i64*
-; CHECK-NEXT:    [[TMP3:%.*]] = load i64, i64* [[TMP1]], align 1
-; CHECK-NEXT:    store i64 [[TMP3]], i64* [[TMP2]], align 1
-; CHECK-NEXT:    ret void
+define void @copy_8_bytes(i8* %d, i8* %s) {
+; ALL-LABEL: @copy_8_bytes(
+; ALL-NEXT:    [[TMP1:%.*]] = bitcast i8* [[S:%.*]] to i64*
+; ALL-NEXT:    [[TMP2:%.*]] = bitcast i8* [[D:%.*]] to i64*
+; ALL-NEXT:    [[TMP3:%.*]] = load i64, i64* [[TMP1]], align 1
+; ALL-NEXT:    store i64 [[TMP3]], i64* [[TMP2]], align 1
+; ALL-NEXT:    ret void
 ;
   call void @llvm.memcpy.p0i8.p0i8.i32(i8* %d, i8* %s, i32 8, i32 1, i1 false)
   ret void
 }
 
-declare void @llvm.memcpy.p0i8.p0i8.i32(i8* nocapture, i8* nocapture, i32, i32, i1) nounwind
+define void @copy_16_bytes(i8* %d, i8* %s) {
+; ALL-LABEL: @copy_16_bytes(
+; ALL-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i32(i8* [[D:%.*]], i8* [[S:%.*]], i32 16, i32 1, i1 false)
+; ALL-NEXT:    ret void
+;
+  call void @llvm.memcpy.p0i8.p0i8.i32(i8* %d, i8* %s, i32 16, i32 1, i1 false)
+  ret void
+}
+