Merge load/store sequences with adresses: base + index + offset

We would also like to merge sequences that involve a variable index like in the
example below.

    int index = *idx++
    int i0 = c[index+0];
    int i1 = c[index+1];
    b[0] = i0;
    b[1] = i1;

By extending the parsing of the base pointer to handle dags that contain a
base, index, and offset we can handle examples like the one above.

The dag for the code above will look something like:

 (load (i64 add (i64 copyfromreg %c)
                (i64 signextend (i8 load %index))))

 (load (i64 add (i64 copyfromreg %c)
                (i64 signextend (i32 add (i32 signextend (i8 load %index))
                                         (i32 1)))))

The code that parses the tree ignores the intermediate sign extensions. However,
if there is a sign extension it needs to be on all indexes.

 (load (i64 add (i64 copyfromreg %c)
                (i64 signextend (add (i8 load %index)
                                     (i8 1))))
 vs

 (load (i64 add (i64 copyfromreg %c)
                (i64 signextend (i32 add (i32 signextend (i8 load %index))
                                         (i32 1)))))
radar://13536387

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@178483 91177308-0d34-0410-b5e6-96231b3b80d8

1 files changed, 96 insertions, 0 deletions

diff --git a/test/CodeGen/X86/MergeConsecutiveStores.ll b/test/CodeGen/X86/MergeConsecutiveStores.ll
index fbe8879ad6..bb227a0185 100644
--- a/test/CodeGen/X86/MergeConsecutiveStores.ll
+++ b/test/CodeGen/X86/MergeConsecutiveStores.ll
@@ -337,3 +337,99 @@ block4:                                       ; preds = %4, %.lr.ph
   ret void
 }
 
+; Make sure that we merge the consecutive load/store sequence below and use a
+; word (16 bit) instead of a byte copy.
+; CHECK: MergeLoadStoreBaseIndexOffset
+; CHECK: movw    (%{{.*}},%{{.*}}), [[REG:%[a-z]+]]
+; CHECK: movw    [[REG]], (%{{.*}})
+define void @MergeLoadStoreBaseIndexOffset(i64* %a, i8* %b, i8* %c, i32 %n) {
+  br label %1
+
+; <label>:1
+  %.09 = phi i32 [ %n, %0 ], [ %11, %1 ]
+  %.08 = phi i8* [ %b, %0 ], [ %10, %1 ]
+  %.0 = phi i64* [ %a, %0 ], [ %2, %1 ]
+  %2 = getelementptr inbounds i64* %.0, i64 1
+  %3 = load i64* %.0, align 1
+  %4 = getelementptr inbounds i8* %c, i64 %3
+  %5 = load i8* %4, align 1
+  %6 = add i64 %3, 1
+  %7 = getelementptr inbounds i8* %c, i64 %6
+  %8 = load i8* %7, align 1
+  store i8 %5, i8* %.08, align 1
+  %9 = getelementptr inbounds i8* %.08, i64 1
+  store i8 %8, i8* %9, align 1
+  %10 = getelementptr inbounds i8* %.08, i64 2
+  %11 = add nsw i32 %.09, -1
+  %12 = icmp eq i32 %11, 0
+  br i1 %12, label %13, label %1
+
+; <label>:13
+  ret void
+}
+
+; Make sure that we merge the consecutive load/store sequence below and use a
+; word (16 bit) instead of a byte copy even if there are intermediate sign
+; extensions.
+; CHECK: MergeLoadStoreBaseIndexOffsetSext
+; CHECK: movw    (%{{.*}},%{{.*}}), [[REG:%[a-z]+]]
+; CHECK: movw    [[REG]], (%{{.*}})
+define void @MergeLoadStoreBaseIndexOffsetSext(i8* %a, i8* %b, i8* %c, i32 %n) {
+  br label %1
+
+; <label>:1
+  %.09 = phi i32 [ %n, %0 ], [ %12, %1 ]
+  %.08 = phi i8* [ %b, %0 ], [ %11, %1 ]
+  %.0 = phi i8* [ %a, %0 ], [ %2, %1 ]
+  %2 = getelementptr inbounds i8* %.0, i64 1
+  %3 = load i8* %.0, align 1
+  %4 = sext i8 %3 to i64
+  %5 = getelementptr inbounds i8* %c, i64 %4
+  %6 = load i8* %5, align 1
+  %7 = add i64 %4, 1
+  %8 = getelementptr inbounds i8* %c, i64 %7
+  %9 = load i8* %8, align 1
+  store i8 %6, i8* %.08, align 1
+  %10 = getelementptr inbounds i8* %.08, i64 1
+  store i8 %9, i8* %10, align 1
+  %11 = getelementptr inbounds i8* %.08, i64 2
+  %12 = add nsw i32 %.09, -1
+  %13 = icmp eq i32 %12, 0
+  br i1 %13, label %14, label %1
+
+; <label>:14
+  ret void
+}
+
+; However, we can only merge ignore sign extensions when they are on all memory
+; computations;
+; CHECK: loadStoreBaseIndexOffsetSextNoSex
+; CHECK-NOT: movw    (%{{.*}},%{{.*}}), [[REG:%[a-z]+]]
+; CHECK-NOT: movw    [[REG]], (%{{.*}})
+define void @loadStoreBaseIndexOffsetSextNoSex(i8* %a, i8* %b, i8* %c, i32 %n) {
+  br label %1
+
+; <label>:1
+  %.09 = phi i32 [ %n, %0 ], [ %12, %1 ]
+  %.08 = phi i8* [ %b, %0 ], [ %11, %1 ]
+  %.0 = phi i8* [ %a, %0 ], [ %2, %1 ]
+  %2 = getelementptr inbounds i8* %.0, i64 1
+  %3 = load i8* %.0, align 1
+  %4 = sext i8 %3 to i64
+  %5 = getelementptr inbounds i8* %c, i64 %4
+  %6 = load i8* %5, align 1
+  %7 = add i8 %3, 1
+  %wrap.4 = sext i8 %7 to i64
+  %8 = getelementptr inbounds i8* %c, i64 %wrap.4
+  %9 = load i8* %8, align 1
+  store i8 %6, i8* %.08, align 1
+  %10 = getelementptr inbounds i8* %.08, i64 1
+  store i8 %9, i8* %10, align 1
+  %11 = getelementptr inbounds i8* %.08, i64 2
+  %12 = add nsw i32 %.09, -1
+  %13 = icmp eq i32 %12, 0
+  br i1 %13, label %14, label %1
+
+; <label>:14
+  ret void
+}