Struct cranelift_frontend::FunctionBuilder
source · pub struct FunctionBuilder<'a> {
pub func: &'a mut Function,
/* private fields */
}
Expand description
Temporary object used to build a single Cranelift IR Function
.
Fields§
§func: &'a mut Function
The function currently being built. This field is public so the function can be re-borrowed.
Implementations§
source§impl<'a> FunctionBuilder<'a>
impl<'a> FunctionBuilder<'a>
This module allows you to create a function in Cranelift IR in a straightforward way, hiding all the complexity of its internal representation.
The module is parametrized by one type which is the representation of variables in your
origin language. It offers a way to conveniently append instruction to your program flow.
You are responsible to split your instruction flow into extended blocks (declared with
create_block
) whose properties are:
- branch and jump instructions can only point at the top of extended blocks;
- the last instruction of each block is a terminator instruction which has no natural successor, and those instructions can only appear at the end of extended blocks.
The parameters of Cranelift IR instructions are Cranelift IR values, which can only be created
as results of other Cranelift IR instructions. To be able to create variables redefined multiple
times in your program, use the def_var
and use_var
command, that will maintain the
correspondence between your variables and Cranelift IR SSA values.
The first block for which you call switch_to_block
will be assumed to be the beginning of
the function.
At creation, a FunctionBuilder
instance borrows an already allocated Function
which it
modifies with the information stored in the mutable borrowed
FunctionBuilderContext
. The function passed in
argument should be newly created with
Function::with_name_signature()
, whereas the
FunctionBuilderContext
can be kept as is between two function translations.
Errors
The functions below will panic in debug mode whenever you try to modify the Cranelift IR
function in a way that violate the coherence of the code. For instance: switching to a new
Block
when you haven’t filled the current one with a terminator instruction, inserting a
return instruction with arguments that don’t match the function’s signature.
sourcepub fn new(
func: &'a mut Function,
func_ctx: &'a mut FunctionBuilderContext
) -> Self
pub fn new(
func: &'a mut Function,
func_ctx: &'a mut FunctionBuilderContext
) -> Self
Creates a new FunctionBuilder structure that will operate on a Function
using a
FunctionBuilderContext
.
sourcepub fn current_block(&self) -> Option<Block>
pub fn current_block(&self) -> Option<Block>
Get the block that this builder is currently at.
sourcepub fn set_srcloc(&mut self, srcloc: SourceLoc)
pub fn set_srcloc(&mut self, srcloc: SourceLoc)
Set the source location that should be assigned to all new instructions.
sourcepub fn create_block(&mut self) -> Block
pub fn create_block(&mut self) -> Block
Creates a new Block
and returns its reference.
Examples found in repository?
111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288
fn build_search_tree(
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();
// 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;
}
let mut stack: Vec<(Option<Block>, Vec<ContiguousCaseRange>)> = Vec::new();
stack.push((None, contiguous_case_ranges));
while let Some((block, contiguous_case_ranges)) = stack.pop() {
if let Some(block) = block {
bx.switch_to_block(block);
}
if contiguous_case_ranges.len() <= 3 {
Self::build_search_branches(
bx,
val,
otherwise,
contiguous_case_ranges,
&mut cases_and_jt_blocks,
);
} else {
let split_point = contiguous_case_ranges.len() / 2;
let mut left = contiguous_case_ranges;
let right = left.split_off(split_point);
let left_block = bx.create_block();
let right_block = bx.create_block();
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.seal_block(left_block);
bx.seal_block(right_block);
stack.push((Some(left_block), left));
stack.push((Some(right_block), right));
}
}
cases_and_jt_blocks
}
/// Linear search for the right `ContiguousCaseRange`.
fn build_search_branches(
bx: &mut FunctionBuilder,
val: Value,
otherwise: Block,
contiguous_case_ranges: Vec<ContiguousCaseRange>,
cases_and_jt_blocks: &mut Vec<(EntryIndex, Block, Vec<Block>)>,
) {
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;
}
(_, _) => {
ins_fallthrough_jump(was_branch, bx);
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.seal_block(jt_block);
cases_and_jt_blocks.push((first_index, jt_block, blocks));
}
}
was_branch = true;
}
bx.ins().jump(otherwise, &[]);
}
/// For every item in `cases_and_jt_blocks` this will create a jump table in the specified block.
fn build_jump_tables(
bx: &mut FunctionBuilder,
val: Value,
otherwise: Block,
cases_and_jt_blocks: Vec<(EntryIndex, Block, Vec<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"
);
let mut jt_data = JumpTableData::new();
for block in blocks {
jt_data.push_entry(block);
}
let jump_table = bx.create_jump_table(jt_data);
bx.switch_to_block(jt_block);
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 {
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,
};
bx.ins().br_table(discr, otherwise, jump_table);
}
}
sourcepub fn set_cold_block(&mut self, block: Block)
pub fn set_cold_block(&mut self, block: Block)
Mark a block as “cold”.
This will try to move it out of the ordinary path of execution when lowered to machine code.
sourcepub fn insert_block_after(&mut self, block: Block, after: Block)
pub fn insert_block_after(&mut self, block: Block, after: Block)
Insert block
in the layout after the existing block after
.
sourcepub fn switch_to_block(&mut self, block: Block)
pub fn switch_to_block(&mut self, block: Block)
After the call to this function, new instructions will be inserted into the designated block, in the order they are declared. You must declare the types of the Block arguments you will use here.
When inserting the terminator instruction (which doesn’t have a fallthrough to its immediate successor), the block will be declared filled and it will not be possible to append instructions to it.
Examples found in repository?
111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288
fn build_search_tree(
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();
// 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;
}
let mut stack: Vec<(Option<Block>, Vec<ContiguousCaseRange>)> = Vec::new();
stack.push((None, contiguous_case_ranges));
while let Some((block, contiguous_case_ranges)) = stack.pop() {
if let Some(block) = block {
bx.switch_to_block(block);
}
if contiguous_case_ranges.len() <= 3 {
Self::build_search_branches(
bx,
val,
otherwise,
contiguous_case_ranges,
&mut cases_and_jt_blocks,
);
} else {
let split_point = contiguous_case_ranges.len() / 2;
let mut left = contiguous_case_ranges;
let right = left.split_off(split_point);
let left_block = bx.create_block();
let right_block = bx.create_block();
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.seal_block(left_block);
bx.seal_block(right_block);
stack.push((Some(left_block), left));
stack.push((Some(right_block), right));
}
}
cases_and_jt_blocks
}
/// Linear search for the right `ContiguousCaseRange`.
fn build_search_branches(
bx: &mut FunctionBuilder,
val: Value,
otherwise: Block,
contiguous_case_ranges: Vec<ContiguousCaseRange>,
cases_and_jt_blocks: &mut Vec<(EntryIndex, Block, Vec<Block>)>,
) {
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;
}
(_, _) => {
ins_fallthrough_jump(was_branch, bx);
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.seal_block(jt_block);
cases_and_jt_blocks.push((first_index, jt_block, blocks));
}
}
was_branch = true;
}
bx.ins().jump(otherwise, &[]);
}
/// For every item in `cases_and_jt_blocks` this will create a jump table in the specified block.
fn build_jump_tables(
bx: &mut FunctionBuilder,
val: Value,
otherwise: Block,
cases_and_jt_blocks: Vec<(EntryIndex, Block, Vec<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"
);
let mut jt_data = JumpTableData::new();
for block in blocks {
jt_data.push_entry(block);
}
let jump_table = bx.create_jump_table(jt_data);
bx.switch_to_block(jt_block);
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 {
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,
};
bx.ins().br_table(discr, otherwise, jump_table);
}
}
sourcepub fn seal_block(&mut self, block: Block)
pub fn seal_block(&mut self, block: Block)
Declares that all the predecessors of this block are known.
Function to call with block
as soon as the last branch instruction to block
has been
created. Forgetting to call this method on every block will cause inconsistencies in the
produced functions.
Examples found in repository?
111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288
fn build_search_tree(
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();
// 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;
}
let mut stack: Vec<(Option<Block>, Vec<ContiguousCaseRange>)> = Vec::new();
stack.push((None, contiguous_case_ranges));
while let Some((block, contiguous_case_ranges)) = stack.pop() {
if let Some(block) = block {
bx.switch_to_block(block);
}
if contiguous_case_ranges.len() <= 3 {
Self::build_search_branches(
bx,
val,
otherwise,
contiguous_case_ranges,
&mut cases_and_jt_blocks,
);
} else {
let split_point = contiguous_case_ranges.len() / 2;
let mut left = contiguous_case_ranges;
let right = left.split_off(split_point);
let left_block = bx.create_block();
let right_block = bx.create_block();
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.seal_block(left_block);
bx.seal_block(right_block);
stack.push((Some(left_block), left));
stack.push((Some(right_block), right));
}
}
cases_and_jt_blocks
}
/// Linear search for the right `ContiguousCaseRange`.
fn build_search_branches(
bx: &mut FunctionBuilder,
val: Value,
otherwise: Block,
contiguous_case_ranges: Vec<ContiguousCaseRange>,
cases_and_jt_blocks: &mut Vec<(EntryIndex, Block, Vec<Block>)>,
) {
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;
}
(_, _) => {
ins_fallthrough_jump(was_branch, bx);
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.seal_block(jt_block);
cases_and_jt_blocks.push((first_index, jt_block, blocks));
}
}
was_branch = true;
}
bx.ins().jump(otherwise, &[]);
}
/// For every item in `cases_and_jt_blocks` this will create a jump table in the specified block.
fn build_jump_tables(
bx: &mut FunctionBuilder,
val: Value,
otherwise: Block,
cases_and_jt_blocks: Vec<(EntryIndex, Block, Vec<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"
);
let mut jt_data = JumpTableData::new();
for block in blocks {
jt_data.push_entry(block);
}
let jump_table = bx.create_jump_table(jt_data);
bx.switch_to_block(jt_block);
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 {
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,
};
bx.ins().br_table(discr, otherwise, jump_table);
}
}
sourcepub fn seal_all_blocks(&mut self)
pub fn seal_all_blocks(&mut self)
Effectively calls seal_block on all unsealed blocks in the function.
It’s more efficient to seal Block
s as soon as possible, during
translation, but for frontends where this is impractical to do, this
function can be used at the end of translating all blocks to ensure
that everything is sealed.
sourcepub fn try_declare_var(
&mut self,
var: Variable,
ty: Type
) -> Result<(), DeclareVariableError>
pub fn try_declare_var(
&mut self,
var: Variable,
ty: Type
) -> Result<(), DeclareVariableError>
Declares the type of a variable, so that it can be used later (by calling
FunctionBuilder::use_var
). This function will return an error if it
was not possible to use the variable.
sourcepub fn declare_var(&mut self, var: Variable, ty: Type)
pub fn declare_var(&mut self, var: Variable, ty: Type)
In order to use a variable (by calling FunctionBuilder::use_var
), you need
to first declare its type with this method.
sourcepub fn try_use_var(&mut self, var: Variable) -> Result<Value, UseVariableError>
pub fn try_use_var(&mut self, var: Variable) -> Result<Value, UseVariableError>
Returns the Cranelift IR necessary to use a previously defined user variable, returning an error if this is not possible.
sourcepub fn use_var(&mut self, var: Variable) -> Value
pub fn use_var(&mut self, var: Variable) -> Value
Returns the Cranelift IR value corresponding to the utilization at the current program position of a previously defined user variable.
sourcepub fn try_def_var(
&mut self,
var: Variable,
val: Value
) -> Result<(), DefVariableError>
pub fn try_def_var(
&mut self,
var: Variable,
val: Value
) -> Result<(), DefVariableError>
Registers a new definition of a user variable. This function will return an error if the value supplied does not match the type the variable was declared to have.
Examples found in repository?
441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457
pub fn def_var(&mut self, var: Variable, val: Value) {
self.try_def_var(var, val)
.unwrap_or_else(|error| match error {
DefVariableError::TypeMismatch(var, val) => {
panic!(
"declared type of variable {:?} doesn't match type of value {}",
var, val
);
}
DefVariableError::DefinedBeforeDeclared(var) => {
panic!(
"variable {:?} is used but its type has not been declared",
var
);
}
})
}
sourcepub fn def_var(&mut self, var: Variable, val: Value)
pub fn def_var(&mut self, var: Variable, val: Value)
Register a new definition of a user variable. The type of the value must be the same as the type registered for the variable.
sourcepub fn set_val_label(&mut self, val: Value, label: ValueLabel)
pub fn set_val_label(&mut self, val: Value, label: ValueLabel)
Set label for Value
This will not do anything unless func.dfg.collect_debug_info
is called first.
sourcepub fn create_jump_table(&mut self, data: JumpTableData) -> JumpTable
pub fn create_jump_table(&mut self, data: JumpTableData) -> JumpTable
Creates a jump table in the function, to be used by br_table
instructions.
Examples found in repository?
232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288
fn build_jump_tables(
bx: &mut FunctionBuilder,
val: Value,
otherwise: Block,
cases_and_jt_blocks: Vec<(EntryIndex, Block, Vec<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"
);
let mut jt_data = JumpTableData::new();
for block in blocks {
jt_data.push_entry(block);
}
let jump_table = bx.create_jump_table(jt_data);
bx.switch_to_block(jt_block);
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 {
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,
};
bx.ins().br_table(discr, otherwise, jump_table);
}
}
sourcepub fn create_sized_stack_slot(&mut self, data: StackSlotData) -> StackSlot
pub fn create_sized_stack_slot(&mut self, data: StackSlotData) -> StackSlot
Creates a sized stack slot in the function, to be used by stack_load
, stack_store
and
stack_addr
instructions.
sourcepub fn create_dynamic_stack_slot(
&mut self,
data: DynamicStackSlotData
) -> DynamicStackSlot
pub fn create_dynamic_stack_slot(
&mut self,
data: DynamicStackSlotData
) -> DynamicStackSlot
Creates a dynamic stack slot in the function, to be used by dynamic_stack_load
,
dynamic_stack_store
and dynamic_stack_addr
instructions.
sourcepub fn import_signature(&mut self, signature: Signature) -> SigRef
pub fn import_signature(&mut self, signature: Signature) -> SigRef
Adds a signature which can later be used to declare an external function import.
Examples found in repository?
729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984
pub fn call_memcpy(
&mut self,
config: TargetFrontendConfig,
dest: Value,
src: Value,
size: Value,
) {
let pointer_type = config.pointer_type();
let signature = {
let mut s = Signature::new(config.default_call_conv);
s.params.push(AbiParam::new(pointer_type));
s.params.push(AbiParam::new(pointer_type));
s.params.push(AbiParam::new(pointer_type));
self.import_signature(s)
};
let libc_memcpy = self.import_function(ExtFuncData {
name: ExternalName::LibCall(LibCall::Memcpy),
signature,
colocated: false,
});
self.ins().call(libc_memcpy, &[dest, src, size]);
}
/// Optimised memcpy or memmove for small copies.
///
/// # Codegen safety
///
/// The following properties must hold to prevent UB:
///
/// * `src_align` and `dest_align` are an upper-bound on the alignment of `src` respectively `dest`.
/// * If `non_overlapping` is true, then this must be correct.
pub fn emit_small_memory_copy(
&mut self,
config: TargetFrontendConfig,
dest: Value,
src: Value,
size: u64,
dest_align: u8,
src_align: u8,
non_overlapping: bool,
mut flags: MemFlags,
) {
// Currently the result of guess work, not actual profiling.
const THRESHOLD: u64 = 4;
if size == 0 {
return;
}
let access_size = greatest_divisible_power_of_two(size);
assert!(
access_size.is_power_of_two(),
"`size` is not a power of two"
);
assert!(
access_size >= u64::from(::core::cmp::min(src_align, dest_align)),
"`size` is smaller than `dest` and `src`'s alignment value."
);
let (access_size, int_type) = if access_size <= 8 {
(access_size, Type::int((access_size * 8) as u16).unwrap())
} else {
(8, types::I64)
};
let load_and_store_amount = size / access_size;
if load_and_store_amount > THRESHOLD {
let size_value = self.ins().iconst(config.pointer_type(), size as i64);
if non_overlapping {
self.call_memcpy(config, dest, src, size_value);
} else {
self.call_memmove(config, dest, src, size_value);
}
return;
}
if u64::from(src_align) >= access_size && u64::from(dest_align) >= access_size {
flags.set_aligned();
}
// Load all of the memory first. This is necessary in case `dest` overlaps.
// It can also improve performance a bit.
let registers: smallvec::SmallVec<[_; THRESHOLD as usize]> = (0..load_and_store_amount)
.map(|i| {
let offset = (access_size * i) as i32;
(self.ins().load(int_type, flags, src, offset), offset)
})
.collect();
for (value, offset) in registers {
self.ins().store(flags, value, dest, offset);
}
}
/// Calls libc.memset
///
/// Writes `size` bytes of i8 value `ch` to memory starting at `buffer`.
pub fn call_memset(
&mut self,
config: TargetFrontendConfig,
buffer: Value,
ch: Value,
size: Value,
) {
let pointer_type = config.pointer_type();
let signature = {
let mut s = Signature::new(config.default_call_conv);
s.params.push(AbiParam::new(pointer_type));
s.params.push(AbiParam::new(types::I32));
s.params.push(AbiParam::new(pointer_type));
self.import_signature(s)
};
let libc_memset = self.import_function(ExtFuncData {
name: ExternalName::LibCall(LibCall::Memset),
signature,
colocated: false,
});
let ch = self.ins().uextend(types::I32, ch);
self.ins().call(libc_memset, &[buffer, ch, size]);
}
/// Calls libc.memset
///
/// Writes `size` bytes of value `ch` to memory starting at `buffer`.
pub fn emit_small_memset(
&mut self,
config: TargetFrontendConfig,
buffer: Value,
ch: u8,
size: u64,
buffer_align: u8,
mut flags: MemFlags,
) {
// Currently the result of guess work, not actual profiling.
const THRESHOLD: u64 = 4;
if size == 0 {
return;
}
let access_size = greatest_divisible_power_of_two(size);
assert!(
access_size.is_power_of_two(),
"`size` is not a power of two"
);
assert!(
access_size >= u64::from(buffer_align),
"`size` is smaller than `dest` and `src`'s alignment value."
);
let (access_size, int_type) = if access_size <= 8 {
(access_size, Type::int((access_size * 8) as u16).unwrap())
} else {
(8, types::I64)
};
let load_and_store_amount = size / access_size;
if load_and_store_amount > THRESHOLD {
let ch = self.ins().iconst(types::I8, i64::from(ch));
let size = self.ins().iconst(config.pointer_type(), size as i64);
self.call_memset(config, buffer, ch, size);
} else {
if u64::from(buffer_align) >= access_size {
flags.set_aligned();
}
let ch = u64::from(ch);
let raw_value = if int_type == types::I64 {
ch * 0x0101010101010101_u64
} else if int_type == types::I32 {
ch * 0x01010101_u64
} else if int_type == types::I16 {
(ch << 8) | ch
} else {
assert_eq!(int_type, types::I8);
ch
};
let value = self.ins().iconst(int_type, raw_value as i64);
for i in 0..load_and_store_amount {
let offset = (access_size * i) as i32;
self.ins().store(flags, value, buffer, offset);
}
}
}
/// Calls libc.memmove
///
/// Copies `size` bytes from memory starting at `source` to memory starting
/// at `dest`. `source` is always read before writing to `dest`.
pub fn call_memmove(
&mut self,
config: TargetFrontendConfig,
dest: Value,
source: Value,
size: Value,
) {
let pointer_type = config.pointer_type();
let signature = {
let mut s = Signature::new(config.default_call_conv);
s.params.push(AbiParam::new(pointer_type));
s.params.push(AbiParam::new(pointer_type));
s.params.push(AbiParam::new(pointer_type));
self.import_signature(s)
};
let libc_memmove = self.import_function(ExtFuncData {
name: ExternalName::LibCall(LibCall::Memmove),
signature,
colocated: false,
});
self.ins().call(libc_memmove, &[dest, source, size]);
}
/// Calls libc.memcmp
///
/// Compares `size` bytes from memory starting at `left` to memory starting
/// at `right`. Returns `0` if all `n` bytes are equal. If the first difference
/// is at offset `i`, returns a positive integer if `ugt(left[i], right[i])`
/// and a negative integer if `ult(left[i], right[i])`.
///
/// Returns a C `int`, which is currently always [`types::I32`].
pub fn call_memcmp(
&mut self,
config: TargetFrontendConfig,
left: Value,
right: Value,
size: Value,
) -> Value {
let pointer_type = config.pointer_type();
let signature = {
let mut s = Signature::new(config.default_call_conv);
s.params.reserve(3);
s.params.push(AbiParam::new(pointer_type));
s.params.push(AbiParam::new(pointer_type));
s.params.push(AbiParam::new(pointer_type));
s.returns.push(AbiParam::new(types::I32));
self.import_signature(s)
};
let libc_memcmp = self.import_function(ExtFuncData {
name: ExternalName::LibCall(LibCall::Memcmp),
signature,
colocated: false,
});
let call = self.ins().call(libc_memcmp, &[left, right, size]);
self.func.dfg.first_result(call)
}
sourcepub fn import_function(&mut self, data: ExtFuncData) -> FuncRef
pub fn import_function(&mut self, data: ExtFuncData) -> FuncRef
Declare an external function import.
Examples found in repository?
729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984
pub fn call_memcpy(
&mut self,
config: TargetFrontendConfig,
dest: Value,
src: Value,
size: Value,
) {
let pointer_type = config.pointer_type();
let signature = {
let mut s = Signature::new(config.default_call_conv);
s.params.push(AbiParam::new(pointer_type));
s.params.push(AbiParam::new(pointer_type));
s.params.push(AbiParam::new(pointer_type));
self.import_signature(s)
};
let libc_memcpy = self.import_function(ExtFuncData {
name: ExternalName::LibCall(LibCall::Memcpy),
signature,
colocated: false,
});
self.ins().call(libc_memcpy, &[dest, src, size]);
}
/// Optimised memcpy or memmove for small copies.
///
/// # Codegen safety
///
/// The following properties must hold to prevent UB:
///
/// * `src_align` and `dest_align` are an upper-bound on the alignment of `src` respectively `dest`.
/// * If `non_overlapping` is true, then this must be correct.
pub fn emit_small_memory_copy(
&mut self,
config: TargetFrontendConfig,
dest: Value,
src: Value,
size: u64,
dest_align: u8,
src_align: u8,
non_overlapping: bool,
mut flags: MemFlags,
) {
// Currently the result of guess work, not actual profiling.
const THRESHOLD: u64 = 4;
if size == 0 {
return;
}
let access_size = greatest_divisible_power_of_two(size);
assert!(
access_size.is_power_of_two(),
"`size` is not a power of two"
);
assert!(
access_size >= u64::from(::core::cmp::min(src_align, dest_align)),
"`size` is smaller than `dest` and `src`'s alignment value."
);
let (access_size, int_type) = if access_size <= 8 {
(access_size, Type::int((access_size * 8) as u16).unwrap())
} else {
(8, types::I64)
};
let load_and_store_amount = size / access_size;
if load_and_store_amount > THRESHOLD {
let size_value = self.ins().iconst(config.pointer_type(), size as i64);
if non_overlapping {
self.call_memcpy(config, dest, src, size_value);
} else {
self.call_memmove(config, dest, src, size_value);
}
return;
}
if u64::from(src_align) >= access_size && u64::from(dest_align) >= access_size {
flags.set_aligned();
}
// Load all of the memory first. This is necessary in case `dest` overlaps.
// It can also improve performance a bit.
let registers: smallvec::SmallVec<[_; THRESHOLD as usize]> = (0..load_and_store_amount)
.map(|i| {
let offset = (access_size * i) as i32;
(self.ins().load(int_type, flags, src, offset), offset)
})
.collect();
for (value, offset) in registers {
self.ins().store(flags, value, dest, offset);
}
}
/// Calls libc.memset
///
/// Writes `size` bytes of i8 value `ch` to memory starting at `buffer`.
pub fn call_memset(
&mut self,
config: TargetFrontendConfig,
buffer: Value,
ch: Value,
size: Value,
) {
let pointer_type = config.pointer_type();
let signature = {
let mut s = Signature::new(config.default_call_conv);
s.params.push(AbiParam::new(pointer_type));
s.params.push(AbiParam::new(types::I32));
s.params.push(AbiParam::new(pointer_type));
self.import_signature(s)
};
let libc_memset = self.import_function(ExtFuncData {
name: ExternalName::LibCall(LibCall::Memset),
signature,
colocated: false,
});
let ch = self.ins().uextend(types::I32, ch);
self.ins().call(libc_memset, &[buffer, ch, size]);
}
/// Calls libc.memset
///
/// Writes `size` bytes of value `ch` to memory starting at `buffer`.
pub fn emit_small_memset(
&mut self,
config: TargetFrontendConfig,
buffer: Value,
ch: u8,
size: u64,
buffer_align: u8,
mut flags: MemFlags,
) {
// Currently the result of guess work, not actual profiling.
const THRESHOLD: u64 = 4;
if size == 0 {
return;
}
let access_size = greatest_divisible_power_of_two(size);
assert!(
access_size.is_power_of_two(),
"`size` is not a power of two"
);
assert!(
access_size >= u64::from(buffer_align),
"`size` is smaller than `dest` and `src`'s alignment value."
);
let (access_size, int_type) = if access_size <= 8 {
(access_size, Type::int((access_size * 8) as u16).unwrap())
} else {
(8, types::I64)
};
let load_and_store_amount = size / access_size;
if load_and_store_amount > THRESHOLD {
let ch = self.ins().iconst(types::I8, i64::from(ch));
let size = self.ins().iconst(config.pointer_type(), size as i64);
self.call_memset(config, buffer, ch, size);
} else {
if u64::from(buffer_align) >= access_size {
flags.set_aligned();
}
let ch = u64::from(ch);
let raw_value = if int_type == types::I64 {
ch * 0x0101010101010101_u64
} else if int_type == types::I32 {
ch * 0x01010101_u64
} else if int_type == types::I16 {
(ch << 8) | ch
} else {
assert_eq!(int_type, types::I8);
ch
};
let value = self.ins().iconst(int_type, raw_value as i64);
for i in 0..load_and_store_amount {
let offset = (access_size * i) as i32;
self.ins().store(flags, value, buffer, offset);
}
}
}
/// Calls libc.memmove
///
/// Copies `size` bytes from memory starting at `source` to memory starting
/// at `dest`. `source` is always read before writing to `dest`.
pub fn call_memmove(
&mut self,
config: TargetFrontendConfig,
dest: Value,
source: Value,
size: Value,
) {
let pointer_type = config.pointer_type();
let signature = {
let mut s = Signature::new(config.default_call_conv);
s.params.push(AbiParam::new(pointer_type));
s.params.push(AbiParam::new(pointer_type));
s.params.push(AbiParam::new(pointer_type));
self.import_signature(s)
};
let libc_memmove = self.import_function(ExtFuncData {
name: ExternalName::LibCall(LibCall::Memmove),
signature,
colocated: false,
});
self.ins().call(libc_memmove, &[dest, source, size]);
}
/// Calls libc.memcmp
///
/// Compares `size` bytes from memory starting at `left` to memory starting
/// at `right`. Returns `0` if all `n` bytes are equal. If the first difference
/// is at offset `i`, returns a positive integer if `ugt(left[i], right[i])`
/// and a negative integer if `ult(left[i], right[i])`.
///
/// Returns a C `int`, which is currently always [`types::I32`].
pub fn call_memcmp(
&mut self,
config: TargetFrontendConfig,
left: Value,
right: Value,
size: Value,
) -> Value {
let pointer_type = config.pointer_type();
let signature = {
let mut s = Signature::new(config.default_call_conv);
s.params.reserve(3);
s.params.push(AbiParam::new(pointer_type));
s.params.push(AbiParam::new(pointer_type));
s.params.push(AbiParam::new(pointer_type));
s.returns.push(AbiParam::new(types::I32));
self.import_signature(s)
};
let libc_memcmp = self.import_function(ExtFuncData {
name: ExternalName::LibCall(LibCall::Memcmp),
signature,
colocated: false,
});
let call = self.ins().call(libc_memcmp, &[left, right, size]);
self.func.dfg.first_result(call)
}
sourcepub fn create_global_value(&mut self, data: GlobalValueData) -> GlobalValue
pub fn create_global_value(&mut self, data: GlobalValueData) -> GlobalValue
Declares a global value accessible to the function.
sourcepub fn create_heap(&mut self, data: HeapData) -> Heap
pub fn create_heap(&mut self, data: HeapData) -> Heap
Declares a heap accessible to the function.
sourcepub fn ins<'short>(&'short mut self) -> FuncInstBuilder<'short, 'a>
pub fn ins<'short>(&'short mut self) -> FuncInstBuilder<'short, 'a>
Returns an object with the InstBuilder
trait that allows to conveniently append an instruction to the current Block
being built.
Examples found in repository?
729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063
pub fn call_memcpy(
&mut self,
config: TargetFrontendConfig,
dest: Value,
src: Value,
size: Value,
) {
let pointer_type = config.pointer_type();
let signature = {
let mut s = Signature::new(config.default_call_conv);
s.params.push(AbiParam::new(pointer_type));
s.params.push(AbiParam::new(pointer_type));
s.params.push(AbiParam::new(pointer_type));
self.import_signature(s)
};
let libc_memcpy = self.import_function(ExtFuncData {
name: ExternalName::LibCall(LibCall::Memcpy),
signature,
colocated: false,
});
self.ins().call(libc_memcpy, &[dest, src, size]);
}
/// Optimised memcpy or memmove for small copies.
///
/// # Codegen safety
///
/// The following properties must hold to prevent UB:
///
/// * `src_align` and `dest_align` are an upper-bound on the alignment of `src` respectively `dest`.
/// * If `non_overlapping` is true, then this must be correct.
pub fn emit_small_memory_copy(
&mut self,
config: TargetFrontendConfig,
dest: Value,
src: Value,
size: u64,
dest_align: u8,
src_align: u8,
non_overlapping: bool,
mut flags: MemFlags,
) {
// Currently the result of guess work, not actual profiling.
const THRESHOLD: u64 = 4;
if size == 0 {
return;
}
let access_size = greatest_divisible_power_of_two(size);
assert!(
access_size.is_power_of_two(),
"`size` is not a power of two"
);
assert!(
access_size >= u64::from(::core::cmp::min(src_align, dest_align)),
"`size` is smaller than `dest` and `src`'s alignment value."
);
let (access_size, int_type) = if access_size <= 8 {
(access_size, Type::int((access_size * 8) as u16).unwrap())
} else {
(8, types::I64)
};
let load_and_store_amount = size / access_size;
if load_and_store_amount > THRESHOLD {
let size_value = self.ins().iconst(config.pointer_type(), size as i64);
if non_overlapping {
self.call_memcpy(config, dest, src, size_value);
} else {
self.call_memmove(config, dest, src, size_value);
}
return;
}
if u64::from(src_align) >= access_size && u64::from(dest_align) >= access_size {
flags.set_aligned();
}
// Load all of the memory first. This is necessary in case `dest` overlaps.
// It can also improve performance a bit.
let registers: smallvec::SmallVec<[_; THRESHOLD as usize]> = (0..load_and_store_amount)
.map(|i| {
let offset = (access_size * i) as i32;
(self.ins().load(int_type, flags, src, offset), offset)
})
.collect();
for (value, offset) in registers {
self.ins().store(flags, value, dest, offset);
}
}
/// Calls libc.memset
///
/// Writes `size` bytes of i8 value `ch` to memory starting at `buffer`.
pub fn call_memset(
&mut self,
config: TargetFrontendConfig,
buffer: Value,
ch: Value,
size: Value,
) {
let pointer_type = config.pointer_type();
let signature = {
let mut s = Signature::new(config.default_call_conv);
s.params.push(AbiParam::new(pointer_type));
s.params.push(AbiParam::new(types::I32));
s.params.push(AbiParam::new(pointer_type));
self.import_signature(s)
};
let libc_memset = self.import_function(ExtFuncData {
name: ExternalName::LibCall(LibCall::Memset),
signature,
colocated: false,
});
let ch = self.ins().uextend(types::I32, ch);
self.ins().call(libc_memset, &[buffer, ch, size]);
}
/// Calls libc.memset
///
/// Writes `size` bytes of value `ch` to memory starting at `buffer`.
pub fn emit_small_memset(
&mut self,
config: TargetFrontendConfig,
buffer: Value,
ch: u8,
size: u64,
buffer_align: u8,
mut flags: MemFlags,
) {
// Currently the result of guess work, not actual profiling.
const THRESHOLD: u64 = 4;
if size == 0 {
return;
}
let access_size = greatest_divisible_power_of_two(size);
assert!(
access_size.is_power_of_two(),
"`size` is not a power of two"
);
assert!(
access_size >= u64::from(buffer_align),
"`size` is smaller than `dest` and `src`'s alignment value."
);
let (access_size, int_type) = if access_size <= 8 {
(access_size, Type::int((access_size * 8) as u16).unwrap())
} else {
(8, types::I64)
};
let load_and_store_amount = size / access_size;
if load_and_store_amount > THRESHOLD {
let ch = self.ins().iconst(types::I8, i64::from(ch));
let size = self.ins().iconst(config.pointer_type(), size as i64);
self.call_memset(config, buffer, ch, size);
} else {
if u64::from(buffer_align) >= access_size {
flags.set_aligned();
}
let ch = u64::from(ch);
let raw_value = if int_type == types::I64 {
ch * 0x0101010101010101_u64
} else if int_type == types::I32 {
ch * 0x01010101_u64
} else if int_type == types::I16 {
(ch << 8) | ch
} else {
assert_eq!(int_type, types::I8);
ch
};
let value = self.ins().iconst(int_type, raw_value as i64);
for i in 0..load_and_store_amount {
let offset = (access_size * i) as i32;
self.ins().store(flags, value, buffer, offset);
}
}
}
/// Calls libc.memmove
///
/// Copies `size` bytes from memory starting at `source` to memory starting
/// at `dest`. `source` is always read before writing to `dest`.
pub fn call_memmove(
&mut self,
config: TargetFrontendConfig,
dest: Value,
source: Value,
size: Value,
) {
let pointer_type = config.pointer_type();
let signature = {
let mut s = Signature::new(config.default_call_conv);
s.params.push(AbiParam::new(pointer_type));
s.params.push(AbiParam::new(pointer_type));
s.params.push(AbiParam::new(pointer_type));
self.import_signature(s)
};
let libc_memmove = self.import_function(ExtFuncData {
name: ExternalName::LibCall(LibCall::Memmove),
signature,
colocated: false,
});
self.ins().call(libc_memmove, &[dest, source, size]);
}
/// Calls libc.memcmp
///
/// Compares `size` bytes from memory starting at `left` to memory starting
/// at `right`. Returns `0` if all `n` bytes are equal. If the first difference
/// is at offset `i`, returns a positive integer if `ugt(left[i], right[i])`
/// and a negative integer if `ult(left[i], right[i])`.
///
/// Returns a C `int`, which is currently always [`types::I32`].
pub fn call_memcmp(
&mut self,
config: TargetFrontendConfig,
left: Value,
right: Value,
size: Value,
) -> Value {
let pointer_type = config.pointer_type();
let signature = {
let mut s = Signature::new(config.default_call_conv);
s.params.reserve(3);
s.params.push(AbiParam::new(pointer_type));
s.params.push(AbiParam::new(pointer_type));
s.params.push(AbiParam::new(pointer_type));
s.returns.push(AbiParam::new(types::I32));
self.import_signature(s)
};
let libc_memcmp = self.import_function(ExtFuncData {
name: ExternalName::LibCall(LibCall::Memcmp),
signature,
colocated: false,
});
let call = self.ins().call(libc_memcmp, &[left, right, size]);
self.func.dfg.first_result(call)
}
/// Optimised [`Self::call_memcmp`] for small copies.
///
/// This implements the byte slice comparison `int_cc(left[..size], right[..size])`.
///
/// `left_align` and `right_align` are the statically-known alignments of the
/// `left` and `right` pointers respectively. These are used to know whether
/// to mark `load`s as aligned. It's always fine to pass `1` for these, but
/// passing something higher than the true alignment may trap or otherwise
/// misbehave as described in [`MemFlags::aligned`].
///
/// Note that `memcmp` is a *big-endian* and *unsigned* comparison.
/// As such, this panics when called with `IntCC::Signed*`.
pub fn emit_small_memory_compare(
&mut self,
config: TargetFrontendConfig,
int_cc: IntCC,
left: Value,
right: Value,
size: u64,
left_align: std::num::NonZeroU8,
right_align: std::num::NonZeroU8,
flags: MemFlags,
) -> Value {
use IntCC::*;
let (zero_cc, empty_imm) = match int_cc {
//
Equal => (Equal, 1),
NotEqual => (NotEqual, 0),
UnsignedLessThan => (SignedLessThan, 0),
UnsignedGreaterThanOrEqual => (SignedGreaterThanOrEqual, 1),
UnsignedGreaterThan => (SignedGreaterThan, 0),
UnsignedLessThanOrEqual => (SignedLessThanOrEqual, 1),
SignedLessThan
| SignedGreaterThanOrEqual
| SignedGreaterThan
| SignedLessThanOrEqual => {
panic!("Signed comparison {} not supported by memcmp", int_cc)
}
};
if size == 0 {
return self.ins().iconst(types::I8, empty_imm);
}
// Future work could consider expanding this to handle more-complex scenarios.
if let Some(small_type) = size.try_into().ok().and_then(Type::int_with_byte_size) {
if let Equal | NotEqual = zero_cc {
let mut left_flags = flags;
if size == left_align.get() as u64 {
left_flags.set_aligned();
}
let mut right_flags = flags;
if size == right_align.get() as u64 {
right_flags.set_aligned();
}
let left_val = self.ins().load(small_type, left_flags, left, 0);
let right_val = self.ins().load(small_type, right_flags, right, 0);
return self.ins().icmp(int_cc, left_val, right_val);
} else if small_type == types::I8 {
// Once the big-endian loads from wasmtime#2492 are implemented in
// the backends, we could easily handle comparisons for more sizes here.
// But for now, just handle single bytes where we don't need to worry.
let mut aligned_flags = flags;
aligned_flags.set_aligned();
let left_val = self.ins().load(small_type, aligned_flags, left, 0);
let right_val = self.ins().load(small_type, aligned_flags, right, 0);
return self.ins().icmp(int_cc, left_val, right_val);
}
}
let pointer_type = config.pointer_type();
let size = self.ins().iconst(pointer_type, size as i64);
let cmp = self.call_memcmp(config, left, right, size);
self.ins().icmp_imm(zero_cc, cmp, 0)
}
More examples
111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326
fn build_search_tree(
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();
// 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;
}
let mut stack: Vec<(Option<Block>, Vec<ContiguousCaseRange>)> = Vec::new();
stack.push((None, contiguous_case_ranges));
while let Some((block, contiguous_case_ranges)) = stack.pop() {
if let Some(block) = block {
bx.switch_to_block(block);
}
if contiguous_case_ranges.len() <= 3 {
Self::build_search_branches(
bx,
val,
otherwise,
contiguous_case_ranges,
&mut cases_and_jt_blocks,
);
} else {
let split_point = contiguous_case_ranges.len() / 2;
let mut left = contiguous_case_ranges;
let right = left.split_off(split_point);
let left_block = bx.create_block();
let right_block = bx.create_block();
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.seal_block(left_block);
bx.seal_block(right_block);
stack.push((Some(left_block), left));
stack.push((Some(right_block), right));
}
}
cases_and_jt_blocks
}
/// Linear search for the right `ContiguousCaseRange`.
fn build_search_branches(
bx: &mut FunctionBuilder,
val: Value,
otherwise: Block,
contiguous_case_ranges: Vec<ContiguousCaseRange>,
cases_and_jt_blocks: &mut Vec<(EntryIndex, Block, Vec<Block>)>,
) {
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;
}
(_, _) => {
ins_fallthrough_jump(was_branch, bx);
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.seal_block(jt_block);
cases_and_jt_blocks.push((first_index, jt_block, blocks));
}
}
was_branch = true;
}
bx.ins().jump(otherwise, &[]);
}
/// For every item in `cases_and_jt_blocks` this will create a jump table in the specified block.
fn build_jump_tables(
bx: &mut FunctionBuilder,
val: Value,
otherwise: Block,
cases_and_jt_blocks: Vec<(EntryIndex, Block, Vec<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"
);
let mut jt_data = JumpTableData::new();
for block in blocks {
jt_data.push_entry(block);
}
let jump_table = bx.create_jump_table(jt_data);
bx.switch_to_block(jt_block);
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 {
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,
};
bx.ins().br_table(discr, otherwise, jump_table);
}
}
/// Build the switch
///
/// # Arguments
///
/// * The function builder to emit to
/// * The value to switch on
/// * The default block
pub fn emit(self, bx: &mut FunctionBuilder, val: Value, otherwise: Block) {
// Validate that the type of `val` is sufficiently wide to address all cases.
let max = self.cases.keys().max().copied().unwrap_or(0);
let val_ty = bx.func.dfg.value_type(val);
let val_ty_max = val_ty.bounds(false).1;
if max > val_ty_max {
panic!(
"The index type {} does not fit the maximum switch entry of {}",
val_ty, max
);
}
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);
}
}
fn icmp_imm_u128(bx: &mut FunctionBuilder, cond: IntCC, x: Value, y: u128) -> Value {
if let Ok(index) = u64::try_from(y) {
bx.ins().icmp_imm(cond, x, index as i64)
} else {
let (lsb, msb) = (y as u64, (y >> 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().icmp(cond, x, index)
}
}
sourcepub fn ensure_inserted_block(&mut self)
pub fn ensure_inserted_block(&mut self)
Make sure that the current block is inserted in the layout.
Examples found in repository?
98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554
fn build(self, data: InstructionData, ctrl_typevar: Type) -> (Inst, &'short mut DataFlowGraph) {
// We only insert the Block in the layout when an instruction is added to it
self.builder.ensure_inserted_block();
let inst = self.builder.func.dfg.make_inst(data.clone());
self.builder.func.dfg.make_inst_results(inst, ctrl_typevar);
self.builder.func.layout.append_inst(inst, self.block);
if !self.builder.srcloc.is_default() {
self.builder.func.set_srcloc(inst, self.builder.srcloc);
}
if data.opcode().is_branch() {
match data.branch_destination() {
Some(dest_block) => {
// If the user has supplied jump arguments we must adapt the arguments of
// the destination block
self.builder.declare_successor(dest_block, inst);
}
None => {
// branch_destination() doesn't detect jump_tables
// If jump table we declare all entries successor
if let InstructionData::BranchTable {
table, destination, ..
} = data
{
// Unlike all other jumps/branches, jump tables are
// capable of having the same successor appear
// multiple times, so we must deduplicate.
let mut unique = EntitySet::<Block>::new();
for dest_block in self
.builder
.func
.jump_tables
.get(table)
.expect("you are referencing an undeclared jump table")
.iter()
.filter(|&dest_block| unique.insert(*dest_block))
{
// Call `declare_block_predecessor` instead of `declare_successor` for
// avoiding the borrow checker.
self.builder
.func_ctx
.ssa
.declare_block_predecessor(*dest_block, inst);
}
self.builder.declare_successor(destination, inst);
}
}
}
}
if data.opcode().is_terminator() {
self.builder.fill_current_block()
}
(inst, &mut self.builder.func.dfg)
}
}
#[derive(Debug, Copy, Clone, PartialEq, Eq)]
/// An error encountered when calling [`FunctionBuilder::try_use_var`].
pub enum UseVariableError {
UsedBeforeDeclared(Variable),
}
impl fmt::Display for UseVariableError {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
match self {
UseVariableError::UsedBeforeDeclared(variable) => {
write!(
f,
"variable {} was used before it was defined",
variable.index()
)?;
}
}
Ok(())
}
}
impl std::error::Error for UseVariableError {}
#[derive(Debug, Copy, Clone, Eq, PartialEq)]
/// An error encountered when calling [`FunctionBuilder::try_declare_var`].
pub enum DeclareVariableError {
DeclaredMultipleTimes(Variable),
}
impl std::error::Error for DeclareVariableError {}
impl fmt::Display for DeclareVariableError {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
match self {
DeclareVariableError::DeclaredMultipleTimes(variable) => {
write!(
f,
"variable {} was declared multiple times",
variable.index()
)?;
}
}
Ok(())
}
}
#[derive(Debug, Copy, Clone, Eq, PartialEq)]
/// An error encountered when defining the initial value of a variable.
pub enum DefVariableError {
/// The variable was instantiated with a value of the wrong type.
///
/// note: to obtain the type of the value, you can call
/// [`cranelift_codegen::ir::dfg::DataFlowGraph::value_type`] (using the
/// [`FunctionBuilder.func.dfg`] field)
TypeMismatch(Variable, Value),
/// The value was defined (in a call to [`FunctionBuilder::def_var`]) before
/// it was declared (in a call to [`FunctionBuilder::declare_var`]).
DefinedBeforeDeclared(Variable),
}
impl fmt::Display for DefVariableError {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
match self {
DefVariableError::TypeMismatch(variable, value) => {
write!(
f,
"the types of variable {} and value {} are not the same.
The `Value` supplied to `def_var` must be of the same type as
the variable was declared to be of in `declare_var`.",
variable.index(),
value.as_u32()
)?;
}
DefVariableError::DefinedBeforeDeclared(variable) => {
write!(
f,
"the value of variabe {} was declared before it was defined",
variable.index()
)?;
}
}
Ok(())
}
}
/// This module allows you to create a function in Cranelift IR in a straightforward way, hiding
/// all the complexity of its internal representation.
///
/// The module is parametrized by one type which is the representation of variables in your
/// origin language. It offers a way to conveniently append instruction to your program flow.
/// You are responsible to split your instruction flow into extended blocks (declared with
/// `create_block`) whose properties are:
///
/// - branch and jump instructions can only point at the top of extended blocks;
/// - the last instruction of each block is a terminator instruction which has no natural successor,
/// and those instructions can only appear at the end of extended blocks.
///
/// The parameters of Cranelift IR instructions are Cranelift IR values, which can only be created
/// as results of other Cranelift IR instructions. To be able to create variables redefined multiple
/// times in your program, use the `def_var` and `use_var` command, that will maintain the
/// correspondence between your variables and Cranelift IR SSA values.
///
/// The first block for which you call `switch_to_block` will be assumed to be the beginning of
/// the function.
///
/// At creation, a `FunctionBuilder` instance borrows an already allocated `Function` which it
/// modifies with the information stored in the mutable borrowed
/// [`FunctionBuilderContext`](struct.FunctionBuilderContext.html). The function passed in
/// argument should be newly created with
/// [`Function::with_name_signature()`](Function::with_name_signature), whereas the
/// `FunctionBuilderContext` can be kept as is between two function translations.
///
/// # Errors
///
/// The functions below will panic in debug mode whenever you try to modify the Cranelift IR
/// function in a way that violate the coherence of the code. For instance: switching to a new
/// `Block` when you haven't filled the current one with a terminator instruction, inserting a
/// return instruction with arguments that don't match the function's signature.
impl<'a> FunctionBuilder<'a> {
/// Creates a new FunctionBuilder structure that will operate on a `Function` using a
/// `FunctionBuilderContext`.
pub fn new(func: &'a mut Function, func_ctx: &'a mut FunctionBuilderContext) -> Self {
debug_assert!(func_ctx.is_empty());
Self {
func,
srcloc: Default::default(),
func_ctx,
position: Default::default(),
}
}
/// Get the block that this builder is currently at.
pub fn current_block(&self) -> Option<Block> {
self.position.expand()
}
/// Set the source location that should be assigned to all new instructions.
pub fn set_srcloc(&mut self, srcloc: ir::SourceLoc) {
self.srcloc = srcloc;
}
/// Creates a new `Block` and returns its reference.
pub fn create_block(&mut self) -> Block {
let block = self.func.dfg.make_block();
self.func_ctx.ssa.declare_block(block);
block
}
/// Mark a block as "cold".
///
/// This will try to move it out of the ordinary path of execution
/// when lowered to machine code.
pub fn set_cold_block(&mut self, block: Block) {
self.func.layout.set_cold(block);
}
/// Insert `block` in the layout *after* the existing block `after`.
pub fn insert_block_after(&mut self, block: Block, after: Block) {
self.func.layout.insert_block_after(block, after);
}
/// After the call to this function, new instructions will be inserted into the designated
/// block, in the order they are declared. You must declare the types of the Block arguments
/// you will use here.
///
/// When inserting the terminator instruction (which doesn't have a fallthrough to its immediate
/// successor), the block will be declared filled and it will not be possible to append
/// instructions to it.
pub fn switch_to_block(&mut self, block: Block) {
// First we check that the previous block has been filled.
debug_assert!(
self.position.is_none()
|| self.is_unreachable()
|| self.is_pristine(self.position.unwrap())
|| self.is_filled(self.position.unwrap()),
"you have to fill your block before switching"
);
// We cannot switch to a filled block
debug_assert!(
!self.is_filled(block),
"you cannot switch to a block which is already filled"
);
// Then we change the cursor position.
self.position = PackedOption::from(block);
}
/// Declares that all the predecessors of this block are known.
///
/// Function to call with `block` as soon as the last branch instruction to `block` has been
/// created. Forgetting to call this method on every block will cause inconsistencies in the
/// produced functions.
pub fn seal_block(&mut self, block: Block) {
let side_effects = self.func_ctx.ssa.seal_block(block, self.func);
self.handle_ssa_side_effects(side_effects);
}
/// Effectively calls seal_block on all unsealed blocks in the function.
///
/// It's more efficient to seal `Block`s as soon as possible, during
/// translation, but for frontends where this is impractical to do, this
/// function can be used at the end of translating all blocks to ensure
/// that everything is sealed.
pub fn seal_all_blocks(&mut self) {
let side_effects = self.func_ctx.ssa.seal_all_blocks(self.func);
self.handle_ssa_side_effects(side_effects);
}
/// Declares the type of a variable, so that it can be used later (by calling
/// [`FunctionBuilder::use_var`]). This function will return an error if it
/// was not possible to use the variable.
pub fn try_declare_var(&mut self, var: Variable, ty: Type) -> Result<(), DeclareVariableError> {
if self.func_ctx.types[var] != types::INVALID {
return Err(DeclareVariableError::DeclaredMultipleTimes(var));
}
self.func_ctx.types[var] = ty;
Ok(())
}
/// In order to use a variable (by calling [`FunctionBuilder::use_var`]), you need
/// to first declare its type with this method.
pub fn declare_var(&mut self, var: Variable, ty: Type) {
self.try_declare_var(var, ty)
.unwrap_or_else(|_| panic!("the variable {:?} has been declared multiple times", var))
}
/// Returns the Cranelift IR necessary to use a previously defined user
/// variable, returning an error if this is not possible.
pub fn try_use_var(&mut self, var: Variable) -> Result<Value, UseVariableError> {
// Assert that we're about to add instructions to this block using the definition of the
// given variable. ssa.use_var is the only part of this crate which can add block parameters
// behind the caller's back. If we disallow calling append_block_param as soon as use_var is
// called, then we enforce a strict separation between user parameters and SSA parameters.
self.ensure_inserted_block();
let (val, side_effects) = {
let ty = *self
.func_ctx
.types
.get(var)
.ok_or(UseVariableError::UsedBeforeDeclared(var))?;
debug_assert_ne!(
ty,
types::INVALID,
"variable {:?} is used but its type has not been declared",
var
);
self.func_ctx
.ssa
.use_var(self.func, var, ty, self.position.unwrap())
};
self.handle_ssa_side_effects(side_effects);
Ok(val)
}
/// Returns the Cranelift IR value corresponding to the utilization at the current program
/// position of a previously defined user variable.
pub fn use_var(&mut self, var: Variable) -> Value {
self.try_use_var(var).unwrap_or_else(|_| {
panic!(
"variable {:?} is used but its type has not been declared",
var
)
})
}
/// Registers a new definition of a user variable. This function will return
/// an error if the value supplied does not match the type the variable was
/// declared to have.
pub fn try_def_var(&mut self, var: Variable, val: Value) -> Result<(), DefVariableError> {
let var_ty = *self
.func_ctx
.types
.get(var)
.ok_or(DefVariableError::DefinedBeforeDeclared(var))?;
if var_ty != self.func.dfg.value_type(val) {
return Err(DefVariableError::TypeMismatch(var, val));
}
self.func_ctx.ssa.def_var(var, val, self.position.unwrap());
Ok(())
}
/// Register a new definition of a user variable. The type of the value must be
/// the same as the type registered for the variable.
pub fn def_var(&mut self, var: Variable, val: Value) {
self.try_def_var(var, val)
.unwrap_or_else(|error| match error {
DefVariableError::TypeMismatch(var, val) => {
panic!(
"declared type of variable {:?} doesn't match type of value {}",
var, val
);
}
DefVariableError::DefinedBeforeDeclared(var) => {
panic!(
"variable {:?} is used but its type has not been declared",
var
);
}
})
}
/// Set label for Value
///
/// This will not do anything unless `func.dfg.collect_debug_info` is called first.
pub fn set_val_label(&mut self, val: Value, label: ValueLabel) {
if let Some(values_labels) = self.func.stencil.dfg.values_labels.as_mut() {
use alloc::collections::btree_map::Entry;
let start = ValueLabelStart {
from: RelSourceLoc::from_base_offset(self.func.params.base_srcloc(), self.srcloc),
label,
};
match values_labels.entry(val) {
Entry::Occupied(mut e) => match e.get_mut() {
ValueLabelAssignments::Starts(starts) => starts.push(start),
_ => panic!("Unexpected ValueLabelAssignments at this stage"),
},
Entry::Vacant(e) => {
e.insert(ValueLabelAssignments::Starts(vec![start]));
}
}
}
}
/// Creates a jump table in the function, to be used by `br_table` instructions.
pub fn create_jump_table(&mut self, data: JumpTableData) -> JumpTable {
self.func.create_jump_table(data)
}
/// Creates a sized stack slot in the function, to be used by `stack_load`, `stack_store` and
/// `stack_addr` instructions.
pub fn create_sized_stack_slot(&mut self, data: StackSlotData) -> StackSlot {
self.func.create_sized_stack_slot(data)
}
/// Creates a dynamic stack slot in the function, to be used by `dynamic_stack_load`,
/// `dynamic_stack_store` and `dynamic_stack_addr` instructions.
pub fn create_dynamic_stack_slot(&mut self, data: DynamicStackSlotData) -> DynamicStackSlot {
self.func.create_dynamic_stack_slot(data)
}
/// Adds a signature which can later be used to declare an external function import.
pub fn import_signature(&mut self, signature: Signature) -> SigRef {
self.func.import_signature(signature)
}
/// Declare an external function import.
pub fn import_function(&mut self, data: ExtFuncData) -> FuncRef {
self.func.import_function(data)
}
/// Declares a global value accessible to the function.
pub fn create_global_value(&mut self, data: GlobalValueData) -> GlobalValue {
self.func.create_global_value(data)
}
/// Declares a heap accessible to the function.
pub fn create_heap(&mut self, data: HeapData) -> Heap {
self.func.create_heap(data)
}
/// Returns an object with the [`InstBuilder`](cranelift_codegen::ir::InstBuilder)
/// trait that allows to conveniently append an instruction to the current `Block` being built.
pub fn ins<'short>(&'short mut self) -> FuncInstBuilder<'short, 'a> {
let block = self
.position
.expect("Please call switch_to_block before inserting instructions");
FuncInstBuilder::new(self, block)
}
/// Make sure that the current block is inserted in the layout.
pub fn ensure_inserted_block(&mut self) {
let block = self.position.unwrap();
if self.is_pristine(block) {
if !self.func.layout.is_block_inserted(block) {
self.func.layout.append_block(block);
}
self.func_ctx.status[block] = BlockStatus::Partial;
} else {
debug_assert!(
!self.is_filled(block),
"you cannot add an instruction to a block already filled"
);
}
}
/// Returns a `FuncCursor` pointed at the current position ready for inserting instructions.
///
/// This can be used to insert SSA code that doesn't need to access locals and that doesn't
/// need to know about `FunctionBuilder` at all.
pub fn cursor(&mut self) -> FuncCursor {
self.ensure_inserted_block();
FuncCursor::new(self.func)
.with_srcloc(self.srcloc)
.at_bottom(self.position.unwrap())
}
sourcepub fn cursor(&mut self) -> FuncCursor<'_>
pub fn cursor(&mut self) -> FuncCursor<'_>
Returns a FuncCursor
pointed at the current position ready for inserting instructions.
This can be used to insert SSA code that doesn’t need to access locals and that doesn’t
need to know about FunctionBuilder
at all.
sourcepub fn append_block_params_for_function_params(&mut self, block: Block)
pub fn append_block_params_for_function_params(&mut self, block: Block)
Append parameters to the given Block
corresponding to the function
parameters. This can be used to set up the block parameters for the
entry block.
sourcepub fn append_block_params_for_function_returns(&mut self, block: Block)
pub fn append_block_params_for_function_returns(&mut self, block: Block)
Append parameters to the given Block
corresponding to the function
return values. This can be used to set up the block parameters for a
function exit block.
source§impl<'a> FunctionBuilder<'a>
impl<'a> FunctionBuilder<'a>
All the functions documented in the previous block are write-only and help you build a valid Cranelift IR functions via multiple debug asserts. However, you might need to improve the performance of your translation perform more complex transformations to your Cranelift IR function. The functions below help you inspect the function you’re creating and modify it in ways that can be unsafe if used incorrectly.
sourcepub fn block_params(&self, block: Block) -> &[Value]
pub fn block_params(&self, block: Block) -> &[Value]
Retrieves all the parameters for a Block
currently inferred from the jump instructions
inserted that target it and the SSA construction.
sourcepub fn signature(&self, sigref: SigRef) -> Option<&Signature>
pub fn signature(&self, sigref: SigRef) -> Option<&Signature>
Retrieves the signature with reference sigref
previously added with import_signature
.
sourcepub fn append_block_param(&mut self, block: Block, ty: Type) -> Value
pub fn append_block_param(&mut self, block: Block, ty: Type) -> Value
Creates a parameter for a specific Block
by appending it to the list of already existing
parameters.
Note: this function has to be called at the creation of the Block
before adding
instructions to it, otherwise this could interfere with SSA construction.
sourcepub fn inst_results(&self, inst: Inst) -> &[Value]
pub fn inst_results(&self, inst: Inst) -> &[Value]
Returns the result values of an instruction.
sourcepub fn change_jump_destination(&mut self, inst: Inst, new_dest: Block)
pub fn change_jump_destination(&mut self, inst: Inst, new_dest: Block)
Changes the destination of a jump instruction after creation.
Note: You are responsible for maintaining the coherence with the arguments of other jump instructions.
sourcepub fn is_unreachable(&self) -> bool
pub fn is_unreachable(&self) -> bool
Returns true
if and only if the current Block
is sealed and has no predecessors declared.
The entry block of a function is never unreachable.
Examples found in repository?
324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341
pub fn switch_to_block(&mut self, block: Block) {
// First we check that the previous block has been filled.
debug_assert!(
self.position.is_none()
|| self.is_unreachable()
|| self.is_pristine(self.position.unwrap())
|| self.is_filled(self.position.unwrap()),
"you have to fill your block before switching"
);
// We cannot switch to a filled block
debug_assert!(
!self.is_filled(block),
"you cannot switch to a block which is already filled"
);
// Then we change the cursor position.
self.position = PackedOption::from(block);
}
source§impl<'a> FunctionBuilder<'a>
impl<'a> FunctionBuilder<'a>
Helper functions
sourcepub fn call_memcpy(
&mut self,
config: TargetFrontendConfig,
dest: Value,
src: Value,
size: Value
)
pub fn call_memcpy(
&mut self,
config: TargetFrontendConfig,
dest: Value,
src: Value,
size: Value
)
Calls libc.memcpy
Copies the size
bytes from src
to dest
, assumes that src + size
won’t overlap onto dest
. If dest
and src
overlap, the behavior is
undefined. Applications in which dest
and src
might overlap should
use call_memmove
instead.
Examples found in repository?
762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824
pub fn emit_small_memory_copy(
&mut self,
config: TargetFrontendConfig,
dest: Value,
src: Value,
size: u64,
dest_align: u8,
src_align: u8,
non_overlapping: bool,
mut flags: MemFlags,
) {
// Currently the result of guess work, not actual profiling.
const THRESHOLD: u64 = 4;
if size == 0 {
return;
}
let access_size = greatest_divisible_power_of_two(size);
assert!(
access_size.is_power_of_two(),
"`size` is not a power of two"
);
assert!(
access_size >= u64::from(::core::cmp::min(src_align, dest_align)),
"`size` is smaller than `dest` and `src`'s alignment value."
);
let (access_size, int_type) = if access_size <= 8 {
(access_size, Type::int((access_size * 8) as u16).unwrap())
} else {
(8, types::I64)
};
let load_and_store_amount = size / access_size;
if load_and_store_amount > THRESHOLD {
let size_value = self.ins().iconst(config.pointer_type(), size as i64);
if non_overlapping {
self.call_memcpy(config, dest, src, size_value);
} else {
self.call_memmove(config, dest, src, size_value);
}
return;
}
if u64::from(src_align) >= access_size && u64::from(dest_align) >= access_size {
flags.set_aligned();
}
// Load all of the memory first. This is necessary in case `dest` overlaps.
// It can also improve performance a bit.
let registers: smallvec::SmallVec<[_; THRESHOLD as usize]> = (0..load_and_store_amount)
.map(|i| {
let offset = (access_size * i) as i32;
(self.ins().load(int_type, flags, src, offset), offset)
})
.collect();
for (value, offset) in registers {
self.ins().store(flags, value, dest, offset);
}
}
sourcepub fn emit_small_memory_copy(
&mut self,
config: TargetFrontendConfig,
dest: Value,
src: Value,
size: u64,
dest_align: u8,
src_align: u8,
non_overlapping: bool,
flags: MemFlags
)
pub fn emit_small_memory_copy(
&mut self,
config: TargetFrontendConfig,
dest: Value,
src: Value,
size: u64,
dest_align: u8,
src_align: u8,
non_overlapping: bool,
flags: MemFlags
)
Optimised memcpy or memmove for small copies.
Codegen safety
The following properties must hold to prevent UB:
src_align
anddest_align
are an upper-bound on the alignment ofsrc
respectivelydest
.- If
non_overlapping
is true, then this must be correct.
sourcepub fn call_memset(
&mut self,
config: TargetFrontendConfig,
buffer: Value,
ch: Value,
size: Value
)
pub fn call_memset(
&mut self,
config: TargetFrontendConfig,
buffer: Value,
ch: Value,
size: Value
)
Calls libc.memset
Writes size
bytes of i8 value ch
to memory starting at buffer
.
Examples found in repository?
858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919
pub fn emit_small_memset(
&mut self,
config: TargetFrontendConfig,
buffer: Value,
ch: u8,
size: u64,
buffer_align: u8,
mut flags: MemFlags,
) {
// Currently the result of guess work, not actual profiling.
const THRESHOLD: u64 = 4;
if size == 0 {
return;
}
let access_size = greatest_divisible_power_of_two(size);
assert!(
access_size.is_power_of_two(),
"`size` is not a power of two"
);
assert!(
access_size >= u64::from(buffer_align),
"`size` is smaller than `dest` and `src`'s alignment value."
);
let (access_size, int_type) = if access_size <= 8 {
(access_size, Type::int((access_size * 8) as u16).unwrap())
} else {
(8, types::I64)
};
let load_and_store_amount = size / access_size;
if load_and_store_amount > THRESHOLD {
let ch = self.ins().iconst(types::I8, i64::from(ch));
let size = self.ins().iconst(config.pointer_type(), size as i64);
self.call_memset(config, buffer, ch, size);
} else {
if u64::from(buffer_align) >= access_size {
flags.set_aligned();
}
let ch = u64::from(ch);
let raw_value = if int_type == types::I64 {
ch * 0x0101010101010101_u64
} else if int_type == types::I32 {
ch * 0x01010101_u64
} else if int_type == types::I16 {
(ch << 8) | ch
} else {
assert_eq!(int_type, types::I8);
ch
};
let value = self.ins().iconst(int_type, raw_value as i64);
for i in 0..load_and_store_amount {
let offset = (access_size * i) as i32;
self.ins().store(flags, value, buffer, offset);
}
}
}
sourcepub fn emit_small_memset(
&mut self,
config: TargetFrontendConfig,
buffer: Value,
ch: u8,
size: u64,
buffer_align: u8,
flags: MemFlags
)
pub fn emit_small_memset(
&mut self,
config: TargetFrontendConfig,
buffer: Value,
ch: u8,
size: u64,
buffer_align: u8,
flags: MemFlags
)
Calls libc.memset
Writes size
bytes of value ch
to memory starting at buffer
.
sourcepub fn call_memmove(
&mut self,
config: TargetFrontendConfig,
dest: Value,
source: Value,
size: Value
)
pub fn call_memmove(
&mut self,
config: TargetFrontendConfig,
dest: Value,
source: Value,
size: Value
)
Calls libc.memmove
Copies size
bytes from memory starting at source
to memory starting
at dest
. source
is always read before writing to dest
.
Examples found in repository?
762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824
pub fn emit_small_memory_copy(
&mut self,
config: TargetFrontendConfig,
dest: Value,
src: Value,
size: u64,
dest_align: u8,
src_align: u8,
non_overlapping: bool,
mut flags: MemFlags,
) {
// Currently the result of guess work, not actual profiling.
const THRESHOLD: u64 = 4;
if size == 0 {
return;
}
let access_size = greatest_divisible_power_of_two(size);
assert!(
access_size.is_power_of_two(),
"`size` is not a power of two"
);
assert!(
access_size >= u64::from(::core::cmp::min(src_align, dest_align)),
"`size` is smaller than `dest` and `src`'s alignment value."
);
let (access_size, int_type) = if access_size <= 8 {
(access_size, Type::int((access_size * 8) as u16).unwrap())
} else {
(8, types::I64)
};
let load_and_store_amount = size / access_size;
if load_and_store_amount > THRESHOLD {
let size_value = self.ins().iconst(config.pointer_type(), size as i64);
if non_overlapping {
self.call_memcpy(config, dest, src, size_value);
} else {
self.call_memmove(config, dest, src, size_value);
}
return;
}
if u64::from(src_align) >= access_size && u64::from(dest_align) >= access_size {
flags.set_aligned();
}
// Load all of the memory first. This is necessary in case `dest` overlaps.
// It can also improve performance a bit.
let registers: smallvec::SmallVec<[_; THRESHOLD as usize]> = (0..load_and_store_amount)
.map(|i| {
let offset = (access_size * i) as i32;
(self.ins().load(int_type, flags, src, offset), offset)
})
.collect();
for (value, offset) in registers {
self.ins().store(flags, value, dest, offset);
}
}
sourcepub fn call_memcmp(
&mut self,
config: TargetFrontendConfig,
left: Value,
right: Value,
size: Value
) -> Value
pub fn call_memcmp(
&mut self,
config: TargetFrontendConfig,
left: Value,
right: Value,
size: Value
) -> Value
Calls libc.memcmp
Compares size
bytes from memory starting at left
to memory starting
at right
. Returns 0
if all n
bytes are equal. If the first difference
is at offset i
, returns a positive integer if ugt(left[i], right[i])
and a negative integer if ult(left[i], right[i])
.
Returns a C int
, which is currently always types::I32
.
Examples found in repository?
998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063
pub fn emit_small_memory_compare(
&mut self,
config: TargetFrontendConfig,
int_cc: IntCC,
left: Value,
right: Value,
size: u64,
left_align: std::num::NonZeroU8,
right_align: std::num::NonZeroU8,
flags: MemFlags,
) -> Value {
use IntCC::*;
let (zero_cc, empty_imm) = match int_cc {
//
Equal => (Equal, 1),
NotEqual => (NotEqual, 0),
UnsignedLessThan => (SignedLessThan, 0),
UnsignedGreaterThanOrEqual => (SignedGreaterThanOrEqual, 1),
UnsignedGreaterThan => (SignedGreaterThan, 0),
UnsignedLessThanOrEqual => (SignedLessThanOrEqual, 1),
SignedLessThan
| SignedGreaterThanOrEqual
| SignedGreaterThan
| SignedLessThanOrEqual => {
panic!("Signed comparison {} not supported by memcmp", int_cc)
}
};
if size == 0 {
return self.ins().iconst(types::I8, empty_imm);
}
// Future work could consider expanding this to handle more-complex scenarios.
if let Some(small_type) = size.try_into().ok().and_then(Type::int_with_byte_size) {
if let Equal | NotEqual = zero_cc {
let mut left_flags = flags;
if size == left_align.get() as u64 {
left_flags.set_aligned();
}
let mut right_flags = flags;
if size == right_align.get() as u64 {
right_flags.set_aligned();
}
let left_val = self.ins().load(small_type, left_flags, left, 0);
let right_val = self.ins().load(small_type, right_flags, right, 0);
return self.ins().icmp(int_cc, left_val, right_val);
} else if small_type == types::I8 {
// Once the big-endian loads from wasmtime#2492 are implemented in
// the backends, we could easily handle comparisons for more sizes here.
// But for now, just handle single bytes where we don't need to worry.
let mut aligned_flags = flags;
aligned_flags.set_aligned();
let left_val = self.ins().load(small_type, aligned_flags, left, 0);
let right_val = self.ins().load(small_type, aligned_flags, right, 0);
return self.ins().icmp(int_cc, left_val, right_val);
}
}
let pointer_type = config.pointer_type();
let size = self.ins().iconst(pointer_type, size as i64);
let cmp = self.call_memcmp(config, left, right, size);
self.ins().icmp_imm(zero_cc, cmp, 0)
}
sourcepub fn emit_small_memory_compare(
&mut self,
config: TargetFrontendConfig,
int_cc: IntCC,
left: Value,
right: Value,
size: u64,
left_align: NonZeroU8,
right_align: NonZeroU8,
flags: MemFlags
) -> Value
pub fn emit_small_memory_compare(
&mut self,
config: TargetFrontendConfig,
int_cc: IntCC,
left: Value,
right: Value,
size: u64,
left_align: NonZeroU8,
right_align: NonZeroU8,
flags: MemFlags
) -> Value
Optimised Self::call_memcmp
for small copies.
This implements the byte slice comparison int_cc(left[..size], right[..size])
.
left_align
and right_align
are the statically-known alignments of the
left
and right
pointers respectively. These are used to know whether
to mark load
s as aligned. It’s always fine to pass 1
for these, but
passing something higher than the true alignment may trap or otherwise
misbehave as described in MemFlags::aligned
.
Note that memcmp
is a big-endian and unsigned comparison.
As such, this panics when called with IntCC::Signed*
.