Rename sse128/sse128i to x86_sse42/x86_m128i

This commit is contained in:
Nigel Tao committed 2021-01-24 21:58:00 +11:00
1 parent a563bb7cf8
commit 5941e7e5cc
9 files changed
+85 -83

No files matched your search

+37 -37
View File
@@ -20,30 +20,30 @@
// than the non-SIMD one.
//
// pri func decoder.filter_1_distance_3_sse128!(curr: slice base.u8),
// choose cpu_arch >= sse128,
// choose cpu_arch >= x86_sse42,
// {
// var c : slice base.u8
// var x128 : base.sse128_i
// var a128 : base.sse128_i
// var x128 : base.x86_m128i
// var a128 : base.x86_m128i
//
// iterate (c = args.curr)(length: 4, advance: 3, unroll: 2) {
// x128.load_u32!(a: c.peek_u32le())
// x128 = x128._mm_add_epi8!(b: a128)
// 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)
// 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,
choose cpu_arch >= x86_sse42,
{
var c : slice base.u8
var x128 : base.sse128_i
var a128 : base.sse128_i
var x128 : base.x86_m128i
var a128 : base.x86_m128i
iterate (c = args.curr)(length: 4, advance: 4, unroll: 2) {
x128.load_u32!(a: c.peek_u32le())
@@ -61,21 +61,21 @@ pri func decoder.filter_1_distance_4_sse128!(curr: slice base.u8),
// 3, distance = 3) doesn't actually bench faster than the non-SIMD one.
//
// pri func decoder.filter_3_distance_3_sse128!(curr: slice base.u8, prev: slice base.u8),
// choose cpu_arch >= sse128,
// choose cpu_arch >= x86_sse42,
// {
// etc
// }
pri func decoder.filter_3_distance_4_sse128!(curr: slice base.u8, prev: slice base.u8),
choose cpu_arch >= sse128,
choose cpu_arch >= x86_sse42,
{
var c : slice base.u8
var p : slice base.u8
var x128 : base.sse128_i
var a128 : base.sse128_i
var b128 : base.sse128_i
var p128 : base.sse128_i
var k128 : base.sse128_i
var x128 : base.x86_m128i
var a128 : base.x86_m128i
var b128 : base.x86_m128i
var p128 : base.x86_m128i
var k128 : base.x86_m128i
if args.prev.length() == 0 {
k128 = k128.create_mm_set1_epi8(a: 0xFE)
@@ -124,7 +124,7 @@ pri func decoder.filter_3_distance_4_sse128!(curr: slice base.u8, prev: slice ba
// Filter 4: Paeth.
pri func decoder.filter_4_distance_3_sse128!(curr: slice base.u8, prev: slice base.u8),
choose cpu_arch >= sse128,
choose cpu_arch >= x86_sse42,
{
// See the comments in filter_4_distance_4_sse128 for an explanation of how
// this works. That function's single loop is copied twice here, once with
@@ -135,16 +135,16 @@ pri func decoder.filter_4_distance_3_sse128!(curr: slice base.u8, prev: slice ba
var c : slice base.u8
var p : slice base.u8
var x128 : base.sse128_i
var a128 : base.sse128_i
var b128 : base.sse128_i
var c128 : base.sse128_i
var p128 : base.sse128_i
var pa128 : base.sse128_i
var pb128 : base.sse128_i
var pc128 : base.sse128_i
var smallest128 : base.sse128_i
var z128 : base.sse128_i
var x128 : base.x86_m128i
var a128 : base.x86_m128i
var b128 : base.x86_m128i
var c128 : base.x86_m128i
var p128 : base.x86_m128i
var pa128 : base.x86_m128i
var pb128 : base.x86_m128i
var pc128 : base.x86_m128i
var smallest128 : base.x86_m128i
var z128 : base.x86_m128i
// § The advance is 3, not 4.
iterate (c = args.curr, p = args.prev)(length: 4, advance: 3, unroll: 2) {
@@ -202,20 +202,20 @@ pri func decoder.filter_4_distance_3_sse128!(curr: slice base.u8, prev: slice ba
}
pri func decoder.filter_4_distance_4_sse128!(curr: slice base.u8, prev: slice base.u8),
choose cpu_arch >= sse128,
choose cpu_arch >= x86_sse42,
{
var c : slice base.u8
var p : slice base.u8
var x128 : base.sse128_i
var a128 : base.sse128_i
var b128 : base.sse128_i
var c128 : base.sse128_i
var p128 : base.sse128_i
var pa128 : base.sse128_i
var pb128 : base.sse128_i
var pc128 : base.sse128_i
var smallest128 : base.sse128_i
var z128 : base.sse128_i
var x128 : base.x86_m128i
var a128 : base.x86_m128i
var b128 : base.x86_m128i
var c128 : base.x86_m128i
var p128 : base.x86_m128i
var pa128 : base.x86_m128i
var pb128 : base.x86_m128i
var pc128 : base.x86_m128i
var smallest128 : base.x86_m128i
var z128 : base.x86_m128i
iterate (c = args.curr, p = args.prev)(length: 4, advance: 4, unroll: 2) {
// Load the pixel from the row above.