github
/
cranelift
mirror of https://github.com/bytecodealliance/wasmtime.git
1
0
Fork 0
Code Issues Projects Releases Wiki Activity
Browse Source

Rework the switch module in cranelift-frontend in terms of brif (#5644) 

Rework the compilation strategy for switch to:
* use brif instead of brz and brnz
* generate tables inline, rather than delyaing them to after the decision tree has been generated
* avoid allocating new vectors by using slices into the sorted contiguous ranges
* avoid generating some unconditional jumps
* output differences in test output using the similar crate for easier debugging
pull/5651/head
Trevor Elliott 2 years ago
committed by GitHub
parent
0f8393508a
commit
b47006d432
No known key found for this signature in database GPG Key ID: 4AEE18F83AFDEB23
3 changed files with 153 additions and 185 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 1
      Cargo.lock
  2. 3
      cranelift/frontend/Cargo.toml
  3. 334
      cranelift/frontend/src/switch.rs

1
Cargo.lock
View File

@ -622,6 +622,7 @@ dependencies = [
"cranelift-codegen",
"hashbrown",
"log",
"similar",
"smallvec",
"target-lexicon",
]

3
cranelift/frontend/Cargo.toml
View File

@ -17,6 +17,9 @@ log = { workspace = true }
hashbrown = { workspace = true, optional = true }
smallvec = { workspace = true }
[dev-dependencies]
similar = { workspace = true }
[features]
default = ["std"]
std = ["cranelift-codegen/std"]

334
cranelift/frontend/src/switch.rs
View File

@ -108,27 +108,19 @@ impl Switch {
}
/// Binary search for the right `ContiguousCaseRange`.
fn build_search_tree(
fn build_search_tree<'a>(
bx: &mut FunctionBuilder,
val: Value,
otherwise: Block,
contiguous_case_ranges: Vec<ContiguousCaseRange>,
) -> Vec<(EntryIndex, Block, Vec<Block>)> {
let mut cases_and_jt_blocks = Vec::new();
contiguous_case_ranges: &'a [ContiguousCaseRange],
) {
// Avoid allocation in the common case
if contiguous_case_ranges.len() <= 3 {
Self::build_search_branches(
bx,
val,
otherwise,
contiguous_case_ranges,
&mut cases_and_jt_blocks,
);
return cases_and_jt_blocks;
Self::build_search_branches(bx, val, otherwise, contiguous_case_ranges);
return;
}
let mut stack: Vec<(Option<Block>, Vec<ContiguousCaseRange>)> = Vec::new();
let mut stack = Vec::new();
stack.push((None, contiguous_case_ranges));
while let Some((block, contiguous_case_ranges)) = stack.pop() {
@ -137,17 +129,10 @@ impl Switch {
}
if contiguous_case_ranges.len() <= 3 {
Self::build_search_branches(
bx,
val,
otherwise,
contiguous_case_ranges,
&mut cases_and_jt_blocks,
);
Self::build_search_branches(bx, val, otherwise, contiguous_case_ranges);
} else {
let split_point = contiguous_case_ranges.len() / 2;
let mut left = contiguous_case_ranges;
let right = left.split_off(split_point);
let (left, right) = contiguous_case_ranges.split_at(split_point);
let left_block = bx.create_block();
let right_block = bx.create_block();
@ -155,8 +140,8 @@ impl Switch {
let first_index = right[0].first_index;
let should_take_right_side =
icmp_imm_u128(bx, IntCC::UnsignedGreaterThanOrEqual, val, first_index);
bx.ins().brnz(should_take_right_side, right_block, &[]);
bx.ins().jump(left_block, &[]);
bx.ins()
.brif(should_take_right_side, right_block, &[], left_block, &[]);
bx.seal_block(left_block);
bx.seal_block(right_block);
@ -165,126 +150,108 @@ impl Switch {
stack.push((Some(right_block), right));
}
}
cases_and_jt_blocks
}
/// Linear search for the right `ContiguousCaseRange`.
fn build_search_branches(
fn build_search_branches<'a>(
bx: &mut FunctionBuilder,
val: Value,
otherwise: Block,
contiguous_case_ranges: Vec<ContiguousCaseRange>,
cases_and_jt_blocks: &mut Vec<(EntryIndex, Block, Vec<Block>)>,
contiguous_case_ranges: &'a [ContiguousCaseRange],
) {
let mut was_branch = false;
let ins_fallthrough_jump = |was_branch: bool, bx: &mut FunctionBuilder| {
if was_branch {
let block = bx.create_block();
bx.ins().jump(block, &[]);
bx.seal_block(block);
bx.switch_to_block(block);
}
};
for ContiguousCaseRange {
first_index,
blocks,
} in contiguous_case_ranges.into_iter().rev()
{
match (blocks.len(), first_index) {
(1, 0) => {
ins_fallthrough_jump(was_branch, bx);
bx.ins().brz(val, blocks[0], &[]);
}
(1, _) => {
ins_fallthrough_jump(was_branch, bx);
let is_good_val = icmp_imm_u128(bx, IntCC::Equal, val, first_index);
bx.ins().brnz(is_good_val, blocks[0], &[]);
}
(_, 0) => {
// if `first_index` is 0, then `icmp_imm uge val, first_index` is trivially true
let jt_block = bx.create_block();
bx.ins().jump(jt_block, &[]);
bx.seal_block(jt_block);
cases_and_jt_blocks.push((first_index, jt_block, blocks));
// `jump otherwise` below must not be hit, because the current block has been
// filled above. This is the last iteration anyway, as 0 is the smallest
// unsigned int, so just return here.
return;
let last_ix = contiguous_case_ranges.len() - 1;
for (ix, range) in contiguous_case_ranges.iter().rev().enumerate() {
let alternate = if ix == last_ix {
otherwise
} else {
bx.create_block()
};
if range.first_index == 0 {
assert_eq!(alternate, otherwise);
if let Some(block) = range.single_block() {
bx.ins().brif(val, otherwise, &[], block, &[]);
} else {
Self::build_jump_table(bx, val, otherwise, 0, &range.blocks);
}
(_, _) => {
ins_fallthrough_jump(was_branch, bx);
} else {
if let Some(block) = range.single_block() {
let is_good_val = icmp_imm_u128(bx, IntCC::Equal, val, range.first_index);
bx.ins().brif(is_good_val, block, &[], alternate, &[]);
} else {
let is_good_val = icmp_imm_u128(
bx,
IntCC::UnsignedGreaterThanOrEqual,
val,
range.first_index,
);
let jt_block = bx.create_block();
let is_good_val =
icmp_imm_u128(bx, IntCC::UnsignedGreaterThanOrEqual, val, first_index);
bx.ins().brnz(is_good_val, jt_block, &[]);
bx.ins().brif(is_good_val, jt_block, &[], alternate, &[]);
bx.seal_block(jt_block);
cases_and_jt_blocks.push((first_index, jt_block, blocks));
bx.switch_to_block(jt_block);
Self::build_jump_table(bx, val, otherwise, range.first_index, &range.blocks);
}
}
was_branch = true;
}
bx.ins().jump(otherwise, &[]);
if alternate != otherwise {
bx.seal_block(alternate);
bx.switch_to_block(alternate);
}
}
}
/// For every item in `cases_and_jt_blocks` this will create a jump table in the specified block.
fn build_jump_tables(
fn build_jump_table(
bx: &mut FunctionBuilder,
val: Value,
otherwise: Block,
cases_and_jt_blocks: Vec<(EntryIndex, Block, Vec<Block>)>,
first_index: EntryIndex,
blocks: &[Block],
) {
for (first_index, jt_block, blocks) in cases_and_jt_blocks.into_iter().rev() {
// There are currently no 128bit systems supported by rustc, but once we do ensure that
// we don't silently ignore a part of the jump table for 128bit integers on 128bit systems.
assert!(
u32::try_from(blocks.len()).is_ok(),
"Jump tables bigger than 2^32-1 are not yet supported"
);
// There are currently no 128bit systems supported by rustc, but once we do ensure that
// we don't silently ignore a part of the jump table for 128bit integers on 128bit systems.
assert!(
u32::try_from(blocks.len()).is_ok(),
"Jump tables bigger than 2^32-1 are not yet supported"
);
let mut jt_data = JumpTableData::new();
for block in blocks {
jt_data.push_entry(block);
}
let jump_table = bx.create_jump_table(jt_data);
let jt_data = JumpTableData::with_blocks(Vec::from(blocks));
let jump_table = bx.create_jump_table(jt_data);
bx.switch_to_block(jt_block);
let discr = if first_index == 0 {
val
let discr = if first_index == 0 {
val
} else {
if let Ok(first_index) = u64::try_from(first_index) {
bx.ins().iadd_imm(val, (first_index as i64).wrapping_neg())
} else {
if let Ok(first_index) = u64::try_from(first_index) {
bx.ins().iadd_imm(val, (first_index as i64).wrapping_neg())
} else {
let (lsb, msb) = (first_index as u64, (first_index >> 64) as u64);
let lsb = bx.ins().iconst(types::I64, lsb as i64);
let msb = bx.ins().iconst(types::I64, msb as i64);
let index = bx.ins().iconcat(lsb, msb);
bx.ins().isub(val, index)
}
};
let (lsb, msb) = (first_index as u64, (first_index >> 64) as u64);
let lsb = bx.ins().iconst(types::I64, lsb as i64);
let msb = bx.ins().iconst(types::I64, msb as i64);
let index = bx.ins().iconcat(lsb, msb);
bx.ins().isub(val, index)
}
};
let discr = match bx.func.dfg.value_type(discr).bits() {
bits if bits > 32 => {
// Check for overflow of cast to u32. This is the max supported jump table entries.
let new_block = bx.create_block();
let bigger_than_u32 =
bx.ins()
.icmp_imm(IntCC::UnsignedGreaterThan, discr, u32::MAX as i64);
bx.ins().brnz(bigger_than_u32, otherwise, &[]);
bx.ins().jump(new_block, &[]);
bx.seal_block(new_block);
bx.switch_to_block(new_block);
// Cast to i32, as br_table is not implemented for i64/i128
bx.ins().ireduce(types::I32, discr)
}
bits if bits < 32 => bx.ins().uextend(types::I32, discr),
_ => discr,
};
let discr = match bx.func.dfg.value_type(discr).bits() {
bits if bits > 32 => {
// Check for overflow of cast to u32. This is the max supported jump table entries.
let new_block = bx.create_block();
let bigger_than_u32 =
bx.ins()
.icmp_imm(IntCC::UnsignedGreaterThan, discr, u32::MAX as i64);
bx.ins()
.brif(bigger_than_u32, otherwise, &[], new_block, &[]);
bx.seal_block(new_block);
bx.switch_to_block(new_block);
// Cast to i32, as br_table is not implemented for i64/i128
bx.ins().ireduce(types::I32, discr)
}
bits if bits < 32 => bx.ins().uextend(types::I32, discr),
_ => discr,
};
bx.ins().br_table(discr, otherwise, jump_table);
}
bx.ins().br_table(discr, otherwise, jump_table);
}
/// Build the switch
@ -307,9 +274,7 @@ impl Switch {
}
let contiguous_case_ranges = self.collect_contiguous_case_ranges();
let cases_and_jt_blocks =
Self::build_search_tree(bx, val, otherwise, contiguous_case_ranges);
Self::build_jump_tables(bx, val, otherwise, cases_and_jt_blocks);
Self::build_search_tree(bx, val, otherwise, &contiguous_case_ranges);
}
}
@ -351,6 +316,15 @@ impl ContiguousCaseRange {
blocks: Vec::new(),
}
}
/// Returns `Some` block when there is only a single block in this range.
fn single_block(&self) -> Option<Block> {
if self.blocks.len() == 1 {
Some(self.blocks[0])
} else {
None
}
}
}
#[cfg(test)]
@ -384,43 +358,52 @@ mod tests {
}};
}
macro_rules! assert_eq_output {
($actual:ident, $expected:literal) => {
if $actual != $expected {
assert!(
false,
"\n{}",
similar::TextDiff::from_lines($expected, &$actual)
.unified_diff()
.header("expected", "actual")
);
}
};
}
#[test]
fn switch_zero() {
let func = setup!(0, [0,]);
assert_eq!(
assert_eq_output!(
func,
"block0:
v0 = iconst.i8 0
brz v0, block1 ; v0 = 0
jump block0"
brif v0, block0, block1 ; v0 = 0"
);
}
#[test]
fn switch_single() {
let func = setup!(0, [1,]);
assert_eq!(
assert_eq_output!(
func,
"block0:
v0 = iconst.i8 0
v1 = icmp_imm eq v0, 1 ; v0 = 0
brnz v1, block1
jump block0"
brif v1, block1, block0"
);
}
#[test]
fn switch_bool() {
let func = setup!(0, [0, 1,]);
assert_eq!(
assert_eq_output!(
func,
" jt0 = jump_table [block1, block2]
block0:
v0 = iconst.i8 0
jump block3
block3:
v1 = uextend.i32 v0 ; v0 = 0
br_table v1, block0, jt0"
);
@ -429,56 +412,50 @@ block3:
#[test]
fn switch_two_gap() {
let func = setup!(0, [0, 2,]);
assert_eq!(
assert_eq_output!(
func,
"block0:
v0 = iconst.i8 0
v1 = icmp_imm eq v0, 2 ; v0 = 0
brnz v1, block2
jump block3
brif v1, block2, block3
block3:
brz.i8 v0, block1 ; v0 = 0
jump block0"
brif.i8 v0, block0, block1 ; v0 = 0"
);
}
#[test]
fn switch_many() {
let func = setup!(0, [0, 1, 5, 7, 10, 11, 12,]);
assert_eq!(
assert_eq_output!(
func,
" jt0 = jump_table [block1, block2]
jt1 = jump_table [block5, block6, block7]
" jt0 = jump_table [block5, block6, block7]
jt1 = jump_table [block1, block2]
block0:
v0 = iconst.i8 0
v1 = icmp_imm uge v0, 7 ; v0 = 0
brnz v1, block9
jump block8
brif v1, block9, block8
block9:
v2 = icmp_imm.i8 uge v0, 10 ; v0 = 0
brnz v2, block10
jump block11
brif v2, block11, block10
block11:
v3 = icmp_imm.i8 eq v0, 7 ; v0 = 0
brnz v3, block4
jump block0
v3 = iadd_imm.i8 v0, -10 ; v0 = 0
v4 = uextend.i32 v3
br_table v4, block0, jt0
block10:
v5 = icmp_imm.i8 eq v0, 7 ; v0 = 0
brif v5, block4, block0
block8:
v4 = icmp_imm.i8 eq v0, 5 ; v0 = 0
brnz v4, block3
jump block12
v6 = icmp_imm.i8 eq v0, 5 ; v0 = 0
brif v6, block3, block12
block12:
v5 = uextend.i32 v0 ; v0 = 0
br_table v5, block0, jt0
block10:
v6 = iadd_imm.i8 v0, -10 ; v0 = 0
v7 = uextend.i32 v6
v7 = uextend.i32 v0 ; v0 = 0
br_table v7, block0, jt1"
);
}
@ -486,51 +463,46 @@ block10:
#[test]
fn switch_min_index_value() {
let func = setup!(0, [i8::MIN as u8 as u128, 1,]);
assert_eq!(
assert_eq_output!(
func,
"block0:
v0 = iconst.i8 0
v1 = icmp_imm eq v0, 128 ; v0 = 0
brnz v1, block1
jump block3
brif v1, block1, block3
block3:
v2 = icmp_imm.i8 eq v0, 1 ; v0 = 0
brnz v2, block2
jump block0"
brif v2, block2, block0"
);
}
#[test]
fn switch_max_index_value() {
let func = setup!(0, [i8::MAX as u8 as u128, 1,]);
assert_eq!(
assert_eq_output!(
func,
"block0:
v0 = iconst.i8 0
v1 = icmp_imm eq v0, 127 ; v0 = 0
brnz v1, block1
jump block3
brif v1, block1, block3
block3:
v2 = icmp_imm.i8 eq v0, 1 ; v0 = 0
brnz v2, block2
jump block0"
brif v2, block2, block0"
)
}
#[test]
fn switch_optimal_codegen() {
let func = setup!(0, [-1i8 as u8 as u128, 0, 1,]);
assert_eq!(
assert_eq_output!(
func,
" jt0 = jump_table [block2, block3]
block0:
v0 = iconst.i8 0
v1 = icmp_imm eq v0, 255 ; v0 = 0
brnz v1, block1
jump block4
brif v1, block1, block4
block4:
v2 = uextend.i32 v0 ; v0 = 0
@ -617,20 +589,16 @@ block4:
.trim_start_matches("function u0:0() fast {\n")
.trim_end_matches("\n}\n")
.to_string();
assert_eq!(
assert_eq_output!(
func,
" jt0 = jump_table [block2, block1]
block0:
v0 = iconst.i64 0
jump block4
v1 = icmp_imm ugt v0, 0xffff_ffff ; v0 = 0
brif v1, block3, block4
block4:
v1 = icmp_imm.i64 ugt v0, 0xffff_ffff ; v0 = 0
brnz v1, block3
jump block5
block5:
v2 = ireduce.i32 v0 ; v0 = 0
br_table v2, block3, jt0"
);
@ -659,21 +627,17 @@ block5:
.trim_start_matches("function u0:0() fast {\n")
.trim_end_matches("\n}\n")
.to_string();
assert_eq!(
assert_eq_output!(
func,
" jt0 = jump_table [block2, block1]
block0:
v0 = iconst.i64 0
v1 = uextend.i128 v0 ; v0 = 0
jump block4
v2 = icmp_imm ugt v1, 0xffff_ffff
brif v2, block3, block4
block4:
v2 = icmp_imm.i128 ugt v1, 0xffff_ffff
brnz v2, block3
jump block5
block5:
v3 = ireduce.i32 v1
br_table v3, block3, jt0"
);

Write Preview
Loading…
Cancel
Save