[WebAssembly] Stackify function prologs and epilogs

The instructions are the same, but fewer locals are used.

Differential Revision: http://reviews.llvm.org/D17428

llvm-svn: 261452

1 files changed, 70 insertions, 68 deletions

diff --git a/llvm/test/CodeGen/WebAssembly/userstack.ll b/llvm/test/CodeGen/WebAssembly/userstack.ll
index 2032507..ac41c94 100644
--- a/llvm/test/CodeGen/WebAssembly/userstack.ll
+++ b/llvm/test/CodeGen/WebAssembly/userstack.ll
@@ -7,54 +7,58 @@ target triple = "wasm32-unknown-unknown"
 
 ; CHECK-LABEL: alloca32:
 ; Check that there is an extra local for the stack pointer.
-; CHECK: .local i32, i32, i32, i32{{$}}
+; CHECK: .local i32{{$}}
 define void @alloca32() {
- ; CHECK: i32.const [[L1:.+]]=, __stack_pointer
- ; CHECK-NEXT: i32.load [[L1]]=, 0([[L1]])
- ; CHECK-NEXT: i32.const [[L2:.+]]=, 16
- ; CHECK-NEXT: i32.sub [[SP:.+]]=, [[L1]], [[L2]]
+ ; CHECK: i32.const $push[[L1:.+]]=, __stack_pointer{{$}}
+ ; CHECK-NEXT: i32.load $push[[L2:.+]]=, 0($pop[[L1]])
+ ; CHECK-NEXT: i32.const $push[[L3:.+]]=, 16
+ ; CHECK-NEXT: i32.sub [[SP:.+]]=, $pop[[L2]], $pop[[L3]]
+ ; CHECK-NEXT: i32.const $push[[L4:.+]]=, __stack_pointer{{$}}
+ ; CHECK-NEXT: i32.store [[SP]]=, 0($pop[[L4]]), [[SP]]
  %retval = alloca i32
- ; CHECK: i32.const $push[[L3:.+]]=, 0
- ; CHECK: i32.store {{.*}}=, 12([[SP]]), $pop[[L3]]
+ ; CHECK: i32.const $push[[L0:.+]]=, 0
+ ; CHECK: i32.store {{.*}}=, 12([[SP]]), $pop[[L0]]
  store i32 0, i32* %retval
- ; CHECK: i32.const [[L4:.+]]=, 16
- ; CHECK-NEXT: i32.add [[SP]]=, [[SP]], [[L4]]
- ; CHECK-NEXT: i32.const [[L5:.+]]=, __stack_pointer
- ; CHECK-NEXT: i32.store [[SP]]=, 0([[L5]]), [[SP]]
+ ; CHECK: i32.const $push[[L5:.+]]=, 16
+ ; CHECK-NEXT: i32.add [[SP]]=, [[SP]], $pop[[L5]]
+ ; CHECK-NEXT: i32.const $push[[L6:.+]]=, __stack_pointer
+ ; CHECK-NEXT: i32.store [[SP]]=, 0($pop[[L6]]), [[SP]]
  ret void
 }
 
 ; CHECK-LABEL: alloca3264:
-; CHECK: .local i32, i32, i32, i32{{$}}
+; CHECK: .local i32{{$}}
 define void @alloca3264() {
- ; CHECK: i32.const [[L1:.+]]=, __stack_pointer
- ; CHECK-NEXT: i32.load [[L1]]=, 0([[L1]])
- ; CHECK-NEXT: i32.const [[L2:.+]]=, 16
- ; CHECK-NEXT: i32.sub [[SP:.+]]=, [[L1]], [[L2]]
+ ; CHECK: i32.const $push[[L1:.+]]=, __stack_pointer
+ ; CHECK-NEXT: i32.load $push[[L2:.+]]=, 0($pop[[L1]])
+ ; CHECK-NEXT: i32.const $push[[L3:.+]]=, 16
+ ; CHECK-NEXT: i32.sub [[SP:.+]]=, $pop[[L2]], $pop[[L3]]
+ ; CHECK-NEXT: i32.const $push[[L4:.+]]=, __stack_pointer{{$}}
+ ; CHECK-NEXT: i32.store [[SP]]=, 0($pop[[L4]]), [[SP]]
  %r1 = alloca i32
  %r2 = alloca double
  ; CHECK: i32.const $push[[L3:.+]]=, 0
  ; CHECK: i32.store {{.*}}=, 12([[SP]]), $pop[[L3]]
  store i32 0, i32* %r1
- ; CHECK: i64.const $push[[L4:.+]]=, 0
- ; CHECK: i64.store {{.*}}=, 0([[SP]]), $pop[[L4]]
+ ; CHECK: i64.const $push[[L0:.+]]=, 0
+ ; CHECK: i64.store {{.*}}=, 0([[SP]]), $pop[[L0]]
  store double 0.0, double* %r2
- ; CHECK: i32.const [[L4:.+]]=, 16
- ; CHECK-NEXT: i32.add [[SP]]=, [[SP]], [[L4]]
- ; CHECK-NEXT: i32.const [[L5:.+]]=, __stack_pointer
- ; CHECK-NEXT: i32.store [[SP]]=, 0([[L5]]), [[SP]]
+ ; CHECK: i32.const $push[[L5:.+]]=, 16
+ ; CHECK-NEXT: i32.add [[SP]]=, [[SP]], $pop[[L5]]
+ ; CHECK-NEXT: i32.const $push[[L6:.+]]=, __stack_pointer
+ ; CHECK-NEXT: i32.store [[SP]]=, 0($pop[[L6]]), [[SP]]
  ret void
 }
 
 ; CHECK-LABEL: allocarray:
-; CHECK: .local i32, i32, i32, i32, i32{{$}}
+; CHECK: .local i32, i32{{$}}
 define void @allocarray() {
- ; CHECK-NEXT: i32.const [[L1:.+]]=, __stack_pointer
- ; CHECK-NEXT: i32.load [[L1]]=, 0([[L1]])
- ; CHECK-NEXT: i32.const [[L2:.+]]=, 32{{$}}
- ; CHECK-NEXT: i32.sub [[SP:.+]]=, [[L1]], [[L2]]
- ; CHECK-NEXT: i32.const [[L2]]=, __stack_pointer{{$}}
- ; CHECK-NEXT: i32.store [[SP]]=, 0([[L2]]), [[SP]]
+ ; CHECK: i32.const $push[[L1:.+]]=, __stack_pointer
+ ; CHECK-NEXT: i32.load $push[[L2:.+]]=, 0($pop[[L1]])
+ ; CHECK-NEXT: i32.const $push[[L3:.+]]=, 32{{$}}
+ ; CHECK-NEXT: i32.sub [[SP:.+]]=, $pop[[L2]], $pop[[L3]]
+ ; CHECK-NEXT: i32.const $push[[L4:.+]]=, __stack_pointer{{$}}
+ ; CHECK-NEXT: i32.store [[SP]]=, 0($pop[[L4]]), [[SP]]
  %r = alloca [5 x i32]
 
  ; CHECK-NEXT: i32.const $push[[L4:.+]]=, 12
@@ -69,18 +73,18 @@ define void @allocarray() {
  %p2 = getelementptr [5 x i32], [5 x i32]* %r, i32 0, i32 3
  store i32 1, i32* %p2
 
- ; CHECK-NEXT: i32.const [[L7:.+]]=, 32
- ; CHECK-NEXT: i32.add [[SP]]=, [[SP]], [[L7]]
- ; CHECK-NEXT: i32.const [[L8:.+]]=, __stack_pointer
- ; CHECK-NEXT: i32.store [[SP]]=, 0([[L8]]), [[SP]]
+ ; CHECK: i32.const $push[[L11:.+]]=, 32
+ ; CHECK-NEXT: i32.add [[SP]]=, [[SP]], $pop[[L11]]
+ ; CHECK-NEXT: i32.const $push[[L12:.+]]=, __stack_pointer
+ ; CHECK-NEXT: i32.store [[SP]]=, 0($pop[[L12]]), [[SP]]
  ret void
 }
 
 declare void @ext_func(i64* %ptr)
 ; CHECK-LABEL: non_mem_use
 define void @non_mem_use(i8** %addr) {
- ; CHECK: i32.const [[L2:.+]]=, 48
- ; CHECK-NEXT: i32.sub [[SP:.+]]=, {{.+}}, [[L2]]
+ ; CHECK: i32.const $push[[L1:.+]]=, 48
+ ; CHECK-NEXT: i32.sub [[SP:.+]]=, {{.+}}, $pop[[L1]]
  %buf = alloca [27 x i8], align 16
  %r = alloca i64
  %r2 = alloca i64
@@ -103,12 +107,12 @@ define void @non_mem_use(i8** %addr) {
 }
 
 ; CHECK-LABEL: allocarray_inbounds:
-; CHECK: .local i32, i32, i32, i32{{$}}
+; CHECK: .local i32{{$}}
 define void @allocarray_inbounds() {
- ; CHECK: i32.const [[L1:.+]]=, __stack_pointer
- ; CHECK-NEXT: i32.load [[L1]]=, 0([[L1]])
- ; CHECK-NEXT: i32.const [[L2:.+]]=, 32
- ; CHECK-NEXT: i32.sub [[SP:.+]]=, [[L1]], [[L2]]
+ ; CHECK: i32.const $push[[L1:.+]]=, __stack_pointer
+ ; CHECK-NEXT: i32.load $push[[L2:.+]]=, 0($pop[[L1]])
+ ; CHECK-NEXT: i32.const $push[[L3:.+]]=, 32{{$}}
+ ; CHECK-NEXT: i32.sub [[SP:.+]]=, $pop[[L2]], $pop[[L3]]
  %r = alloca [5 x i32]
  ; CHECK: i32.const $push[[L3:.+]]=, 1
  ; CHECK: i32.store {{.*}}=, 12([[SP]]), $pop[[L3]]
@@ -118,45 +122,45 @@ define void @allocarray_inbounds() {
  ; CHECK-NEXT: i32.store {{.*}}=, 24([[SP]]), $pop
  %p2 = getelementptr inbounds [5 x i32], [5 x i32]* %r, i32 0, i32 3
  store i32 1, i32* %p2
- ; CHECK: i32.const [[L7:.+]]=, 32
- ; CHECK-NEXT: i32.add [[SP]]=, [[SP]], [[L7]]
- ; CHECK-NEXT: i32.const [[L8:.+]]=, __stack_pointer
- ; CHECK-NEXT: i32.store [[SP]]=, 0([[L7]]), [[SP]]
+ ; CHECK: i32.const $push[[L5:.+]]=, 32
+ ; CHECK-NEXT: i32.add [[SP]]=, [[SP]], $pop[[L5]]
+ ; CHECK-NEXT: i32.const $push[[L6:.+]]=, __stack_pointer
+ ; CHECK-NEXT: i32.store [[SP]]=, 0($pop[[L6]]), [[SP]]
  ret void
 }
 
 ; CHECK-LABEL: dynamic_alloca:
 define void @dynamic_alloca(i32 %alloc) {
- ; CHECK: i32.const [[L0:.+]]=, __stack_pointer
- ; CHECK-NEXT: i32.load [[SP:.+]]=, 0([[L0]])
+ ; CHECK: i32.const $push[[L1:.+]]=, __stack_pointer
+ ; CHECK-NEXT: i32.load [[SP:.+]]=, 0($pop[[L1]])
  ; CHECK-NEXT: copy_local [[FP:.+]]=, [[SP]]
  ; Target independent codegen bumps the stack pointer
  ; FIXME: we need to write the value back to memory
  %r = alloca i32, i32 %alloc
  ; Target-independent codegen also calculates the store addr
  store i32 0, i32* %r
- ; CHECK: i32.const [[L3:.+]]=, __stack_pointer
- ; CHECK-NEXT: i32.store [[SP]]=, 0([[L3]]), [[FP]]
+ ; CHECK: i32.const $push[[L3:.+]]=, __stack_pointer
+ ; CHECK-NEXT: i32.store [[SP]]=, 0($pop[[L3]]), [[FP]]
  ret void
 }
 
 
 ; CHECK-LABEL: dynamic_static_alloca:
 define void @dynamic_static_alloca(i32 %alloc) {
- ; CHECK: i32.const [[L0:.+]]=, __stack_pointer
- ; CHECK-NEXT: i32.load [[L0]]=, 0([[L0]])
- ; CHECK-NEXT: i32.const [[L2:.+]]=, 16
- ; CHECK-NEXT: i32.sub [[SP:.+]]=, [[L0]], [[L2]]
+ ; CHECK: i32.const $push[[L1:.+]]=, __stack_pointer
+ ; CHECK-NEXT: i32.load $push[[L2:.+]]=, 0($pop[[L1]])
+ ; CHECK-NEXT: i32.const $push[[L3:.+]]=, 16
+ ; CHECK-NEXT: i32.sub [[SP:.+]]=, $pop[[L2]], $pop[[L3]]
  ; CHECK-NEXT: copy_local [[FP:.+]]=, [[SP]]
- ; CHECK-NEXT: i32.const [[L3:.+]]=, __stack_pointer
- ; CHECK-NEXT: i32.store {{.*}}=, 0([[L3]]), [[SP]]
+ ; CHECK-NEXT: i32.const $push[[L4:.+]]=, __stack_pointer
+ ; CHECK-NEXT: i32.store {{.*}}=, 0($pop[[L4]]), [[SP]]
  %r1 = alloca i32
  %r = alloca i32, i32 %alloc
  store i32 0, i32* %r
- ; CHECK: i32.const [[L3:.+]]=, 16
- ; CHECK: i32.add [[SP]]=, [[FP]], [[L3]]
- ; CHECK: i32.const [[L4:.+]]=, __stack_pointer
- ; CHECK-NEXT: i32.store [[SP]]=, 0([[L4]]), [[SP]]
+ ; CHECK: i32.const $push[[L5:.+]]=, 16
+ ; CHECK-NEXT: i32.add [[SP]]=, [[FP]], $pop[[L5]]
+ ; CHECK-NEXT: i32.const $push[[L6:.+]]=, __stack_pointer
+ ; CHECK-NEXT: i32.store [[SP]]=, 0($pop[[L6]]), [[SP]]
  ret void
 }
 
@@ -165,8 +169,8 @@ define void @dynamic_static_alloca(i32 %alloc) {
 ; CHECK-LABEL: copytoreg_fi:
 define void @copytoreg_fi(i1 %cond, i32* %b) {
 entry:
- ; CHECK: i32.const [[L2:.+]]=, 16
- ; CHECK-NEXT: i32.sub [[SP:.+]]=, {{.+}}, [[L2]]
+ ; CHECK: i32.const $push[[L1:.+]]=, 16
+ ; CHECK-NEXT: i32.sub [[SP:.+]]=, {{.+}}, $pop[[L1]]
  %addr = alloca i32
  ; CHECK: i32.const [[OFF:.+]]=, 12
  ; CHECK-NEXT: i32.add [[ADDR:.+]]=, [[SP]], [[OFF]]
@@ -185,15 +189,13 @@ declare void @use_i8_star(i8*)
 declare i8* @llvm.frameaddress(i32)
 
 ; Test __builtin_frame_address(0).
-; TODO: When the prolog/epilog sequences are optimized, refine these checks to
-; be more specific.
-
 ; CHECK-LABEL: frameaddress_0:
-; CHECK: __stack_pointer
-; CHECK: load
-; CHECK: call use_i8_star
-; CHECK: __stack_pointer
-; CHECK: store
+; CHECK: i32.const $push[[L1:.+]]=, __stack_pointer
+; CHECK-NEXT: i32.load [[SP:.+]]=, 0($pop[[L1]])
+; CHECK-NEXT: copy_local [[FP:.+]]=, [[SP]]
+; CHECK-NEXT: call use_i8_star@FUNCTION, [[FP]]
+; CHECK-NEXT: i32.const $push[[L6:.+]]=, __stack_pointer
+; CHECK-NEXT: i32.store [[SP]]=, 0($pop[[L6]]), [[FP]]
 define void @frameaddress_0() {
   %t = call i8* @llvm.frameaddress(i32 0)
   call void @use_i8_star(i8* %t)