aaru-dps/Aaru
1
0
Fork 0
mirror of https://github.com/aaru-dps/Aaru.git synced 2025-12-16 19:24:25 +00:00
Code Issues Packages Projects Releases Wiki Activity

General refactor and cleanup.

Browse Source
This commit is contained in:
Natalia Portillo
2024-05-01 04:05:22 +01:00
parent 185a8c3fd5
commit e46d21bde6
922 changed files with 36437 additions and 29485 deletions
Download Patch File Download Diff File Expand all files Collapse all files

77
Aaru.Checksums/Adler32/neon.cs
View File

@@ -73,8 +73,7 @@ static class Neon
{
uint n = Adler32Context.NMAX / blockSize; /* The NMAX constraint. */
if(n > blocks)
n = blocks;
if(n > blocks) n = blocks;
blocks -= n;
/*
@@ -93,17 +92,41 @@ static class Neon
/*
* Load 32 input bytes.
*/
var bytes1 = Vector128.Create(buf[bufPos], buf[bufPos + 1], buf[bufPos + 2], buf[bufPos + 3],
buf[bufPos + 4], buf[bufPos + 5], buf[bufPos + 6], buf[bufPos + 7],
buf[bufPos + 8], buf[bufPos + 9], buf[bufPos + 10], buf[bufPos + 11],
buf[bufPos + 12], buf[bufPos + 13], buf[bufPos + 14], buf[bufPos + 15]);
var bytes1 = Vector128.Create(buf[bufPos],
buf[bufPos + 1],
buf[bufPos + 2],
buf[bufPos + 3],
buf[bufPos + 4],
buf[bufPos + 5],
buf[bufPos + 6],
buf[bufPos + 7],
buf[bufPos + 8],
buf[bufPos + 9],
buf[bufPos + 10],
buf[bufPos + 11],
buf[bufPos + 12],
buf[bufPos + 13],
buf[bufPos + 14],
buf[bufPos + 15]);
bufPos += 16;
var bytes2 = Vector128.Create(buf[bufPos], buf[bufPos + 1], buf[bufPos + 2], buf[bufPos + 3],
buf[bufPos + 4], buf[bufPos + 5], buf[bufPos + 6], buf[bufPos + 7],
buf[bufPos + 8], buf[bufPos + 9], buf[bufPos + 10], buf[bufPos + 11],
buf[bufPos + 12], buf[bufPos + 13], buf[bufPos + 14], buf[bufPos + 15]);
var bytes2 = Vector128.Create(buf[bufPos],
buf[bufPos + 1],
buf[bufPos + 2],
buf[bufPos + 3],
buf[bufPos + 4],
buf[bufPos + 5],
buf[bufPos + 6],
buf[bufPos + 7],
buf[bufPos + 8],
buf[bufPos + 9],
buf[bufPos + 10],
buf[bufPos + 11],
buf[bufPos + 12],
buf[bufPos + 13],
buf[bufPos + 14],
buf[bufPos + 15]);
bufPos += 16;
/*
@@ -114,11 +137,11 @@ static class Neon
/*
* Horizontally add the bytes for s1.
*/
vS1 =
AdvSimd.AddPairwiseWideningAndAdd(vS1,
AdvSimd.
AddPairwiseWideningAndAdd(AdvSimd.AddPairwiseWidening(bytes1),
bytes2));
vS1 = AdvSimd.AddPairwiseWideningAndAdd(vS1,
AdvSimd
.AddPairwiseWideningAndAdd(AdvSimd
.AddPairwiseWidening(bytes1),
bytes2));
/*
* Vertically add the bytes for s2.
@@ -134,22 +157,28 @@ static class Neon
/*
* Multiply-add bytes by [ 32, 31, 30, ... ] for s2.
*/
vS2 = AdvSimd.MultiplyWideningLowerAndAdd(vS2, vColumnSum1.GetLower(),
vS2 = AdvSimd.MultiplyWideningLowerAndAdd(vS2,
vColumnSum1.GetLower(),
Vector64.Create((ushort)32, 31, 30, 29));
vS2 = AdvSimd.MultiplyWideningLowerAndAdd(vS2, vColumnSum1.GetUpper(),
vS2 = AdvSimd.MultiplyWideningLowerAndAdd(vS2,
vColumnSum1.GetUpper(),
Vector64.Create((ushort)28, 27, 26, 25));
vS2 = AdvSimd.MultiplyWideningLowerAndAdd(vS2, vColumnSum2.GetLower(),
vS2 = AdvSimd.MultiplyWideningLowerAndAdd(vS2,
vColumnSum2.GetLower(),
Vector64.Create((ushort)24, 23, 22, 21));
vS2 = AdvSimd.MultiplyWideningLowerAndAdd(vS2, vColumnSum2.GetUpper(),
vS2 = AdvSimd.MultiplyWideningLowerAndAdd(vS2,
vColumnSum2.GetUpper(),
Vector64.Create((ushort)20, 19, 18, 17));
vS2 = AdvSimd.MultiplyWideningLowerAndAdd(vS2, vColumnSum3.GetLower(),
vS2 = AdvSimd.MultiplyWideningLowerAndAdd(vS2,
vColumnSum3.GetLower(),
Vector64.Create((ushort)16, 15, 14, 13));
vS2 = AdvSimd.MultiplyWideningLowerAndAdd(vS2, vColumnSum3.GetUpper(),
vS2 = AdvSimd.MultiplyWideningLowerAndAdd(vS2,
vColumnSum3.GetUpper(),
Vector64.Create((ushort)12, 11, 10, 9));
vS2 = AdvSimd.MultiplyWideningLowerAndAdd(vS2, vColumnSum4.GetLower(), Vector64.Create((ushort)8, 7, 6, 5));
@@ -197,11 +226,9 @@ static class Neon
len -= 16;
}
while(len-- != 0)
s2 += s1 += buf[bufPos++];
while(len-- != 0) s2 += s1 += buf[bufPos++];
if(s1 >= Adler32Context.ADLER_MODULE)
s1 -= Adler32Context.ADLER_MODULE;
if(s1 >= Adler32Context.ADLER_MODULE) s1 -= Adler32Context.ADLER_MODULE;
s2 %= Adler32Context.ADLER_MODULE;
}