Rename some std/png c/p variables

This commit is contained in:
Nigel Tao committed 2021-02-18 14:32:35 +11:00
1 parent fd071dfc90
commit 869be76fae
4 files changed
+649 -634

No files matched your search

+32 -28
View File
@@ -31,14 +31,15 @@
pri func decoder.filter_1_distance_4_arm_neon!(curr: slice base.u8),
choose cpu_arch >= arm_neon,
{
var c : slice base.u8
var curr : slice base.u8
var fa : base.arm_neon_64
var fx : base.arm_neon_64
iterate (c = args.curr)(length: 4, advance: 4, unroll: 2) {
fx = fx.create_vdup_n_u32(a: c.peek_u32le())
iterate (curr = args.curr)(length: 4, advance: 4, unroll: 2) {
fx = fx.create_vdup_n_u32(a: curr.peek_u32le())
fx = fx.vadd_u8(b: fa)
c.poke_u32le!(a: fx.vget_lane_u32(b: 0))
curr.poke_u32le!(a: fx.vget_lane_u32(b: 0))
fa = fx
}
}
@@ -50,26 +51,27 @@ pri func decoder.filter_1_distance_4_arm_neon!(curr: slice base.u8),
pri func decoder.filter_3_distance_4_arm_neon!(curr: slice base.u8, prev: slice base.u8),
choose cpu_arch >= arm_neon,
{
var c : slice base.u8
var p : slice base.u8
var curr : slice base.u8
var prev : slice base.u8
var fa : base.arm_neon_64
var fb : base.arm_neon_64
var fx : base.arm_neon_64
if args.prev.length() == 0 {
iterate (c = args.curr)(length: 4, advance: 4, unroll: 2) {
fx = fx.create_vdup_n_u32(a: c.peek_u32le())
iterate (curr = args.curr)(length: 4, advance: 4, unroll: 2) {
fx = fx.create_vdup_n_u32(a: curr.peek_u32le())
fx = fx.vadd_u8(b: fa.vhadd_u8(b: fb))
c.poke_u32le!(a: fx.vget_lane_u32(b: 0))
curr.poke_u32le!(a: fx.vget_lane_u32(b: 0))
fa = fx
}
} else {
iterate (c = args.curr, p = args.prev)(length: 4, advance: 4, unroll: 2) {
fb = fb.create_vdup_n_u32(a: p.peek_u32le())
fx = fx.create_vdup_n_u32(a: c.peek_u32le())
iterate (curr = args.curr, prev = args.prev)(length: 4, advance: 4, unroll: 2) {
fb = fb.create_vdup_n_u32(a: prev.peek_u32le())
fx = fx.create_vdup_n_u32(a: curr.peek_u32le())
fx = fx.vadd_u8(b: fa.vhadd_u8(b: fb))
c.poke_u32le!(a: fx.vget_lane_u32(b: 0))
curr.poke_u32le!(a: fx.vget_lane_u32(b: 0))
fa = fx
}
}
@@ -89,8 +91,9 @@ pri func decoder.filter_4_distance_3_arm_neon!(curr: slice base.u8, prev: slice
//
// Differences between that function and this one are marked with a §.
var c : slice base.u8
var p : slice base.u8
var curr : slice base.u8
var prev : slice base.u8
var fa : base.arm_neon_64
var fb : base.arm_neon_64
var fc : base.arm_neon_64
@@ -106,9 +109,9 @@ pri func decoder.filter_4_distance_3_arm_neon!(curr: slice base.u8, prev: slice
var pickb : base.arm_neon_64
// § The advance is 3, not 4.
iterate (c = args.curr, p = args.prev)(length: 4, advance: 3, unroll: 2) {
fb = fb.create_vdup_n_u32(a: p.peek_u32le())
fx = fx.create_vdup_n_u32(a: c.peek_u32le())
iterate (curr = args.curr, prev = args.prev)(length: 4, advance: 3, unroll: 2) {
fb = fb.create_vdup_n_u32(a: prev.peek_u32le())
fx = fx.create_vdup_n_u32(a: curr.peek_u32le())
fafb = fa.vaddl_u8(b: fb)
fcfc = fc.vaddl_u8(b: fc)
pa = fb.vabdl_u8(b: fc)
@@ -122,14 +125,14 @@ pri func decoder.filter_4_distance_3_arm_neon!(curr: slice base.u8, prev: slice
b: picka.vbsl_u8(b: fa,
c: pickb.vbsl_u8(b: fb, c: fc)))
// § poke_u24le replaces poke_u32le.
c.poke_u24le!(a: fx.vget_lane_u32(b: 0))
curr.poke_u24le!(a: fx.vget_lane_u32(b: 0))
fc = fb
fa = fx
} else (length: 3, advance: 3, unroll: 1) {
// § peek_u24le_as_u32 replaces peek_u32le.
fb = fb.create_vdup_n_u32(a: p.peek_u24le_as_u32())
fb = fb.create_vdup_n_u32(a: prev.peek_u24le_as_u32())
// § peek_u24le_as_u32 replaces peek_u32le.
fx = fx.create_vdup_n_u32(a: c.peek_u24le_as_u32())
fx = fx.create_vdup_n_u32(a: curr.peek_u24le_as_u32())
fafb = fa.vaddl_u8(b: fb)
fcfc = fc.vaddl_u8(b: fc)
pa = fb.vabdl_u8(b: fc)
@@ -143,7 +146,7 @@ pri func decoder.filter_4_distance_3_arm_neon!(curr: slice base.u8, prev: slice
b: picka.vbsl_u8(b: fa,
c: pickb.vbsl_u8(b: fb, c: fc)))
// § poke_u24le replaces poke_u32le.
c.poke_u24le!(a: fx.vget_lane_u32(b: 0))
curr.poke_u24le!(a: fx.vget_lane_u32(b: 0))
// § These assignments are unnecessary; this is the last iteration.
// fc = fb
// fa = fx
@@ -153,8 +156,9 @@ pri func decoder.filter_4_distance_3_arm_neon!(curr: slice base.u8, prev: slice
pri func decoder.filter_4_distance_4_arm_neon!(curr: slice base.u8, prev: slice base.u8),
choose cpu_arch >= arm_neon,
{
var c : slice base.u8
var p : slice base.u8
var curr : slice base.u8
var prev : slice base.u8
var fa : base.arm_neon_64
var fb : base.arm_neon_64
var fc : base.arm_neon_64
@@ -169,9 +173,9 @@ pri func decoder.filter_4_distance_4_arm_neon!(curr: slice base.u8, prev: slice
var picka : base.arm_neon_64
var pickb : base.arm_neon_64
iterate (c = args.curr, p = args.prev)(length: 4, advance: 4, unroll: 2) {
fb = fb.create_vdup_n_u32(a: p.peek_u32le())
fx = fx.create_vdup_n_u32(a: c.peek_u32le())
iterate (curr = args.curr, prev = args.prev)(length: 4, advance: 4, unroll: 2) {
fb = fb.create_vdup_n_u32(a: prev.peek_u32le())
fx = fx.create_vdup_n_u32(a: curr.peek_u32le())
fafb = fa.vaddl_u8(b: fb) // fafb = (fa + fb)
fcfc = fc.vaddl_u8(b: fc) // fcfc = (fc + fc)
@@ -191,7 +195,7 @@ pri func decoder.filter_4_distance_4_arm_neon!(curr: slice base.u8, prev: slice
b: picka.vbsl_u8(b: fa,
c: pickb.vbsl_u8(b: fb, c: fc)))
c.poke_u32le!(a: fx.vget_lane_u32(b: 0))
curr.poke_u32le!(a: fx.vget_lane_u32(b: 0))
fc = fb
fa = fx
}
+87 -81
View File
@@ -42,37 +42,39 @@ pri func decoder.filter_1!(curr: slice base.u8),
}
pri func decoder.filter_1_distance_3_fallback!(curr: slice base.u8) {
var c : slice base.u8
var curr : slice base.u8
var fa0 : base.u8
var fa1 : base.u8
var fa2 : base.u8
iterate (c = args.curr)(length: 3, advance: 3, unroll: 2) {
fa0 = fa0 ~mod+ c[0]
c[0] = fa0
fa1 = fa1 ~mod+ c[1]
c[1] = fa1
fa2 = fa2 ~mod+ c[2]
c[2] = fa2
iterate (curr = args.curr)(length: 3, advance: 3, unroll: 2) {
fa0 = fa0 ~mod+ curr[0]
curr[0] = fa0
fa1 = fa1 ~mod+ curr[1]
curr[1] = fa1
fa2 = fa2 ~mod+ curr[2]
curr[2] = fa2
}
}
pri func decoder.filter_1_distance_4_fallback!(curr: slice base.u8) {
var c : slice base.u8
var curr : slice base.u8
var fa0 : base.u8
var fa1 : base.u8
var fa2 : base.u8
var fa3 : base.u8
iterate (c = args.curr)(length: 4, advance: 4, unroll: 1) {
fa0 = fa0 ~mod+ c[0]
c[0] = fa0
fa1 = fa1 ~mod+ c[1]
c[1] = fa1
fa2 = fa2 ~mod+ c[2]
c[2] = fa2
fa3 = fa3 ~mod+ c[3]
c[3] = fa3
iterate (curr = args.curr)(length: 4, advance: 4, unroll: 1) {
fa0 = fa0 ~mod+ curr[0]
curr[0] = fa0
fa1 = fa1 ~mod+ curr[1]
curr[1] = fa1
fa2 = fa2 ~mod+ curr[2]
curr[2] = fa2
fa3 = fa3 ~mod+ curr[3]
curr[3] = fa3
}
}
@@ -154,64 +156,66 @@ pri func decoder.filter_3!(curr: slice base.u8, prev: slice base.u8),
}
pri func decoder.filter_3_distance_3_fallback!(curr: slice base.u8, prev: slice base.u8) {
var c : slice base.u8
var p : slice base.u8
var curr : slice base.u8
var prev : slice base.u8
var fa0 : base.u8
var fa1 : base.u8
var fa2 : base.u8
if args.prev.length() == 0 {
iterate (c = args.curr)(length: 3, advance: 3, unroll: 2) {
fa0 = (fa0 / 2) ~mod+ c[0]
c[0] = fa0
fa1 = (fa1 / 2) ~mod+ c[1]
c[1] = fa1
fa2 = (fa2 / 2) ~mod+ c[2]
c[2] = fa2
iterate (curr = args.curr)(length: 3, advance: 3, unroll: 2) {
fa0 = (fa0 / 2) ~mod+ curr[0]
curr[0] = fa0
fa1 = (fa1 / 2) ~mod+ curr[1]
curr[1] = fa1
fa2 = (fa2 / 2) ~mod+ curr[2]
curr[2] = fa2
}
} else {
iterate (c = args.curr, p = args.prev)(length: 3, advance: 3, unroll: 2) {
fa0 = ((((fa0 as base.u32) + (p[0] as base.u32)) / 2) as base.u8) ~mod+ c[0]
c[0] = fa0
fa1 = ((((fa1 as base.u32) + (p[1] as base.u32)) / 2) as base.u8) ~mod+ c[1]
c[1] = fa1
fa2 = ((((fa2 as base.u32) + (p[2] as base.u32)) / 2) as base.u8) ~mod+ c[2]
c[2] = fa2
iterate (curr = args.curr, prev = args.prev)(length: 3, advance: 3, unroll: 2) {
fa0 = ((((fa0 as base.u32) + (prev[0] as base.u32)) / 2) as base.u8) ~mod+ curr[0]
curr[0] = fa0
fa1 = ((((fa1 as base.u32) + (prev[1] as base.u32)) / 2) as base.u8) ~mod+ curr[1]
curr[1] = fa1
fa2 = ((((fa2 as base.u32) + (prev[2] as base.u32)) / 2) as base.u8) ~mod+ curr[2]
curr[2] = fa2
}
}
}
pri func decoder.filter_3_distance_4_fallback!(curr: slice base.u8, prev: slice base.u8) {
var c : slice base.u8
var p : slice base.u8
var curr : slice base.u8
var prev : slice base.u8
var fa0 : base.u8
var fa1 : base.u8
var fa2 : base.u8
var fa3 : base.u8
if args.prev.length() == 0 {
iterate (c = args.curr)(length: 4, advance: 4, unroll: 1) {
fa0 = (fa0 / 2) ~mod+ c[0]
c[0] = fa0
fa1 = (fa1 / 2) ~mod+ c[1]
c[1] = fa1
fa2 = (fa2 / 2) ~mod+ c[2]
c[2] = fa2
fa3 = (fa3 / 2) ~mod+ c[3]
c[3] = fa3
iterate (curr = args.curr)(length: 4, advance: 4, unroll: 1) {
fa0 = (fa0 / 2) ~mod+ curr[0]
curr[0] = fa0
fa1 = (fa1 / 2) ~mod+ curr[1]
curr[1] = fa1
fa2 = (fa2 / 2) ~mod+ curr[2]
curr[2] = fa2
fa3 = (fa3 / 2) ~mod+ curr[3]
curr[3] = fa3
}
} else {
iterate (c = args.curr, p = args.prev)(length: 4, advance: 4, unroll: 1) {
fa0 = ((((fa0 as base.u32) + (p[0] as base.u32)) / 2) as base.u8) ~mod+ c[0]
c[0] = fa0
fa1 = ((((fa1 as base.u32) + (p[1] as base.u32)) / 2) as base.u8) ~mod+ c[1]
c[1] = fa1
fa2 = ((((fa2 as base.u32) + (p[2] as base.u32)) / 2) as base.u8) ~mod+ c[2]
c[2] = fa2
fa3 = ((((fa3 as base.u32) + (p[3] as base.u32)) / 2) as base.u8) ~mod+ c[3]
c[3] = fa3
iterate (curr = args.curr, prev = args.prev)(length: 4, advance: 4, unroll: 1) {
fa0 = ((((fa0 as base.u32) + (prev[0] as base.u32)) / 2) as base.u8) ~mod+ curr[0]
curr[0] = fa0
fa1 = ((((fa1 as base.u32) + (prev[1] as base.u32)) / 2) as base.u8) ~mod+ curr[1]
curr[1] = fa1
fa2 = ((((fa2 as base.u32) + (prev[2] as base.u32)) / 2) as base.u8) ~mod+ curr[2]
curr[2] = fa2
fa3 = ((((fa3 as base.u32) + (prev[3] as base.u32)) / 2) as base.u8) ~mod+ curr[3]
curr[3] = fa3
}
}
}
@@ -289,8 +293,9 @@ pri func decoder.filter_4!(curr: slice base.u8, prev: slice base.u8),
}
pri func decoder.filter_4_distance_3_fallback!(curr: slice base.u8, prev: slice base.u8) {
var c : slice base.u8
var p : slice base.u8
var curr : slice base.u8
var prev : slice base.u8
var fa0 : base.u32
var fa1 : base.u32
var fa2 : base.u32
@@ -313,8 +318,8 @@ pri func decoder.filter_4_distance_3_fallback!(curr: slice base.u8, prev: slice
var pc1 : base.u32
var pc2 : base.u32
iterate (c = args.curr, p = args.prev)(length: 3, advance: 3, unroll: 1) {
fb0 = p[0] as base.u32
iterate (curr = args.curr, prev = args.prev)(length: 3, advance: 3, unroll: 1) {
fb0 = prev[0] as base.u32
pp0 = (fa0 ~mod+ fb0) ~mod- fc0
pa0 = pp0 ~mod- fa0
if pa0 >= 0x8000_0000 {
@@ -335,11 +340,11 @@ pri func decoder.filter_4_distance_3_fallback!(curr: slice base.u8, prev: slice
} else {
fa0 = fc0
}
c[0] = c[0] ~mod+ ((fa0 & 0xFF) as base.u8)
fa0 = c[0] as base.u32
curr[0] = curr[0] ~mod+ ((fa0 & 0xFF) as base.u8)
fa0 = curr[0] as base.u32
fc0 = fb0
fb1 = p[1] as base.u32
fb1 = prev[1] as base.u32
pp1 = (fa1 ~mod+ fb1) ~mod- fc1
pa1 = pp1 ~mod- fa1
if pa1 >= 0x8000_0000 {
@@ -360,11 +365,11 @@ pri func decoder.filter_4_distance_3_fallback!(curr: slice base.u8, prev: slice
} else {
fa1 = fc1
}
c[1] = c[1] ~mod+ ((fa1 & 0xFF) as base.u8)
fa1 = c[1] as base.u32
curr[1] = curr[1] ~mod+ ((fa1 & 0xFF) as base.u8)
fa1 = curr[1] as base.u32
fc1 = fb1
fb2 = p[2] as base.u32
fb2 = prev[2] as base.u32
pp2 = (fa2 ~mod+ fb2) ~mod- fc2
pa2 = pp2 ~mod- fa2
if pa2 >= 0x8000_0000 {
@@ -385,15 +390,16 @@ pri func decoder.filter_4_distance_3_fallback!(curr: slice base.u8, prev: slice
} else {
fa2 = fc2
}
c[2] = c[2] ~mod+ ((fa2 & 0xFF) as base.u8)
fa2 = c[2] as base.u32
curr[2] = curr[2] ~mod+ ((fa2 & 0xFF) as base.u8)
fa2 = curr[2] as base.u32
fc2 = fb2
}
}
pri func decoder.filter_4_distance_4_fallback!(curr: slice base.u8, prev: slice base.u8) {
var c : slice base.u8
var p : slice base.u8
var curr : slice base.u8
var prev : slice base.u8
var fa0 : base.u32
var fa1 : base.u32
var fa2 : base.u32
@@ -423,8 +429,8 @@ pri func decoder.filter_4_distance_4_fallback!(curr: slice base.u8, prev: slice
var pc2 : base.u32
var pc3 : base.u32
iterate (c = args.curr, p = args.prev)(length: 4, advance: 4, unroll: 1) {
fb0 = p[0] as base.u32
iterate (curr = args.curr, prev = args.prev)(length: 4, advance: 4, unroll: 1) {
fb0 = prev[0] as base.u32
pp0 = (fa0 ~mod+ fb0) ~mod- fc0
pa0 = pp0 ~mod- fa0
if pa0 >= 0x8000_0000 {
@@ -445,11 +451,11 @@ pri func decoder.filter_4_distance_4_fallback!(curr: slice base.u8, prev: slice
} else {
fa0 = fc0
}
c[0] = c[0] ~mod+ ((fa0 & 0xFF) as base.u8)
fa0 = c[0] as base.u32
curr[0] = curr[0] ~mod+ ((fa0 & 0xFF) as base.u8)
fa0 = curr[0] as base.u32
fc0 = fb0
fb1 = p[1] as base.u32
fb1 = prev[1] as base.u32
pp1 = (fa1 ~mod+ fb1) ~mod- fc1
pa1 = pp1 ~mod- fa1
if pa1 >= 0x8000_0000 {
@@ -470,11 +476,11 @@ pri func decoder.filter_4_distance_4_fallback!(curr: slice base.u8, prev: slice
} else {
fa1 = fc1
}
c[1] = c[1] ~mod+ ((fa1 & 0xFF) as base.u8)
fa1 = c[1] as base.u32
curr[1] = curr[1] ~mod+ ((fa1 & 0xFF) as base.u8)
fa1 = curr[1] as base.u32
fc1 = fb1
fb2 = p[2] as base.u32
fb2 = prev[2] as base.u32
pp2 = (fa2 ~mod+ fb2) ~mod- fc2
pa2 = pp2 ~mod- fa2
if pa2 >= 0x8000_0000 {
@@ -495,11 +501,11 @@ pri func decoder.filter_4_distance_4_fallback!(curr: slice base.u8, prev: slice
} else {
fa2 = fc2
}
c[2] = c[2] ~mod+ ((fa2 & 0xFF) as base.u8)
fa2 = c[2] as base.u32
curr[2] = curr[2] ~mod+ ((fa2 & 0xFF) as base.u8)
fa2 = curr[2] as base.u32
fc2 = fb2
fb3 = p[3] as base.u32
fb3 = prev[3] as base.u32
pp3 = (fa3 ~mod+ fb3) ~mod- fc3
pa3 = pp3 ~mod- fa3
if pa3 >= 0x8000_0000 {
@@ -520,8 +526,8 @@ pri func decoder.filter_4_distance_4_fallback!(curr: slice base.u8, prev: slice
} else {
fa3 = fc3
}
c[3] = c[3] ~mod+ ((fa3 & 0xFF) as base.u8)
fa3 = c[3] as base.u32
curr[3] = curr[3] ~mod+ ((fa3 & 0xFF) as base.u8)
fa3 = curr[3] as base.u32
fc3 = fb3
}
}
+39 -34
View File
@@ -22,19 +22,20 @@
// pri func decoder.filter_1_distance_3_x86_sse42!(curr: slice base.u8),
// choose cpu_arch >= x86_sse42,
// {
// var c : slice base.u8
// var curr : slice base.u8
//
// 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())
// iterate (curr = args.curr)(length: 4, advance: 3, unroll: 2) {
// x128.load_u32!(a: curr.peek_u32le())
// x128 = x128._mm_add_epi8(b: a128)
// a128 = x128
// c.poke_u24le!(a: x128.truncate_u32())
// curr.poke_u24le!(a: x128.truncate_u32())
// } else (length: 3, advance: 3, unroll: 1) {
// x128.load_u32!(a: c.peek_u24le_as_u32())
// x128.load_u32!(a: curr.peek_u24le_as_u32())
// x128 = x128._mm_add_epi8(b: a128)
// c.poke_u24le!(a: x128.truncate_u32())
// curr.poke_u24le!(a: x128.truncate_u32())
// }
// }
//
@@ -45,15 +46,16 @@
pri func decoder.filter_1_distance_4_x86_sse42!(curr: slice base.u8),
choose cpu_arch >= x86_sse42,
{
var c : slice base.u8
var curr : slice base.u8
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())
iterate (curr = args.curr)(length: 4, advance: 4, unroll: 2) {
x128.load_u32!(a: curr.peek_u32le())
x128 = x128._mm_add_epi8(b: a128)
a128 = x128
c.poke_u32le!(a: x128.truncate_u32())
curr.poke_u32le!(a: x128.truncate_u32())
}
}
@@ -73,8 +75,9 @@ pri func decoder.filter_1_distance_4_x86_sse42!(curr: slice base.u8),
pri func decoder.filter_3_distance_4_x86_sse42!(curr: slice base.u8, prev: slice base.u8),
choose cpu_arch >= x86_sse42,
{
var c : slice base.u8
var p : slice base.u8
var curr : slice base.u8
var prev : slice base.u8
var x128 : base.x86_m128i
var a128 : base.x86_m128i
var b128 : base.x86_m128i
@@ -83,7 +86,7 @@ pri func decoder.filter_3_distance_4_x86_sse42!(curr: slice base.u8, prev: slice
if args.prev.length() == 0 {
k128 = k128.create_mm_set1_epi8(a: 0xFE)
iterate (c = args.curr)(length: 4, advance: 4, unroll: 2) {
iterate (curr = args.curr)(length: 4, advance: 4, unroll: 2) {
// The predictor, p128, is just half (rounded down) of the previous
// pixel, a128. In this branch, b128 stays zero so the average of
// a128 and b128 is just half of a128. _mm_avg_epu8 rounds up, but
@@ -92,17 +95,17 @@ pri func decoder.filter_3_distance_4_x86_sse42!(curr: slice base.u8, prev: slice
// Add the predictor to the residual and, for the next iteration,
// set its previous pixel, a128, to this one, x128.
x128.load_u32!(a: c.peek_u32le())
x128.load_u32!(a: curr.peek_u32le())
x128 = x128._mm_add_epi8(b: p128)
a128 = x128
c.poke_u32le!(a: x128.truncate_u32())
curr.poke_u32le!(a: x128.truncate_u32())
}
} else {
k128 = k128.create_mm_set1_epi8(a: 0x01)
iterate (c = args.curr, p = args.prev)(length: 4, advance: 4, unroll: 2) {
iterate (curr = args.curr, prev = args.prev)(length: 4, advance: 4, unroll: 2) {
// Load the pixel from the row above.
b128.load_u32!(a: p.peek_u32le())
b128.load_u32!(a: prev.peek_u32le())
// The predictor, p128, is the average (rounded down) of the
// previous pixel, a128, and the pixel above, b128.
@@ -115,10 +118,10 @@ pri func decoder.filter_3_distance_4_x86_sse42!(curr: slice base.u8, prev: slice
// Add the predictor to the residual and, for the next iteration,
// set its previous pixel, a128, to this one, x128.
x128.load_u32!(a: c.peek_u32le())
x128.load_u32!(a: curr.peek_u32le())
x128 = x128._mm_add_epi8(b: p128)
a128 = x128
c.poke_u32le!(a: x128.truncate_u32())
curr.poke_u32le!(a: x128.truncate_u32())
}
}
}
@@ -137,8 +140,9 @@ pri func decoder.filter_4_distance_3_x86_sse42!(curr: slice base.u8, prev: slice
//
// Differences between that function and this one are marked with a §.
var c : slice base.u8
var p : slice base.u8
var curr : slice base.u8
var prev : slice base.u8
var x128 : base.x86_m128i
var a128 : base.x86_m128i
var b128 : base.x86_m128i
@@ -151,8 +155,8 @@ pri func decoder.filter_4_distance_3_x86_sse42!(curr: slice base.u8, prev: slice
var z128 : base.x86_m128i
// § The advance is 3, not 4.
iterate (c = args.curr, p = args.prev)(length: 4, advance: 3, unroll: 2) {
b128.load_u32!(a: p.peek_u32le())
iterate (curr = args.curr, prev = args.prev)(length: 4, advance: 3, unroll: 2) {
b128.load_u32!(a: prev.peek_u32le())
b128 = b128._mm_unpacklo_epi8(b: z128)
pa128 = b128._mm_sub_epi16(b: c128)
pb128 = a128._mm_sub_epi16(b: c128)
@@ -166,19 +170,19 @@ pri func decoder.filter_4_distance_3_x86_sse42!(curr: slice base.u8, prev: slice
mask: smallest128._mm_cmpeq_epi16(b: pb128))._mm_blendv_epi8(
b: a128,
mask: smallest128._mm_cmpeq_epi16(b: pa128))
x128.load_u32!(a: c.peek_u32le())
x128.load_u32!(a: curr.peek_u32le())
x128 = x128._mm_unpacklo_epi8(b: z128)
x128 = x128._mm_add_epi8(b: p128)
a128 = x128
c128 = b128
x128 = x128._mm_packus_epi16(b: x128)
// § poke_u24le replaces poke_u32le.
c.poke_u24le!(a: x128.truncate_u32())
curr.poke_u24le!(a: x128.truncate_u32())
// § The length and advance are both 3, not 4.
} else (length: 3, advance: 3, unroll: 1) {
// § peek_u24le_as_u32 replaces peek_u32le.
b128.load_u32!(a: p.peek_u24le_as_u32())
b128.load_u32!(a: prev.peek_u24le_as_u32())
b128 = b128._mm_unpacklo_epi8(b: z128)
pa128 = b128._mm_sub_epi16(b: c128)
pb128 = a128._mm_sub_epi16(b: c128)
@@ -193,7 +197,7 @@ pri func decoder.filter_4_distance_3_x86_sse42!(curr: slice base.u8, prev: slice
b: a128,
mask: smallest128._mm_cmpeq_epi16(b: pa128))
// § peek_u24le_as_u32 replaces peek_u32le.
x128.load_u32!(a: c.peek_u24le_as_u32())
x128.load_u32!(a: curr.peek_u24le_as_u32())
x128 = x128._mm_unpacklo_epi8(b: z128)
x128 = x128._mm_add_epi8(b: p128)
// § These assignments are unnecessary; this is the last iteration.
@@ -201,15 +205,16 @@ pri func decoder.filter_4_distance_3_x86_sse42!(curr: slice base.u8, prev: slice
// c128 = b128
x128 = x128._mm_packus_epi16(b: x128)
// § poke_u24le replaces poke_u32le.
c.poke_u24le!(a: x128.truncate_u32())
curr.poke_u24le!(a: x128.truncate_u32())
}
}
pri func decoder.filter_4_distance_4_x86_sse42!(curr: slice base.u8, prev: slice base.u8),
choose cpu_arch >= x86_sse42,
{
var c : slice base.u8
var p : slice base.u8
var curr : slice base.u8
var prev : slice base.u8
var x128 : base.x86_m128i
var a128 : base.x86_m128i
var b128 : base.x86_m128i
@@ -221,9 +226,9 @@ pri func decoder.filter_4_distance_4_x86_sse42!(curr: slice base.u8, prev: slice
var smallest128 : base.x86_m128i
var z128 : base.x86_m128i
iterate (c = args.curr, p = args.prev)(length: 4, advance: 4, unroll: 2) {
iterate (curr = args.curr, prev = args.prev)(length: 4, advance: 4, unroll: 2) {
// Load the pixel from the row above.
b128.load_u32!(a: p.peek_u32le())
b128.load_u32!(a: prev.peek_u32le())
// Convert from u8 to i16 by unpacking it with zeroes.
b128 = b128._mm_unpacklo_epi8(b: z128)
@@ -258,12 +263,12 @@ pri func decoder.filter_4_distance_4_x86_sse42!(curr: slice base.u8, prev: slice
// its previous pixels, a128 and c128, to x128 and b128.
//
// Again, convert between u8 and i16 by unpacking and re-packing.
x128.load_u32!(a: c.peek_u32le())
x128.load_u32!(a: curr.peek_u32le())
x128 = x128._mm_unpacklo_epi8(b: z128)
x128 = x128._mm_add_epi8(b: p128)
a128 = x128
c128 = b128
x128 = x128._mm_packus_epi16(b: x128)
c.poke_u32le!(a: x128.truncate_u32())
curr.poke_u32le!(a: x128.truncate_u32())
}
}