fluencelabs/wasmer
1
0
Fork 0
mirror of https://github.com/fluencelabs/wasmer synced 2025-06-12 08:31:21 +00:00
Code Issues Projects Releases Wiki Activity

Improve performance of memory access.

Browse Source
This commit is contained in:
Heyang Zhou
2019-05-13 05:11:08 -07:00
parent 620a6ddd85
commit af1ac9af96
3 changed files with 100 additions and 44 deletions
Download Patch File Download Diff File Expand all files Collapse all files

59
lib/singlepass-backend/src/codegen_x64.rs
View File

@ -1201,41 +1201,6 @@ impl X64FunctionCode {
value_size: usize,
cb: F,
) {
let tmp_addr = m.acquire_temp_gpr().unwrap();
let tmp_base = m.acquire_temp_gpr().unwrap();
let tmp_bound = m.acquire_temp_gpr().unwrap();
// Loads both base and bound into temporary registers.
a.emit_mov(
Size::S64,
Location::Memory(
Machine::get_vmctx_reg(),
match MemoryIndex::new(0).local_or_import(module_info) {
LocalOrImport::Local(_) => vm::Ctx::offset_memories(),
LocalOrImport::Import(_) => vm::Ctx::offset_imported_memories(),
} as i32,
),
Location::GPR(tmp_base),
);
a.emit_mov(
Size::S64,
Location::Memory(tmp_base, 0),
Location::GPR(tmp_base),
);
a.emit_mov(
Size::S32,
Location::Memory(tmp_base, LocalMemory::offset_bound() as i32),
Location::GPR(tmp_bound),
);
a.emit_mov(
Size::S64,
Location::Memory(tmp_base, LocalMemory::offset_base() as i32),
Location::GPR(tmp_base),
);
// Adds base to bound so `tmp_bound` now holds the end of linear memory.
a.emit_add(Size::S64, Location::GPR(tmp_base), Location::GPR(tmp_bound));
// If the memory is dynamic, we need to do bound checking at runtime.
let mem_desc = match MemoryIndex::new(0).local_or_import(module_info) {
LocalOrImport::Local(local_mem_index) => &module_info.memories[local_mem_index],
@ -1251,8 +1216,32 @@ impl X64FunctionCode {
MemoryBoundCheckMode::Enable => true,
MemoryBoundCheckMode::Disable => false,
};
let tmp_addr = m.acquire_temp_gpr().unwrap();
let tmp_base = m.acquire_temp_gpr().unwrap();
let tmp_bound = m.acquire_temp_gpr().unwrap();
// Load base into temporary register.
a.emit_mov(
Size::S64,
Location::Memory(
Machine::get_vmctx_reg(),
vm::Ctx::offset_memory_base() as i32,
),
Location::GPR(tmp_base),
);
if need_check {
a.emit_mov(
Size::S64,
Location::Memory(
Machine::get_vmctx_reg(),
vm::Ctx::offset_memory_bound() as i32,
),
Location::GPR(tmp_bound),
);
// Adds base to bound so `tmp_bound` now holds the end of linear memory.
a.emit_add(Size::S64, Location::GPR(tmp_base), Location::GPR(tmp_bound));
a.emit_mov(Size::S32, addr, Location::GPR(tmp_addr));
// This branch is used for emitting "faster" code for the special case of (offset + value_size) not exceeding u32 range.