diff --git a/internal/cgen/builtin.go b/internal/cgen/builtin.go index d31af4cd..9965cff4 100644 --- a/internal/cgen/builtin.go +++ b/internal/cgen/builtin.go @@ -593,7 +593,11 @@ func (g *gen) writeBuiltinSlice(b *buffer, recv *a.Expr, method t.ID, args []*a. } if (t.IDPeekU8 <= method) && (method <= t.IDPeekU64LE) { - b.printf("wuffs_base__%s__no_bounds_check(", method.Str(g.tm)) + s := method.Str(g.tm) + if i := strings.Index(s, "_as_"); i >= 0 { + s = s[:i] + } + b.printf("wuffs_base__%s__no_bounds_check(", s) if err := g.writeExpr(b, recv, depth); err != nil { return err } diff --git a/lang/builtin/builtin.go b/lang/builtin/builtin.go index e117e68b..b7d713a3 100644 --- a/lang/builtin/builtin.go +++ b/lang/builtin/builtin.go @@ -612,16 +612,16 @@ var SliceU8Funcs = []string{ "GENERIC T1.peek_u8() u8", "GENERIC T1.peek_u16be() u16", "GENERIC T1.peek_u16le() u16", - "GENERIC T1.peek_u24be() u32", - "GENERIC T1.peek_u24le() u32", + "GENERIC T1.peek_u24be_as_u32() u32", + "GENERIC T1.peek_u24le_as_u32() u32", "GENERIC T1.peek_u32be() u32", "GENERIC T1.peek_u32le() u32", - "GENERIC T1.peek_u40be() u64", - "GENERIC T1.peek_u40le() u64", - "GENERIC T1.peek_u48be() u64", - "GENERIC T1.peek_u48le() u64", - "GENERIC T1.peek_u56be() u64", - "GENERIC T1.peek_u56le() u64", + "GENERIC T1.peek_u40be_as_u64() u64", + "GENERIC T1.peek_u40le_as_u64() u64", + "GENERIC T1.peek_u48be_as_u64() u64", + "GENERIC T1.peek_u48le_as_u64() u64", + "GENERIC T1.peek_u56be_as_u64() u64", + "GENERIC T1.peek_u56le_as_u64() u64", "GENERIC T1.peek_u64be() u64", "GENERIC T1.peek_u64le() u64", diff --git a/std/png/decode_filter_sse128.wuffs b/std/png/decode_filter_sse128.wuffs index 746afcdf..7858541d 100644 --- a/std/png/decode_filter_sse128.wuffs +++ b/std/png/decode_filter_sse128.wuffs @@ -16,6 +16,28 @@ // Filter 1: Sub. +// This (filter = 1, distance = 3) implementation doesn't actually bench faster +// than the non-SIMD one. +// +// pri func decoder.filter_1_distance_3_sse128!(curr: slice base.u8), +// choose cpu_arch >= sse128, +// { +// var c : slice base.u8 +// var x128 : base.sse128_i +// var a128 : base.sse128_i +// +// iterate (c = args.curr)(length: 4, advance: 3, unroll: 1) { +// x128.load_u32!(a: c.peek_u32le()) +// x128 = x128._mm_add_epi8!(b: a128) +// a128 = x128 +// c.poke_u24le!(a: x128.truncate_u32()) +// } else (length: 3, advance: 3, unroll: 1) { +// x128.load_u32!(a: c.peek_u24le_as_u32()) +// x128 = x128._mm_add_epi8!(b: a128) +// c.poke_u24le!(a: x128.truncate_u32()) +// } +// } + pri func decoder.filter_1_distance_4_sse128!(curr: slice base.u8), choose cpu_arch >= sse128, {