morgan/FreeRDP
1
0
Fork 0
mirror of https://github.com/morgan9e/FreeRDP synced 2026-04-15 00:44:19 +09:00
Code Issues Packages Projects Releases Wiki Activity

[primitives,copy] remove alignment check

Browse Source
This commit is contained in:
akallabeth
2025-03-06 16:09:06 +01:00
parent 5139a3a3e5
commit aa4384a286
2 changed files with 25 additions and 32 deletions
Download Patch File Download Diff File Expand all files Collapse all files

36
libfreerdp/primitives/sse/prim_copy_avx2.c
View File

@@ -61,7 +61,6 @@ static INLINE pstatus_t avx2_image_copy_bgr24_bgrx32(BYTE* WINPR_RESTRICT pDstDa
const SSIZE_T width = nWidth - rem;
const size_t align = nSrcStep % 32;
const BOOL fast = (align == 0) ? TRUE : (align >= 8 - MIN(8, (size_t)rem) ? TRUE : FALSE);
for (SSIZE_T y = 0; y < nHeight; y++)
{
const BYTE* WINPR_RESTRICT srcLine =
@@ -72,29 +71,26 @@ static INLINE pstatus_t avx2_image_copy_bgr24_bgrx32(BYTE* WINPR_RESTRICT pDstDa
SSIZE_T x = 0;
/* Ensure alignment requirements can be met */
if (fast)
for (; x < width; x += 8)
{
for (; x < width; x += 8)
{
const __m256i* src = (const __m256i*)&srcLine[(x + nXSrc) * srcByte];
__m256i* dst = (__m256i*)&dstLine[(x + nXDst) * dstByte];
const __m256i s0 = _mm256_loadu_si256(src);
__m256i s1 = _mm256_shuffle_epi8(s0, smask);
const __m256i* src = (const __m256i*)&srcLine[(x + nXSrc) * srcByte];
__m256i* dst = (__m256i*)&dstLine[(x + nXDst) * dstByte];
const __m256i s0 = _mm256_loadu_si256(src);
__m256i s1 = _mm256_shuffle_epi8(s0, smask);
/* _mm256_shuffle_epi8 can not cross 128bit lanes.
* manually copy these bytes with extract/insert */
const __m256i sx = _mm256_broadcastsi128_si256(_mm256_extractf128_si256(s0, 0));
const __m256i sxx = _mm256_shuffle_epi8(sx, shelpmask);
const __m256i bmask =
_mm256_set_epi32(0x00000000, 0x00000000, 0x000000FF, 0x00FFFFFF, 0x00000000,
0x00000000, 0x00000000, 0x00000000);
const __m256i merged = _mm256_blendv_epi8(s1, sxx, bmask);
/* _mm256_shuffle_epi8 can not cross 128bit lanes.
* manually copy these bytes with extract/insert */
const __m256i sx = _mm256_broadcastsi128_si256(_mm256_extractf128_si256(s0, 0));
const __m256i sxx = _mm256_shuffle_epi8(sx, shelpmask);
const __m256i bmask = _mm256_set_epi32(0x00000000, 0x00000000, 0x000000FF, 0x00FFFFFF,
0x00000000, 0x00000000, 0x00000000, 0x00000000);
const __m256i merged = _mm256_blendv_epi8(s1, sxx, bmask);
const __m256i s2 = _mm256_loadu_si256(dst);
__m256i d0 = _mm256_blendv_epi8(merged, s2, mask);
_mm256_storeu_si256(dst, d0);
}
const __m256i s2 = _mm256_loadu_si256(dst);
__m256i d0 = _mm256_blendv_epi8(merged, s2, mask);
_mm256_storeu_si256(dst, d0);
}
for (; x < nWidth; x++)
{
const BYTE* src = &srcLine[(x + nXSrc) * srcByte];

21
libfreerdp/primitives/sse/prim_copy_sse4_1.c
View File

@@ -50,7 +50,6 @@ static INLINE pstatus_t sse_image_copy_bgr24_bgrx32(BYTE* WINPR_RESTRICT pDstDat
const UINT32 rem = nWidth % 4;
const size_t align = nSrcStep % 64;
const BOOL fast = (align == 0) ? TRUE : (align >= 16 - MIN(16, (size_t)rem) ? TRUE : FALSE);
const SSIZE_T width = nWidth - rem;
for (SSIZE_T y = 0; y < nHeight; y++)
{
@@ -61,20 +60,18 @@ static INLINE pstatus_t sse_image_copy_bgr24_bgrx32(BYTE* WINPR_RESTRICT pDstDat
SSIZE_T x = 0;
/* Ensure alignment requirements can be met */
if (fast)
for (; x < width; x += 4)
{
for (; x < width; x += 4)
{
const __m128i* src = (const __m128i*)&srcLine[(x + nXSrc) * srcByte];
__m128i* dst = (__m128i*)&dstLine[(x + nXDst) * dstByte];
const __m128i s0 = LOAD_SI128(src);
const __m128i s1 = _mm_shuffle_epi8(s0, smask);
const __m128i s2 = LOAD_SI128(dst);
const __m128i* src = (const __m128i*)&srcLine[(x + nXSrc) * srcByte];
__m128i* dst = (__m128i*)&dstLine[(x + nXDst) * dstByte];
const __m128i s0 = LOAD_SI128(src);
const __m128i s1 = _mm_shuffle_epi8(s0, smask);
const __m128i s2 = LOAD_SI128(dst);
__m128i d0 = _mm_blendv_epi8(s1, s2, mask);
STORE_SI128(dst, d0);
}
__m128i d0 = _mm_blendv_epi8(s1, s2, mask);
STORE_SI128(dst, d0);
}
for (; x < nWidth; x++)
{
const BYTE* src = &srcLine[(x + nXSrc) * srcByte];