mirrors/kitty
1
0
Fork 0
mirror of https://github.com/kovidgoyal/kitty synced 2026-07-07 00:15:12 +02:00
Code Issues Packages Projects Releases Wiki Activity

Make a re-useable template for generating ASM index functions with different tests

Browse Source
This commit is contained in:
Kovid Goyal
2024-01-31 19:59:09 +05:30
parent fa9a2b1e2e
commit 82b7b4fcce
1 changed files with 39 additions and 32 deletions
Download Patch File Download Diff File Expand all files Collapse all files

71
tools/simdstring/generate.go
View File

@@ -1079,46 +1079,20 @@ func (isa *ISA) LEA() string {
return "LEAQ"
}
func (s *State) indexbyte2() {
f := s.NewFunction("index_byte2_asm", "Find the index of either of two bytes", []FunctionParam{{"data", ByteSlice}, {"b1", types.Byte}, {"b2", types.Byte}}, []FunctionParam{{"ans", types.Int}})
if s.ISA.HasSIMD {
s.indexbyte2_body(f)
}
f = s.NewFunction("index_byte2_string_asm", "Find the index of either of two bytes", []FunctionParam{{"data", types.String}, {"b1", types.Byte}, {"b2", types.Byte}}, []FunctionParam{{"ans", types.Int}})
if s.ISA.HasSIMD {
s.indexbyte2_body(f)
}
}
func (s *State) indexbyte2_body(f *Function) {
b1 := f.Vec()
b2 := f.Vec()
test_ans := f.Vec()
f.Set1Epi8FromParam("b1", b1)
f.Set1Epi8FromParam("b2", b2)
func (s *State) index_func(f *Function, test_bytes_impl func(pos, test_ans Register, aligned bool, byte_found_label string)) {
pos := f.Reg()
test_ans := f.Vec()
data_start := f.LoadParam(`data`)
limit := f.LoadParamLen(`data`)
f.AddToSelf(limit, data_start)
test_bytes := func(aligned bool) {
bytes_to_test := f.Vec()
defer f.ReleaseReg(bytes_to_test)
if aligned {
f.LoadPointerAligned(pos, bytes_to_test)
} else {
f.LoadPointerUnaligned(pos, bytes_to_test)
}
f.CmpEqEpi8(bytes_to_test, b1, test_ans)
f.CmpEqEpi8(bytes_to_test, b2, bytes_to_test)
f.Or(test_ans, bytes_to_test, test_ans)
f.JumpIfNonZero(test_ans, "byte_found")
}
vecsz := f.ISA.Bits / 8
f.CopyRegister(data_start, pos)
test_bytes := func(aligned bool) {
test_bytes_impl(pos, test_ans, aligned, "byte_found")
}
f.Comment("Load next vector of possibly unaligned bytes and check if either of the bytes are present in it")
test_bytes(false)
@@ -1164,6 +1138,39 @@ func (s *State) indexbyte2_body(f *Function) {
}
func (s *State) indexbyte2_body(f *Function) {
b1 := f.Vec()
b2 := f.Vec()
f.Set1Epi8FromParam("b1", b1)
f.Set1Epi8FromParam("b2", b2)
test_bytes := func(pos, test_ans Register, aligned bool, byte_found_label string) {
bytes_to_test := f.Vec()
defer f.ReleaseReg(bytes_to_test)
if aligned {
f.LoadPointerAligned(pos, bytes_to_test)
} else {
f.LoadPointerUnaligned(pos, bytes_to_test)
}
f.CmpEqEpi8(bytes_to_test, b1, test_ans)
f.CmpEqEpi8(bytes_to_test, b2, bytes_to_test)
f.Or(test_ans, bytes_to_test, test_ans)
f.JumpIfNonZero(test_ans, byte_found_label)
}
s.index_func(f, test_bytes)
}
func (s *State) indexbyte2() {
f := s.NewFunction("index_byte2_asm", "Find the index of either of two bytes", []FunctionParam{{"data", ByteSlice}, {"b1", types.Byte}, {"b2", types.Byte}}, []FunctionParam{{"ans", types.Int}})
if s.ISA.HasSIMD {
s.indexbyte2_body(f)
}
f = s.NewFunction("index_byte2_string_asm", "Find the index of either of two bytes", []FunctionParam{{"data", types.String}, {"b1", types.Byte}, {"b2", types.Byte}}, []FunctionParam{{"ans", types.Int}})
if s.ISA.HasSIMD {
s.indexbyte2_body(f)
}
}
func (s *State) Generate() {
s.test_load()
s.test_set1_epi8()