Skip to content

x64: reduce work in non-stack using leaf functions #5352

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
abrown wants to merge 5 commits into
base: main
Choose a base branch
from
+64 −1,827
Draft

x64: reduce work in non-stack using leaf functions #5352

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
c47ecbf
x64: reduce work in non-stack using leaf functions
abrown Nov 29, 2022
094c665
test: update golden output tests
abrown Nov 29, 2022
bd011d8
test: update systemv unwinding test output
abrown Nov 29, 2022
b2d3e98
test: update CLIF precise output
abrown Nov 29, 2022
d9945d4
fix: disable frame setup elision on macos
abrown Nov 30, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
23 changes: 15 additions & 8 deletions cranelift/codegen/src/isa/x64/abi.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -468,9 +468,9 @@ impl ABIMachineSpec for X64ABIMachineSpec {

if flags.unwind_info() && setup_frame {
// Emit unwind info: start the frame. The frame (from unwind
// consumers' point of view) starts at clobbbers, just below
// the FP and return address. Spill slots and stack slots are
// part of our actual frame but do not concern the unwinder.
// consumers' point of view) starts at clobbers, just below the FP
// and return address. Spill slots and stack slots are part of our
// actual frame but do not concern the unwinder.
insts.push(Inst::Unwind {
inst: UnwindInst::DefineNewFrame {
offset_downward_to_clobbers: clobbered_size,
Expand Down Expand Up @@ -732,12 +732,19 @@ impl ABIMachineSpec for X64ABIMachineSpec {
}

fn is_frame_setup_needed(
_is_leaf: bool,
_stack_args_size: u32,
_num_clobbered_callee_saves: usize,
_frame_storage_size: u32,
is_leaf: bool,
stack_args_size: u32,
num_clobbered_callee_saves: usize,
frame_storage_size: u32,
) -> bool {
true
(!is_leaf
// The function arguments that are passed on the stack are addressed
// relative to the Frame Pointer.
|| stack_args_size > 0
|| num_clobbered_callee_saves > 0
|| frame_storage_size > 0)
// TODO
&& !cfg!(target_os = "macos")
Copy link
Contributor

@bjorn3 bjorn3 Dec 2, 2022

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Maybe my suggestion at #4469 (comment) would work? "non-leaf" could be used by default and "always" could be used on macOS.

}
}

Expand Down
9 changes: 7 additions & 2 deletions cranelift/codegen/src/isa/x64/inst/unwind/systemv.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -130,7 +130,12 @@ mod tests {
_ => panic!("expected unwind information"),
};

assert_eq!(format!("{:?}", fde), "FrameDescriptionEntry { address: Constant(1234), length: 17, lsda: None, instructions: [(1, CfaOffset(16)), (1, Offset(Register(6), -16)), (4, CfaRegister(Register(6)))] }");
if cfg!(target_os = "macos") {
// TODO
assert_eq!(format!("{:?}", fde), "FrameDescriptionEntry { address: Constant(1234), length: 9, lsda: None, instructions: [] }");
} else {
assert_eq!(format!("{:?}", fde), "FrameDescriptionEntry { address: Constant(1234), length: 17, lsda: None, instructions: [(1, CfaOffset(16)), (1, Offset(Register(6), -16)), (4, CfaRegister(Register(6)))] }");
}
}

fn create_function(call_conv: CallConv, stack_slot: Option<StackSlotData>) -> Function {
Expand Down Expand Up @@ -169,7 +174,7 @@ mod tests {
_ => panic!("expected unwind information"),
};

assert_eq!(format!("{:?}", fde), "FrameDescriptionEntry { address: Constant(4321), length: 22, lsda: None, instructions: [(1, CfaOffset(16)), (1, Offset(Register(6), -16)), (4, CfaRegister(Register(6)))] }");
assert_eq!(format!("{:?}", fde), "FrameDescriptionEntry { address: Constant(4321), length: 10, lsda: None, instructions: [] }");
}

fn create_multi_return_function(call_conv: CallConv) -> Function {
Expand Down
97 changes: 42 additions & 55 deletions cranelift/codegen/src/isa/x64/mod.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -343,32 +343,28 @@ mod test {
// command on it:
// > objdump -b binary -D <file> -m i386:x86-64 -M intel
//
// 0: 55 push rbp
// 1: 48 89 e5 mov rbp,rsp
// 4: 48 89 fe mov rsi,rdi
// 7: 81 c6 34 12 00 00 add esi,0x1234
// d: 85 f6 test esi,esi
// f: 0f 84 1c 00 00 00 je 0x31
// 15: 49 89 f0 mov r8,rsi
// 18: 48 89 f0 mov rax,rsi
// 1b: 81 e8 34 12 00 00 sub eax,0x1234
// 21: 44 01 c0 add eax,r8d
// 24: 85 f6 test esi,esi
// 26: 0f 85 05 00 00 00 jne 0x31
// 2c: 48 89 ec mov rsp,rbp
// 2f: 5d pop rbp
// 30: c3 ret
// 31: 49 89 f0 mov r8,rsi
// 34: 41 81 c0 34 12 00 00 add r8d,0x1234
// 3b: 45 85 c0 test r8d,r8d
// 3e: 0f 85 ed ff ff ff jne 0x31
// 44: e9 cf ff ff ff jmp 0x18
// 0: 48 89 fe mov rsi,rdi
// 3: 81 c6 34 12 00 00 add esi,0x1234
// 9: 85 f6 test esi,esi
// b: 0f 84 18 00 00 00 je 0x29
// 11: 49 89 f0 mov r8,rsi
// 14: 48 89 f0 mov rax,rsi
// 17: 81 e8 34 12 00 00 sub eax,0x1234
// 1d: 44 01 c0 add eax,r8d
// 20: 85 f6 test esi,esi
// 22: 0f 85 01 00 00 00 jne 0x29
// 28: c3 ret
// 29: 49 89 f0 mov r8,rsi
// 2c: 41 81 c0 34 12 00 00 add r8d,0x1234
// 33: 45 85 c0 test r8d,r8d
// 36: 0f 85 ed ff ff ff jne 0x29
// 3c: e9 d3 ff ff ff jmp 0x14

let golden = vec![
85, 72, 137, 229, 72, 137, 254, 129, 198, 52, 18, 0, 0, 133, 246, 15, 132, 28, 0, 0, 0,
73, 137, 240, 72, 137, 240, 129, 232, 52, 18, 0, 0, 68, 1, 192, 133, 246, 15, 133, 5,
0, 0, 0, 72, 137, 236, 93, 195, 73, 137, 240, 65, 129, 192, 52, 18, 0, 0, 69, 133, 192,
15, 133, 237, 255, 255, 255, 233, 207, 255, 255, 255,
72, 137, 254, 129, 198, 52, 18, 0, 0, 133, 246, 15, 132, 24, 0, 0, 0, 73, 137, 240, 72,
137, 240, 129, 232, 52, 18, 0, 0, 68, 1, 192, 133, 246, 15, 133, 1, 0, 0, 0, 195, 73,
137, 240, 65, 129, 192, 52, 18, 0, 0, 69, 133, 192, 15, 133, 237, 255, 255, 255, 233,
211, 255, 255, 255,
];

assert_eq!(code, &golden[..]);
Expand Down Expand Up @@ -448,39 +444,30 @@ mod test {
// command on it:
// > objdump -b binary -D <file> -m i386:x86-64 -M intel
//
// 0: 55 push rbp
// 1: 48 89 e5 mov rbp,rsp
// 4: 83 ff 02 cmp edi,0x2
// 7: 0f 83 27 00 00 00 jae 0x34
// d: 44 8b d7 mov r10d,edi
// 10: 41 b9 00 00 00 00 mov r9d,0x0
// 16: 4d 0f 43 d1 cmovae r10,r9
// 1a: 4c 8d 0d 0b 00 00 00 lea r9,[rip+0xb] # 0x2c
// 21: 4f 63 54 91 00 movsxd r10,DWORD PTR [r9+r10*4+0x0]
// 26: 4d 01 d1 add r9,r10
// 29: 41 ff e1 jmp r9
// 2c: 12 00 adc al,BYTE PTR [rax]
// 2e: 00 00 add BYTE PTR [rax],al
// 30: 1c 00 sbb al,0x0
// 32: 00 00 add BYTE PTR [rax],al
// 34: b8 03 00 00 00 mov eax,0x3
// 39: 48 89 ec mov rsp,rbp
// 3c: 5d pop rbp
// 3d: c3 ret
// 3e: b8 01 00 00 00 mov eax,0x1
// 43: 48 89 ec mov rsp,rbp
// 46: 5d pop rbp
// 47: c3 ret
// 48: b8 02 00 00 00 mov eax,0x2
// 4d: 48 89 ec mov rsp,rbp
// 50: 5d pop rbp
// 51: c3 ret
// 0: 83 ff 02 cmp edi,0x2
// 3: 0f 83 27 00 00 00 jae 0x30
// 9: 44 8b d7 mov r10d,edi
// c: 41 b9 00 00 00 00 mov r9d,0x0
// 12: 4d 0f 43 d1 cmovae r10,r9
// 16: 4c 8d 0d 0b 00 00 00 lea r9,[rip+0xb] # 0x28
// 1d: 4f 63 54 91 00 movsxd r10,DWORD PTR [r9+r10*4+0x0]
// 22: 4d 01 d1 add r9,r10
// 25: 41 ff e1 jmp r9
// 28: 0e (bad)
// 29: 00 00 add BYTE PTR [rax],al
// 2b: 00 14 00 add BYTE PTR [rax+rax*1],dl
// 2e: 00 00 add BYTE PTR [rax],al
// 30: b8 03 00 00 00 mov eax,0x3
// 35: c3 ret
// 36: b8 01 00 00 00 mov eax,0x1
// 3b: c3 ret
// 3c: b8 02 00 00 00 mov eax,0x2
// 41: c3 ret

let golden = vec![
85, 72, 137, 229, 131, 255, 2, 15, 131, 39, 0, 0, 0, 68, 139, 215, 65, 185, 0, 0, 0, 0,
77, 15, 67, 209, 76, 141, 13, 11, 0, 0, 0, 79, 99, 84, 145, 0, 77, 1, 209, 65, 255,
225, 18, 0, 0, 0, 28, 0, 0, 0, 184, 3, 0, 0, 0, 72, 137, 236, 93, 195, 184, 1, 0, 0, 0,
72, 137, 236, 93, 195, 184, 2, 0, 0, 0, 72, 137, 236, 93, 195,
131, 255, 2, 15, 131, 39, 0, 0, 0, 68, 139, 215, 65, 185, 0, 0, 0, 0, 77, 15, 67, 209,
76, 141, 13, 11, 0, 0, 0, 79, 99, 84, 145, 0, 77, 1, 209, 65, 255, 225, 14, 0, 0, 0,
20, 0, 0, 0, 184, 3, 0, 0, 0, 195, 184, 1, 0, 0, 0, 195, 184, 2, 0, 0, 0, 195,
];

assert_eq!(code, &golden[..]);
Expand Down
4 changes: 0 additions & 4 deletions cranelift/filetests/filetests/egraph/not_a_load.clif
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -12,12 +12,8 @@ function u0:1302(i64) -> i64 system_v {
return v0
}

; pushq %rbp
; movq %rsp, %rbp
; block0:
; atomically { 64_bits_at_[%r9]) Add= %r10; %rax = old_value_at_[%r9]; %r11, %rflags = trash }
; movq %rdi, %rax
; movq %rbp, %rsp
; popq %rbp
; ret

37 changes: 0 additions & 37 deletions cranelift/filetests/filetests/isa/x64/amode-opt.clif
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8,12 +8,8 @@ block0(v0: i64, v1: i64):
return v3
}

; pushq %rbp
; movq %rsp, %rbp
; block0:
; movq 0(%rdi,%rsi,1), %rax
; movq %rbp, %rsp
; popq %rbp
; ret

function %amode_add_imm(i64) -> i64 {
Expand All @@ -24,12 +20,8 @@ block0(v0: i64):
return v3
}

; pushq %rbp
; movq %rsp, %rbp
; block0:
; movq 42(%rdi), %rax
; movq %rbp, %rsp
; popq %rbp
; ret

function %amode_add_imm_order(i64) -> i64 {
Expand All @@ -40,12 +32,8 @@ block0(v0: i64):
return v3
}

; pushq %rbp
; movq %rsp, %rbp
; block0:
; movq 42(%rdi), %rax
; movq %rbp, %rsp
; popq %rbp
; ret

function %amode_add_uext_imm(i64) -> i64 {
Expand All @@ -57,12 +45,8 @@ block0(v0: i64):
return v4
}

; pushq %rbp
; movq %rsp, %rbp
; block0:
; movq 42(%rdi), %rax
; movq %rbp, %rsp
; popq %rbp
; ret

function %amode_reg_reg_imm(i64, i64) -> i64 {
Expand All @@ -74,12 +58,8 @@ block0(v0: i64, v1: i64):
return v5
}

; pushq %rbp
; movq %rsp, %rbp
; block0:
; movq 320(%rdi,%rsi,1), %rax
; movq %rbp, %rsp
; popq %rbp
; ret

function %amode_reg_reg_imm_negative(i64, i64) -> i64 {
Expand All @@ -91,12 +71,8 @@ block0(v0: i64, v1: i64):
return v5
}

; pushq %rbp
; movq %rsp, %rbp
; block0:
; movq -1(%rdi,%rsi,1), %rax
; movq %rbp, %rsp
; popq %rbp
; ret

function %amode_reg_reg_imm_scaled(i64, i64) -> i64 {
Expand All @@ -109,15 +85,10 @@ block0(v0: i64, v1: i64):
return v6
}

; pushq %rbp
; movq %rsp, %rbp
; block0:
; movq -1(%rdi,%rsi,8), %rax
; movq %rbp, %rsp
; popq %rbp
; ret


function %amode_reg_reg_imm_uext_scaled(i64, i32) -> i64 {
block0(v0: i64, v1: i32):
v2 = iconst.i64 -1
Expand All @@ -129,13 +100,9 @@ block0(v0: i64, v1: i32):
return v7
}

; pushq %rbp
; movq %rsp, %rbp
; block0:
; movl %esi, %ecx
; movq -1(%rdi,%rcx,8), %rax
; movq %rbp, %rsp
; popq %rbp
; ret

function %amode_reg_reg_imm_uext_scaled_add(i64, i32, i32) -> i64 {
Expand All @@ -150,13 +117,9 @@ block0(v0: i64, v1: i32, v2: i32):
return v9
}

; pushq %rbp
; movq %rsp, %rbp
; block0:
; movq %rsi, %r8
; addl %r8d, %edx, %r8d
; movq -1(%rdi,%r8,4), %rax
; movq %rbp, %rsp
; popq %rbp
; ret

4 changes: 0 additions & 4 deletions cranelift/filetests/filetests/isa/x64/basic.clif
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,12 +7,8 @@ block0(v0: i32, v1: i32):
return v2
}

; pushq %rbp
; movq %rsp, %rbp
; block0:
; movq %rdi, %rax
; addl %eax, %esi, %eax
; movq %rbp, %rsp
; popq %rbp
; ret

16 changes: 0 additions & 16 deletions cranelift/filetests/filetests/isa/x64/bitcast.clif
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,12 +7,8 @@ block0(v0: f32):
return v1
}

; pushq %rbp
; movq %rsp, %rbp
; block0:
; movd %xmm0, %eax
; movq %rbp, %rsp
; popq %rbp
; ret

function %f2(i32) -> f32 {
Expand All @@ -21,12 +17,8 @@ block0(v0: i32):
return v1
}

; pushq %rbp
; movq %rsp, %rbp
; block0:
; movd %edi, %xmm0
; movq %rbp, %rsp
; popq %rbp
; ret

function %f3(f64) -> i64 {
Expand All @@ -35,12 +27,8 @@ block0(v0: f64):
return v1
}

; pushq %rbp
; movq %rsp, %rbp
; block0:
; movq %xmm0, %rax
; movq %rbp, %rsp
; popq %rbp
; ret

function %f4(i64) -> f64 {
Expand All @@ -49,11 +37,7 @@ block0(v0: i64):
return v1
}

; pushq %rbp
; movq %rsp, %rbp
; block0:
; movq %rdi, %xmm0
; movq %rbp, %rsp
; popq %rbp
; ret

Loading