A few missed optimizations; the last one could have a significant impact on

code with lots of bitfields.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@95809 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/README.txt b/lib/Target/X86/README.txt
index aa7bb3d9788924974c059e1c8fecbcc2241c1253..3c6138bb1e6a64089439a068c328c0e4834b5cc3 100644 (file)
--- a/lib/Target/X86/README.txt
+++ b/lib/Target/X86/README.txt
@@ -1868,3 +1868,69 @@ carried over to machine instructions. Asm printer (or JIT) can use this
 information to add the "lock" prefix.
 
 //===---------------------------------------------------------------------===//
+
+_Bool bar(int *x) { return *x & 1; }
+
+define zeroext i1 @bar(i32* nocapture %x) nounwind readonly {
+entry:
+  %tmp1 = load i32* %x                            ; <i32> [#uses=1]
+  %and = and i32 %tmp1, 1                         ; <i32> [#uses=1]
+  %tobool = icmp ne i32 %and, 0                   ; <i1> [#uses=1]
+  ret i1 %tobool
+}
+
+bar:                                                        # @bar
+# BB#0:                                                     # %entry
+       movl    4(%esp), %eax
+       movb    (%eax), %al
+       andb    $1, %al
+       movzbl  %al, %eax
+       ret
+
+Missed optimization: should be movl+andl.
+
+//===---------------------------------------------------------------------===//
+
+Consider the following two functions compiled with clang:
+_Bool foo(int *x) { return !(*x & 4); }
+unsigned bar(int *x) { return !(*x & 4); }
+
+foo:
+       movl    4(%esp), %eax
+       testb   $4, (%eax)
+       sete    %al
+       movzbl  %al, %eax
+       ret
+
+bar:
+       movl    4(%esp), %eax
+       movl    (%eax), %eax
+       shrl    $2, %eax
+       andl    $1, %eax
+       xorl    $1, %eax
+       ret
+
+The second function generates more code even though the two functions are
+are functionally identical.
+
+//===---------------------------------------------------------------------===//
+
+Take the following C code:
+int x(int y) { return (y & 63) << 14; }
+
+Code produced by gcc:
+       andl    $63, %edi
+       sall    $14, %edi
+       movl    %edi, %eax
+       ret
+
+Code produced by clang:
+       shll    $14, %edi
+       movl    %edi, %eax
+       andl    $1032192, %eax
+       ret
+
+The code produced by gcc is 3 bytes shorter.  This sort of construct often
+shows up with bitfields.
+
+//===---------------------------------------------------------------------===//