go/sqlite3
1
0
Fork 0
mirror of https://github.com/ncruces/go-sqlite3.git synced 2026-01-12 05:59:14 +00:00
Code Issues Packages Projects Releases Wiki Activity

SIMD.

Browse Source
This commit is contained in:
Nuno Cruces
2025-04-04 17:46:21 +01:00
parent e6c9f18934
commit 4c19387535
5 changed files with 309 additions and 98 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
sqlite3/libc/build.sh
View File

@@ -19,7 +19,8 @@ trap 'rm -f libc.tmp' EXIT
-Wl,--initial-memory=16777216 \
-Wl,--export=memset \
-Wl,--export=memcpy \
-Wl,--export=memcmp
-Wl,--export=memcmp \
-Wl,--export=strncmp
"$BINARYEN/wasm-ctor-eval" -g -c _initialize libc.wasm -o libc.tmp
"$BINARYEN/wasm-opt" -g --strip --strip-producers -c -O3 \

BIN
sqlite3/libc/libc.wasm
View File

Binary file not shown.

263
sqlite3/libc/libc.wat
View File

@@ -6,6 +6,7 @@
(export "memset" (func $memset))
(export "memcpy" (func $memcpy))
(export "memcmp" (func $memcmp))
(export "strncmp" (func $strncmp))
(func $memset (param $0 i32) (param $1 i32) (param $2 i32) (result i32)
(memory.fill
(local.get $0)
@@ -25,61 +26,54 @@
(func $memcmp (param $0 i32) (param $1 i32) (param $2 i32) (result i32)
(local $3 i32)
(local $4 i32)
(block $block2
(block $block1
(block $block
(br_if $block
(i32.lt_u
(local.get $2)
(i32.const 8)
)
(block $block1
(block $block
(if
(i32.ge_u
(local.get $2)
(i32.const 16)
)
(br_if $block
(i32.and
(i32.or
(local.get $0)
(local.get $1)
)
(i32.const 7)
)
)
(loop $label
(br_if $block1
(i64.ne
(i64.load
(local.get $0)
)
(i64.load
(local.get $1)
)
)
)
(local.set $1
(i32.add
(local.get $1)
(i32.const 8)
)
)
(local.set $0
(i32.add
(local.get $0)
(i32.const 8)
)
)
(br_if $label
(i32.gt_u
(local.tee $2
(i32.sub
(local.get $2)
(i32.const 8)
(then
(loop $label
(br_if $block
(v128.any_true
(v128.xor
(v128.load align=1
(local.get $1)
)
(v128.load align=1
(local.get $0)
)
)
)
(i32.const 7)
)
(local.set $1
(i32.add
(local.get $1)
(i32.const 16)
)
)
(local.set $0
(i32.add
(local.get $0)
(i32.const 16)
)
)
(br_if $label
(i32.gt_u
(local.tee $2
(i32.sub
(local.get $2)
(i32.const 16)
)
)
(i32.const 15)
)
)
)
)
)
(br_if $block2
(br_if $block1
(i32.eqz
(local.get $2)
)
@@ -132,6 +126,179 @@
)
(i32.const 0)
)
(func $strncmp (param $0 i32) (param $1 i32) (param $2 i32) (result i32)
(local $3 i32)
(local $4 i32)
(local $5 i32)
(local $6 v128)
(block $block2
(block $block1
(block $block
(if
(i32.ge_u
(local.get $2)
(i32.const 16)
)
(then
(loop $label
(br_if $block
(v128.any_true
(v128.xor
(v128.load align=1
(local.get $1)
)
(local.tee $6
(v128.load align=1
(local.get $0)
)
)
)
)
)
(if
(i32.eqz
(i8x16.all_true
(local.get $6)
)
)
(then
(return
(i32.const 0)
)
)
)
(local.set $1
(i32.add
(local.get $1)
(i32.const 16)
)
)
(local.set $0
(i32.add
(local.get $0)
(i32.const 16)
)
)
(br_if $label
(i32.gt_u
(local.tee $2
(i32.sub
(local.get $2)
(i32.const 16)
)
)
(i32.const 15)
)
)
)
)
)
(br_if $block
(local.get $2)
)
(local.set $4
(i32.load8_u
(local.get $1)
)
)
(local.set $3
(i32.load8_u
(local.get $0)
)
)
(br $block1)
)
(br_if $block1
(i32.ne
(local.tee $3
(i32.load8_u
(local.get $0)
)
)
(local.tee $4
(i32.load8_u
(local.get $1)
)
)
)
)
(local.set $5
(i32.add
(local.get $1)
(i32.const 1)
)
)
(local.set $0
(i32.add
(local.get $0)
(i32.const 1)
)
)
(local.set $1
(i32.sub
(local.get $2)
(i32.const 1)
)
)
(loop $label1
(local.set $2
(i32.const 0)
)
(br_if $block2
(i32.eqz
(local.get $1)
)
)
(br_if $block2
(i32.eqz
(local.get $3)
)
)
(local.set $1
(i32.sub
(local.get $1)
(i32.const 1)
)
)
(local.set $4
(i32.load8_u
(local.get $5)
)
)
(local.set $3
(i32.load8_u
(local.get $0)
)
)
(local.set $5
(i32.add
(local.get $5)
(i32.const 1)
)
)
(local.set $0
(i32.add
(local.get $0)
(i32.const 1)
)
)
(br_if $label1
(i32.eq
(local.get $3)
(local.get $4)
)
)
)
)
(local.set $2
(i32.sub
(local.get $3)
(local.get $4)
)
)
)
(local.get $2)
)
;; features section: mutable-globals, nontrapping-float-to-int, simd, bulk-memory, sign-ext, reference-types, multivalue, bulk-memory-opt
)

63
sqlite3/libc/libc_test.go
View File

@@ -20,15 +20,17 @@ const (
)
var (
memory []byte
module api.Module
memset api.Function
memcpy api.Function
memcmp api.Function
memory []byte
module api.Module
memset api.Function
memcpy api.Function
memcmp api.Function
strncmp api.Function
)
func call(fn api.Function, arg ...uint64) {
func call(fn api.Function, arg ...uint64) uint32 {
fn.CallWithStack(context.Background(), arg)
return uint32(arg[0])
}
func TestMain(m *testing.M) {
@@ -44,6 +46,7 @@ func TestMain(m *testing.M) {
memset = mod.ExportedFunction("memset")
memcpy = mod.ExportedFunction("memcpy")
memcmp = mod.ExportedFunction("memcmp")
strncmp = mod.ExportedFunction("strncmp")
memory, _ = mod.Memory().Read(0, mod.Memory().Size())
os.Exit(m.Run())
@@ -58,9 +61,9 @@ func Benchmark_memset(b *testing.B) {
}
b.StopTimer()
for i, v := range memory[ptr1 : ptr1+size] {
if v != 3 {
b.Fatal(i, v)
for i, got := range memory[ptr1 : ptr1+size] {
if got != 3 {
b.Fatal(i, got)
}
}
}
@@ -75,9 +78,9 @@ func Benchmark_memcpy(b *testing.B) {
}
b.StopTimer()
for i, v := range memory[ptr1 : ptr1+size] {
if v != 5 {
b.Fatal(i, v)
for i, got := range memory[ptr1 : ptr1+size] {
if got != 5 {
b.Fatal(i, got)
}
}
}
@@ -93,4 +96,40 @@ func Benchmark_memcmp(b *testing.B) {
call(memcmp, ptr1, ptr2, size)
}
b.StopTimer()
// ptr1 > ptr2
if got := int32(call(memcmp, ptr1, ptr2, size)); got <= 0 {
b.Fatal(got)
}
// ptr1[:size/2] == ptr2[:size/2]
if got := int32(call(memcmp, ptr1, ptr2, size/2)); got != 0 {
b.Fatal(got)
}
}
func Benchmark_strncmp(b *testing.B) {
clear(memory)
call(memset, ptr1, 7, size)
call(memset, ptr2, 7, size)
call(memset, ptr2+size/2, 5, size)
b.ResetTimer()
for range b.N {
call(strncmp, ptr1, ptr2, size)
}
b.StopTimer()
// ptr1 > ptr2
if got := int32(call(memcmp, ptr1, ptr2, size)); got <= 0 {
b.Fatal(got)
}
// make ptr1 < ptr2
memory[ptr1+size/2] = 0
if got := int32(call(memcmp, ptr1, ptr2, size)); got >= 0 {
b.Fatal(got)
}
// ptr1[:size/2] == ptr2[:size/2]
if got := int32(call(memcmp, ptr1, ptr2, size/2)); got != 0 {
b.Fatal(got)
}
}

78
sqlite3/strings.c
View File

@@ -1,8 +1,9 @@
#include <limits.h>
#include <stddef.h>
#include <stdint.h>
#include <wasm_simd128.h>
#if defined(__wasm_bulk_memory__)
#ifdef __wasm_bulk_memory__
void *memset(void *dest, int c, size_t n) {
return __builtin_memset(dest, c, n);
@@ -18,22 +19,17 @@ void *memmove(void *dest, const void *src, size_t n) {
#endif
#define ONES (~(uintmax_t)(0) / UCHAR_MAX)
#define HIGHS (ONES * (UCHAR_MAX / 2 + 1))
#define HASZERO(x) ((x) - (typeof(x))(ONES) & ~(x) & (typeof(x))(HIGHS))
#define UNALIGNED(x) ((uintptr_t)(x) & (sizeof(*x) - 1))
#ifdef __wasm_simd128__
int memcmp(const void *v1, const void *v2, size_t n) {
typedef uint64_t __attribute__((__may_alias__)) word;
const word *w1 = v1;
const word *w2 = v2;
if (!(UNALIGNED(w1) | UNALIGNED(w2))) {
while (n >= sizeof(word) && *w1 == *w2) {
n -= sizeof(word);
w1++;
w2++;
const v128_t *w1 = v1;
const v128_t *w2 = v2;
for (; n >= sizeof(v128_t); n -= sizeof(v128_t)) {
if (wasm_v128_any_true(wasm_v128_load(w1) ^ wasm_v128_load(w2))) {
break; // *w1 != *w2
}
w1++;
w2++;
}
const unsigned char *u1 = (const void *)w1;
@@ -46,6 +42,37 @@ int memcmp(const void *v1, const void *v2, size_t n) {
return 0;
}
int strncmp(const char *c1, const char *c2, size_t n) {
const v128_t *w1 = (const void *)c1;
const v128_t *w2 = (const void *)c2;
for (; n >= sizeof(v128_t); n -= sizeof(v128_t)) {
if (wasm_v128_any_true(wasm_v128_load(w1) ^ wasm_v128_load(w2))) {
break; // *w1 != *w2
}
if (!wasm_i8x16_all_true(wasm_v128_load(w1))) {
return 0; // *w1 == *w2 and they have a NUL
}
w1++;
w2++;
}
c1 = (const void *)w1;
c2 = (const void *)w2;
while (n-- && *c1 == *c2) {
if (n == 0 || *c1 == 0) return 0;
c1++;
c2++;
}
return *(unsigned char *)c1 - *(unsigned char *)c2;
}
#endif
#define ONES (~(uintmax_t)(0) / UCHAR_MAX)
#define HIGHS (ONES * (UCHAR_MAX / 2 + 1))
#define HASZERO(x) ((x) - (typeof(x))(ONES) & ~(x) & (typeof(x))(HIGHS))
#define UNALIGNED(x) ((uintptr_t)(x) & (sizeof(*x) - 1))
int strcmp(const char *c1, const char *c2) {
typedef uintptr_t __attribute__((__may_alias__)) word;
@@ -68,29 +95,6 @@ int strcmp(const char *c1, const char *c2) {
return *(unsigned char *)c1 - *(unsigned char *)c2;
}
int strncmp(const char *c1, const char *c2, size_t n) {
typedef uintptr_t __attribute__((__may_alias__)) word;
const word *w1 = (const void *)c1;
const word *w2 = (const void *)c2;
if (!(UNALIGNED(w1) | UNALIGNED(w2))) {
while (n >= sizeof(word) && *w1 == *w2) {
if ((n -= sizeof(word)) == 0 || HASZERO(*w1)) return 0;
w1++;
w2++;
}
c1 = (const void *)w1;
c2 = (const void *)w2;
}
while (n-- && *c1 == *c2) {
if (n == 0 || *c1 == 0) return 0;
c1++;
c2++;
}
return *(unsigned char *)c1 - *(unsigned char *)c2;
}
#undef UNALIGNED
#undef HASZERO
#undef HIGHS