The current Intel Atom microarchitecture has a feature whereby when a function

returns early then it is slightly faster to execute a sequence of NOP
instructions to wait until the return address is ready,
as opposed to simply stalling on the ret instruction
until the return address is ready.

When compiling for X86 Atom only, this patch will run a pass, called
"X86PadShortFunction" which will add NOP instructions where less than four
cycles elapse between function entry and return.

It includes tests.

Patch by Andy Zhang.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171524 91177308-0d34-0410-b5e6-96231b3b80d8

4 files changed, 76 insertions, 5 deletions

diff --git a/test/CodeGen/X86/atom-pad-short-functions.ll b/test/CodeGen/X86/atom-pad-short-functions.ll
new file mode 100644
index 0000000000..3e59934597
--- /dev/null
+++ b/test/CodeGen/X86/atom-pad-short-functions.ll
@@ -0,0 +1,71 @@
+; RUN: llc < %s -mcpu=atom -mtriple=i686-linux  | FileCheck %s
+
+declare void @external_function(...)
+
+define i32 @test_return_val(i32 %a) nounwind {
+; CHECK: test_return_val
+; CHECK: movl
+; CHECK: nop
+; CHECK: nop
+; CHECK: nop
+; CHECK: nop
+; CHECK: nop
+; CHECK: nop
+; CHECK: ret
+  ret i32 %a
+}
+
+define i32 @test_add(i32 %a, i32 %b) nounwind {
+; CHECK: test_add
+; CHECK: addl
+; CHECK: nop
+; CHECK: nop
+; CHECK: nop
+; CHECK: nop
+; CHECK: ret
+  %result = add i32 %a, %b
+  ret i32 %result
+}
+
+define i32 @test_multiple_ret(i32 %a, i32 %b, i1 %c) nounwind {
+; CHECK: @test_multiple_ret
+; CHECK: je
+
+; CHECK: nop
+; CHECK: nop
+; CHECK: ret
+
+; CHECK: nop
+; CHECK: nop
+; CHECK: ret
+
+  br i1 %c, label %bb1, label %bb2
+
+bb1:
+  ret i32 %a
+
+bb2:
+  ret i32 %b
+}
+
+define void @test_call_others(i32 %x) nounwind
+{
+; CHECK: test_call_others
+; CHECK: je
+  %tobool = icmp eq i32 %x, 0
+  br i1 %tobool, label %if.end, label %true.case
+
+; CHECK: jmp external_function
+true.case:
+  tail call void bitcast (void (...)* @external_function to void ()*)() nounwind
+  br label %if.end
+
+; CHECK: nop
+; CHECK: nop
+; CHECK: nop
+; CHECK: nop
+; CHECK: ret
+if.end:
+  ret void
+
+}
diff --git a/test/CodeGen/X86/fast-isel-x86-64.ll b/test/CodeGen/X86/fast-isel-x86-64.ll
index cdfaf7f4c1..ad1520ef81 100644
--- a/test/CodeGen/X86/fast-isel-x86-64.ll
+++ b/test/CodeGen/X86/fast-isel-x86-64.ll
@@ -1,5 +1,5 @@
-; RUN: llc < %s -mattr=-avx -fast-isel -O0 -regalloc=fast -asm-verbose=0 -fast-isel-abort | FileCheck %s
-; RUN: llc < %s -mattr=+avx -fast-isel -O0 -regalloc=fast -asm-verbose=0 -fast-isel-abort | FileCheck %s --check-prefix=AVX
+; RUN: llc < %s -mattr=-avx -fast-isel -mcpu=core2 -O0 -regalloc=fast -asm-verbose=0 -fast-isel-abort | FileCheck %s
+; RUN: llc < %s -mattr=+avx -fast-isel -mcpu=core2 -O0 -regalloc=fast -asm-verbose=0 -fast-isel-abort | FileCheck %s --check-prefix=AVX
 
 target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64"
 target triple = "x86_64-apple-darwin10.0.0"
diff --git a/test/CodeGen/X86/ret-mmx.ll b/test/CodeGen/X86/ret-mmx.ll
index 865e147a4a..778e4722cd 100644
--- a/test/CodeGen/X86/ret-mmx.ll
+++ b/test/CodeGen/X86/ret-mmx.ll
@@ -1,4 +1,4 @@
-; RUN: llc < %s -mtriple=x86_64-apple-darwin11 -mattr=+mmx,+sse2 | FileCheck %s
+; RUN: llc < %s -mtriple=x86_64-apple-darwin11 -mcpu=core2 -mattr=+mmx,+sse2 | FileCheck %s
 ; rdar://6602459
 
 @g_v1di = external global <1 x i64>
diff --git a/test/CodeGen/X86/select.ll b/test/CodeGen/X86/select.ll
index 3bec3acdbf..09ca07b31a 100644
--- a/test/CodeGen/X86/select.ll
+++ b/test/CodeGen/X86/select.ll
@@ -282,7 +282,7 @@ define i32 @test13(i32 %a, i32 %b) nounwind {
 ; ATOM: test13:
 ; ATOM: cmpl
 ; ATOM-NEXT: sbbl
-; ATOM-NEXT: ret
+; ATOM: ret
 }
 
 define i32 @test14(i32 %a, i32 %b) nounwind {
@@ -299,7 +299,7 @@ define i32 @test14(i32 %a, i32 %b) nounwind {
 ; ATOM: cmpl
 ; ATOM-NEXT: sbbl
 ; ATOM-NEXT: notl
-; ATOM-NEXT: ret
+; ATOM: ret
 }
 
 ; rdar://10961709