[X86] Merge adjacent stack adjustments in eliminateCallFramePseudoInstr (PR27140)

For code such as:

  void f(int, int);
  void g() {
      f(1, 2);
  }

compiled for 32-bit X86 Linux, Clang would previously generate:

  subl    $12, %esp
  subl    $8, %esp
  pushl   $2
  pushl   $1
  calll   f
  addl    $16, %esp
  addl    $12, %esp
  retl

This patch fixes that by merging adjacent stack adjustments in
eliminateCallFramePseudoInstr().

Differential Revision: http://reviews.llvm.org/D18627

llvm-svn: 265039

9 files changed, 65 insertions, 30 deletions

diff --git a/llvm/lib/Target/X86/X86FrameLowering.cpp b/llvm/lib/Target/X86/X86FrameLowering.cpp
index 413195f09c6..d8a376a6200 100644
--- a/llvm/lib/Target/X86/X86FrameLowering.cpp
+++ b/llvm/lib/Target/X86/X86FrameLowering.cpp
@@ -2534,13 +2534,22 @@ eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB,
       BuildCFI(MBB, I, DL, 
                MCCFIInstruction::createAdjustCfaOffset(nullptr, -InternalAmt));
 
-    if (Amount) {
-      // Add Amount to SP to destroy a frame, and subtract to setup.
-      int Offset = isDestroy ? Amount : -Amount;
+    // Add Amount to SP to destroy a frame, or subtract to setup.
+    int64_t StackAdjustment = isDestroy ? Amount : -Amount;
 
-      if (!(Fn->optForMinSize() && 
-            adjustStackWithPops(MBB, I, DL, Offset)))
-        BuildStackAdjustment(MBB, I, DL, Offset, /*InEpilogue=*/false);
+    if (StackAdjustment) {
+      // Merge with any previous or following adjustment instruction.
+      StackAdjustment += mergeSPUpdates(MBB, I, true);
+      StackAdjustment += mergeSPUpdates(MBB, I, false);
+
+      if (!StackAdjustment) {
+        // This and the merged instruction canceled out each other.
+        return I;
+      }
+
+      if (!(Fn->optForMinSize() &&
+            adjustStackWithPops(MBB, I, DL, StackAdjustment)))
+        BuildStackAdjustment(MBB, I, DL, StackAdjustment, /*InEpilogue=*/false);
     }
 
     if (DwarfCFI && !hasFP(MF)) {
@@ -2550,14 +2559,12 @@ eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB,
       // CFI only for EH purposes or for debugging. EH only requires the CFA
       // offset to be correct at each call site, while for debugging we want
       // it to be more precise.
-      int CFAOffset = Amount;
+
       // TODO: When not using precise CFA, we also need to adjust for the
       // InternalAmt here.
-
-      if (CFAOffset) {
-        CFAOffset = isDestroy ? -CFAOffset : CFAOffset;
-        BuildCFI(MBB, I, DL, 
-                 MCCFIInstruction::createAdjustCfaOffset(nullptr, CFAOffset));
+      if (StackAdjustment) {
+        BuildCFI(MBB, I, DL, MCCFIInstruction::createAdjustCfaOffset(
+                                 nullptr, -StackAdjustment));
       }
     }
 
diff --git a/llvm/test/CodeGen/X86/2006-05-02-InstrSched1.ll b/llvm/test/CodeGen/X86/2006-05-02-InstrSched1.ll
index eae0ec21c09..acd32e49e60 100644
--- a/llvm/test/CodeGen/X86/2006-05-02-InstrSched1.ll
+++ b/llvm/test/CodeGen/X86/2006-05-02-InstrSched1.ll
@@ -1,6 +1,6 @@
 ; REQUIRES: asserts
 ; RUN: llc < %s -mtriple=i686-unknown-linux -relocation-model=static -stats 2>&1 | \
-; RUN:   grep asm-printer | grep 15
+; RUN:   grep asm-printer | grep 14
 ;
 ; It's possible to schedule this in 14 instructions by avoiding
 ; callee-save registers, but the scheduler isn't currently that
diff --git a/llvm/test/CodeGen/X86/fold-push.ll b/llvm/test/CodeGen/X86/fold-push.ll
index eaf91351021..9d3afd1c449 100644
--- a/llvm/test/CodeGen/X86/fold-push.ll
+++ b/llvm/test/CodeGen/X86/fold-push.ll
@@ -14,7 +14,7 @@ define void @test(i32 %a, i32 %b) optsize nounwind {
 ; SLM: movl (%esp), [[RELOAD:%e..]]
 ; SLM-NEXT: pushl [[RELOAD]]
 ; CHECK: calll
-; CHECK-NEXT: addl $4, %esp
+; CHECK-NEXT: addl $8, %esp
   %c = add i32 %a, %b
   call void @foo(i32 %c)
   call void asm sideeffect "nop", "~{ax},~{bx},~{cx},~{dx},~{bp},~{si},~{di}"()
diff --git a/llvm/test/CodeGen/X86/force-align-stack-alloca.ll b/llvm/test/CodeGen/X86/force-align-stack-alloca.ll
index d0cf3417008..8d42680e199 100644
--- a/llvm/test/CodeGen/X86/force-align-stack-alloca.ll
+++ b/llvm/test/CodeGen/X86/force-align-stack-alloca.ll
@@ -32,15 +32,21 @@ define i64 @g(i32 %i) nounwind {
 ; CHECK:      movl   %{{...}}, %esp
 ; CHECK-NOT:         {{[^ ,]*}}, %esp
 ;
-; Next we set up the memset call, and then undo it.
+; Next we set up the memset call.
 ; CHECK:      subl   $20, %esp
 ; CHECK-NOT:         {{[^ ,]*}}, %esp
+; CHECK:      pushl
+; CHECK:      pushl
+; CHECK:      pushl
 ; CHECK:      calll  memset
-; CHECK-NEXT: addl   $32, %esp
+;
+; Deallocating 32 bytes of outgoing call frame for memset and
+; allocating 28 bytes for calling f yields a 4-byte adjustment:
+; CHECK-NEXT: addl   $4, %esp
 ; CHECK-NOT:         {{[^ ,]*}}, %esp
 ;
-; Next we set up the call to 'f'.
-; CHECK:      subl   $28, %esp
+; And move on to call 'f', and then restore the stack.
+; CHECK:      pushl
 ; CHECK-NOT:         {{[^ ,]*}}, %esp
 ; CHECK:      calll  f
 ; CHECK-NEXT: addl   $32, %esp
diff --git a/llvm/test/CodeGen/X86/localescape.ll b/llvm/test/CodeGen/X86/localescape.ll
index 07c3b7f4a35..10ab8dd9672 100644
--- a/llvm/test/CodeGen/X86/localescape.ll
+++ b/llvm/test/CodeGen/X86/localescape.ll
@@ -137,6 +137,5 @@ define void @alloc_func_no_frameaddr() {
 ; X86: movl $13, (%esp)
 ; X86: pushl $0
 ; X86: calll _print_framealloc_from_fp
-; X86: addl    $4, %esp
-; X86: addl    $8, %esp
+; X86: addl    $12, %esp
 ; X86: retl
diff --git a/llvm/test/CodeGen/X86/memset-2.ll b/llvm/test/CodeGen/X86/memset-2.ll
index 7f37b62a28b..e9a7b566b1d 100644
--- a/llvm/test/CodeGen/X86/memset-2.ll
+++ b/llvm/test/CodeGen/X86/memset-2.ll
@@ -5,7 +5,7 @@ declare void @llvm.memset.i32(i8*, i8, i32, i32) nounwind
 
 define fastcc void @t1() nounwind {
 ; CHECK-LABEL: t1:
-; CHECK:         subl $12, %esp
+; CHECK:         subl $16, %esp
 ; CHECK:         pushl $188
 ; CHECK-NEXT:    pushl $0
 ; CHECK-NEXT:    pushl $0
diff --git a/llvm/test/CodeGen/X86/movtopush.ll b/llvm/test/CodeGen/X86/movtopush.ll
index 5dd465ef5a3..ceefb0aa58d 100644
--- a/llvm/test/CodeGen/X86/movtopush.ll
+++ b/llvm/test/CodeGen/X86/movtopush.ll
@@ -2,6 +2,7 @@
 ; RUN: llc < %s -mtriple=i686-windows -no-x86-call-frame-opt | FileCheck %s -check-prefix=NOPUSH
 ; RUN: llc < %s -mtriple=x86_64-windows | FileCheck %s -check-prefix=X64
 ; RUN: llc < %s -mtriple=i686-windows -stackrealign -stack-alignment=32 | FileCheck %s -check-prefix=ALIGNED
+; RUN: llc < %s -mtriple=i686-pc-linux | FileCheck %s -check-prefix=LINUX
 
 %class.Class = type { i32 }
 %struct.s = type { i64 }
@@ -223,8 +224,7 @@ entry:
 ; NORMAL-NEXT: pushl $2
 ; NORMAL-NEXT: pushl $1
 ; NORMAL-NEXT: call
-; NORMAL-NEXT: addl $16, %esp
-; NORMAL-NEXT: subl $20, %esp
+; NORMAL-NEXT: subl $4, %esp
 ; NORMAL-NEXT: movl 20(%esp), [[E1:%e..]]
 ; NORMAL-NEXT: movl 24(%esp), [[E2:%e..]]
 ; NORMAL-NEXT: movl    [[E2]], 4(%esp)
@@ -261,7 +261,7 @@ entry:
 ; NORMAL-NEXT: pushl $2
 ; NORMAL-NEXT: pushl $1
 ; NORMAL-NEXT: calll *16(%esp)
-; NORMAL-NEXT: addl $16, %esp
+; NORMAL-NEXT: addl $24, %esp
 define void @test10() optsize {
   %stack_fptr = alloca void (i32, i32, i32, i32)*
   store void (i32, i32, i32, i32)* @good, void (i32, i32, i32, i32)** %stack_fptr
@@ -314,8 +314,7 @@ entry:
 ; NORMAL-NEXT: pushl    $2
 ; NORMAL-NEXT: pushl    $1
 ; NORMAL-NEXT: calll _good
-; NORMAL-NEXT: addl    $16, %esp
-; NORMAL-NEXT: subl    $20, %esp
+; NORMAL-NEXT: subl    $4, %esp
 ; NORMAL: movl    $8, 16(%esp)
 ; NORMAL-NEXT: movl    $7, 12(%esp)
 ; NORMAL-NEXT: movl    $6, 8(%esp)
@@ -358,3 +357,27 @@ entry:
   call void @good(i32 %val1, i32 %val2, i32 %val3, i32 %add)
   ret i32* %ptr3
 }
+
+; Make sure to fold adjacent stack adjustments.
+; LINUX-LABEL: pr27140:
+; LINUX: subl    $12, %esp
+; LINUX: .cfi_def_cfa_offset 16
+; LINUX-NOT: sub
+; LINUX: pushl   $4
+; LINUX: .cfi_adjust_cfa_offset 4
+; LINUX: pushl   $3
+; LINUX: .cfi_adjust_cfa_offset 4
+; LINUX: pushl   $2
+; LINUX: .cfi_adjust_cfa_offset 4
+; LINUX: pushl   $1
+; LINUX: .cfi_adjust_cfa_offset 4
+; LINUX: calll   good
+; LINUX: addl    $28, %esp
+; LINUX: .cfi_adjust_cfa_offset -28
+; LINUX-NOT: add
+; LINUX: retl
+define void @pr27140() optsize {
+entry:
+  tail call void @good(i32 1, i32 2, i32 3, i32 4)
+  ret void
+}
diff --git a/llvm/test/CodeGen/X86/push-cfi-debug.ll b/llvm/test/CodeGen/X86/push-cfi-debug.ll
index cc00fab525a..ee4cb8e0e60 100644
--- a/llvm/test/CodeGen/X86/push-cfi-debug.ll
+++ b/llvm/test/CodeGen/X86/push-cfi-debug.ll
@@ -23,8 +23,8 @@ declare x86_stdcallcc void @stdfoo(i32, i32) #0
 ; CHECK: .cfi_adjust_cfa_offset 4
 ; CHECK: calll stdfoo
 ; CHECK: .cfi_adjust_cfa_offset -8
-; CHECK: addl $8, %esp
-; CHECK: .cfi_adjust_cfa_offset -8
+; CHECK: addl $20, %esp
+; CHECK: .cfi_adjust_cfa_offset -20
 define void @test1() #0 !dbg !4 {
 entry:
   tail call void @foo(i32 1, i32 2) #1, !dbg !10
diff --git a/llvm/test/CodeGen/X86/push-cfi.ll b/llvm/test/CodeGen/X86/push-cfi.ll
index 6389708f42c..5498af51f23 100644
--- a/llvm/test/CodeGen/X86/push-cfi.ll
+++ b/llvm/test/CodeGen/X86/push-cfi.ll
@@ -82,8 +82,8 @@ cleanup:
 ; LINUX-NEXT: Ltmp{{[0-9]+}}:
 ; LINUX-NEXT: .cfi_adjust_cfa_offset 4
 ; LINUX-NEXT: call
-; LINUX-NEXT: addl $16, %esp
-; LINUX: .cfi_adjust_cfa_offset -16
+; LINUX-NEXT: addl $28, %esp
+; LINUX: .cfi_adjust_cfa_offset -28
 ; DARWIN-NOT: .cfi_escape
 ; DARWIN-NOT: pushl
 define void @test2_nofp() #0 personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) {