Change some more s/sse128/sse42/

This commit is contained in:
Nigel Tao committed 2021-01-24 22:04:53 +11:00
1 parent 5941e7e5cc
commit 73d95387f5
7 files changed
+125 -125

No files matched your search

+8 -8
View File
@@ -19,7 +19,7 @@
// 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),
// pri func decoder.filter_1_distance_3_sse42!(curr: slice base.u8),
// choose cpu_arch >= x86_sse42,
// {
// var c : slice base.u8
@@ -38,7 +38,7 @@
// }
// }
pri func decoder.filter_1_distance_4_sse128!(curr: slice base.u8),
pri func decoder.filter_1_distance_4_sse42!(curr: slice base.u8),
choose cpu_arch >= x86_sse42,
{
var c : slice base.u8
@@ -57,16 +57,16 @@ pri func decoder.filter_1_distance_4_sse128!(curr: slice base.u8),
// Filter 3: Average.
// Similar to filter_1_distance_3_sse128, the SIMD implementation for (filter =
// Similar to filter_1_distance_3_sse42, the SIMD implementation for (filter =
// 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),
// pri func decoder.filter_3_distance_3_sse42!(curr: slice base.u8, prev: slice base.u8),
// choose cpu_arch >= x86_sse42,
// {
// etc
// }
pri func decoder.filter_3_distance_4_sse128!(curr: slice base.u8, prev: slice base.u8),
pri func decoder.filter_3_distance_4_sse42!(curr: slice base.u8, prev: slice base.u8),
choose cpu_arch >= x86_sse42,
{
var c : slice base.u8
@@ -123,10 +123,10 @@ 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),
pri func decoder.filter_4_distance_3_sse42!(curr: slice base.u8, prev: slice base.u8),
choose cpu_arch >= x86_sse42,
{
// See the comments in filter_4_distance_4_sse128 for an explanation of how
// See the comments in filter_4_distance_4_sse42 for an explanation of how
// this works. That function's single loop is copied twice here, once with
// "length: 4" and once with "length: 3". It's generally faster to load 4
// bytes at a time instead of 3.
@@ -201,7 +201,7 @@ 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),
pri func decoder.filter_4_distance_4_sse42!(curr: slice base.u8, prev: slice base.u8),
choose cpu_arch >= x86_sse42,
{
var c : slice base.u8
+4 -4
View File
@@ -313,17 +313,17 @@ pri func decoder.choose_filter_implementations!() {
choose filter_1 = [filter_1_distance_3_fallback]
choose filter_3 = [filter_3_distance_3_fallback]
choose filter_4 = [
filter_4_distance_3_sse128,
filter_4_distance_3_sse42,
filter_4_distance_3_fallback]
} else if this.filter_distance == 4 {
choose filter_1 = [
filter_1_distance_4_sse128,
filter_1_distance_4_sse42,
filter_1_distance_4_fallback]
choose filter_3 = [
filter_3_distance_4_sse128,
filter_3_distance_4_sse42,
filter_3_distance_4_fallback]
choose filter_4 = [
filter_4_distance_4_sse128,
filter_4_distance_4_sse42,
filter_4_distance_4_fallback]
}
}