Merge branch 'simd/master' into simd/7.0.x

CHANGES.SIMD.rst

@@ -1,6 +1,11 @@
 Changelog (Pillow-SIMD)
 =======================
 
+7.0.0.post4
+-----------
+
+- Filter: fixed wrong offset handling for 3x3 single-band version
+
 7.0.0.post3
 -----------
 

README.md

@@ -40,7 +40,7 @@ The project is supported by Uploadcare, a SAAS for cloud-based image storing and
 
 [![Uploadcare][uploadcare.logo]][uploadcare.com]
 
-In fact, Uploadcare has been running Pillow-SIMD for about three years now.
+In fact, Uploadcare has been running Pillow-SIMD since 2015.
 
 The following image operations are currently SIMD-accelerated:
 
@@ -96,6 +96,7 @@ So there is no easy way to compile such library, especially with setuptools.
 
 If there's a copy of the original Pillow installed, it has to be removed first
 with `$ pip uninstall -y pillow`.
+Please install [prerequisites](https://pillow.readthedocs.io/en/stable/installation.html#building-from-source) for your platform.
 The installation itself is simple just as running `$ pip install pillow-simd`, 
 and if you're using SSE4-capable CPU everything should run smoothly.
 If you'd like to install the AVX2-enabled version, 
@@ -124,4 +125,4 @@ All bugfixes to the original Pillow will then be transferred to the next Pillow-
   [pillow-perf-page]: https://python-pillow.github.io/pillow-perf/
   [pillow-perf-repo]: https://github.com/python-pillow/pillow-perf
   [uploadcare.com]: https://uploadcare.com/?utm_source=github&utm_medium=description&utm_campaign=pillow-simd
-  [uploadcare.logo]: https://ucarecdn.com/74c4d283-f7cf-45d7-924c-fc77345585af/uploadcare.svg
+  [uploadcare.logo]: https://ucarecdn.com/8eca784b-bbe5-4f7e-8cdf-98d75aab8cec/logotransparent.svg

src/PIL/_version.py

@@ -1,2 +1,2 @@
 # Master version for Pillow
-__version__ = "7.0.0.post3"
+__version__ = "7.0.0.post4"

src/libImaging/FilterSIMD_3x3f_u8.c

@@ -3,14 +3,12 @@ ImagingFilter3x3f_u8(Imaging imOut, Imaging im, const float* kernel,
                      float offset)
 {
 #define MM_KERNEL1x3_SUM1(ss, row, kernel) \
-    ss = _mm_set1_ps(offset); \
-    ss = _mm_add_ps(ss, _mm_mul_ps(pix0##row, kernel0##kernel)); \
+    ss = _mm_mul_ps(pix0##row, kernel0##kernel); \
     ss = _mm_add_ps(ss, _mm_mul_ps(pix1##row, kernel1##kernel)); \
     ss = _mm_add_ps(ss, _mm_mul_ps(pix2##row, kernel2##kernel));
 
 #define MM_KERNEL1x3_SUM1_2(ss, row, kernel) \
-    ss = _mm_set1_ps(offset); \
-    ss = _mm_add_ps(ss, _mm_mul_ps(pix3##row, kernel0##kernel)); \
+    ss = _mm_mul_ps(pix3##row, kernel0##kernel); \
     ss = _mm_add_ps(ss, _mm_mul_ps(pix0##row, kernel1##kernel)); \
     ss = _mm_add_ps(ss, _mm_mul_ps(pix1##row, kernel2##kernel));
 
@@ -26,6 +24,7 @@ ImagingFilter3x3f_u8(Imaging imOut, Imaging im, const float* kernel,
     __m128 kernel01 = _mm_set_ps(kernel[2], kernel[1], kernel[0], 0);
     __m128 kernel11 = _mm_set_ps(kernel[5], kernel[4], kernel[3], 0);
     __m128 kernel21 = _mm_set_ps(kernel[8], kernel[7], kernel[6], 0);
+    __m128 mm_offset = _mm_set1_ps(offset);
 
     memcpy(imOut->image8[0], im->image8[0], im->linesize);
     y = 1;
@@ -64,12 +63,14 @@ ImagingFilter3x3f_u8(Imaging imOut, Imaging im, const float* kernel,
             ss4 = _mm_hadd_ps(ss4, ss5);
 
             ss0 = _mm_hadd_ps(ss0, ss1);
+            ss0 = _mm_add_ps(ss0, mm_offset);
             ssi0 = _mm_cvtps_epi32(ss0);
             ssi0 = _mm_packs_epi32(ssi0, ssi0);
             ssi0 = _mm_packus_epi16(ssi0, ssi0);
             *((UINT32*) &out0[x]) = _mm_cvtsi128_si32(ssi0);
 
             ss3 = _mm_hadd_ps(ss3, ss4);
+            ss3 = _mm_add_ps(ss3, mm_offset);
             ssi0 = _mm_cvtps_epi32(ss3);
             ssi0 = _mm_packs_epi32(ssi0, ssi0);
             ssi0 = _mm_packus_epi16(ssi0, ssi0);
@@ -89,13 +90,15 @@ ImagingFilter3x3f_u8(Imaging imOut, Imaging im, const float* kernel,
 
             ss0 = _mm_hadd_ps(ss0, ss0);
             ss0 = _mm_hadd_ps(ss0, ss0);
+            ss0 = _mm_add_ps(ss0, mm_offset);
             ssi0 = _mm_cvtps_epi32(ss0);
             ssi0 = _mm_packs_epi32(ssi0, ssi0);
             ssi0 = _mm_packus_epi16(ssi0, ssi0);
             out0[x] = _mm_cvtsi128_si32(ssi0);
 
             ss1 = _mm_hadd_ps(ss1, ss1);
             ss1 = _mm_hadd_ps(ss1, ss1);
+            ss1 = _mm_add_ps(ss1, mm_offset);
             ssi0 = _mm_cvtps_epi32(ss1);
             ssi0 = _mm_packs_epi32(ssi0, ssi0);
             ssi0 = _mm_packus_epi16(ssi0, ssi0);
@@ -122,6 +125,7 @@ ImagingFilter3x3f_u8(Imaging imOut, Imaging im, const float* kernel,
 
             ss = _mm_hadd_ps(ss, ss);
             ss = _mm_hadd_ps(ss, ss);
+            ss = _mm_add_ps(ss, mm_offset);
             ssi0 = _mm_cvtps_epi32(ss);
             ssi0 = _mm_packs_epi32(ssi0, ssi0);
             ssi0 = _mm_packus_epi16(ssi0, ssi0);
@@ -139,6 +143,7 @@ ImagingFilter3x3f_u8(Imaging imOut, Imaging im, const float* kernel,
 
             ss = _mm_hadd_ps(ss, ss);
             ss = _mm_hadd_ps(ss, ss);
+            ss = _mm_add_ps(ss, mm_offset);
             ssi0 = _mm_cvtps_epi32(ss);
             ssi0 = _mm_packs_epi32(ssi0, ssi0);
             ssi0 = _mm_packus_epi16(ssi0, ssi0);

src/libImaging/FilterSIMD_5x5f_u8.c

@@ -10,8 +10,7 @@ ImagingFilter5x5f_u8(Imaging imOut, Imaging im, const float* kernel,
     pix4##row = _mm_cvtepi32_ps(mm_cvtepu8_epi32(&in_2[x]));
 
 #define MM_KERNEL1x5_SUM(ss, row, krow) \
-    ss = _mm_setzero_ps(); \
-    ss = _mm_add_ps(ss, _mm_mul_ps(pix0##row, kernel0##krow)); \
+    ss = _mm_mul_ps(pix0##row, kernel0##krow); \
     ss = _mm_add_ps(ss, _mm_mul_ps(pix1##row, kernel1##krow)); \
     ss = _mm_add_ps(ss, _mm_mul_ps(pix2##row, kernel2##krow)); \
     ss = _mm_add_ps(ss, _mm_mul_ps(pix3##row, kernel3##krow)); \