From 5e74365e556e9688b9167b005a6f7dc883bbb853 Mon Sep 17 00:00:00 2001
From: Jeremy Rand <[EMAIL REDACTED]>
Date: Sat, 6 May 2023 03:24:02 +0000
Subject: [PATCH] SDL_blit_N.c: Move ppc64le swizzle outside of loop
An in-place swizzle mutation was erroneously inside of a loop, which
caused each consecutive 4-pixel vector to alternate between correct and
incorrect endianness.
The bug was introduced in 715e070d299fc547cce8c52915b75e6316af6bff.
Thanks to RobbieAB for reporting the bug.
Fixes https://github.com/libsdl-org/SDL/issues/3428
(cherry picked from commit 9142292f4ad6d9bc8a2bc7c874e2fde54befeb9b)
---
src/video/SDL_blit_N.c | 27 +++++++++++++++------------
1 file changed, 15 insertions(+), 12 deletions(-)
diff --git a/src/video/SDL_blit_N.c b/src/video/SDL_blit_N.c
index 4bde9df49c8b..39dfeefe61fc 100644
--- a/src/video/SDL_blit_N.c
+++ b/src/video/SDL_blit_N.c
@@ -624,6 +624,11 @@ static void Blit32to32KeyAltivec(SDL_BlitInfo *info)
((unsigned int *)(char *)&vrgbmask)[0] = rgbmask;
vrgbmask = vec_splat(vrgbmask, 0);
+#if defined(__powerpc__) && (SDL_BYTEORDER == SDL_LIL_ENDIAN)
+ /* reorder bytes for PowerPC little endian */
+ vpermute = reorder_ppc64le_vec(vpermute);
+#endif
+
while (height--) {
#define ONE_PIXEL_BLEND(condition, widthvar) \
if (copy_alpha) { \
@@ -673,10 +678,6 @@ static void Blit32to32KeyAltivec(SDL_BlitInfo *info)
/* vsel is set for items that match the key */
vsel = (vector unsigned char)vec_and(vs, vrgbmask);
vsel = (vector unsigned char)vec_cmpeq(vs, vckey);
-#if defined(__powerpc__) && (SDL_BYTEORDER == SDL_LIL_ENDIAN)
- /* reorder bytes for PowerPC little endian */
- vpermute = reorder_ppc64le_vec(vpermute);
-#endif
/* permute the src vec to the dest format */
vs = vec_perm(vs, valpha, vpermute);
/* load the destination vec */
@@ -724,6 +725,11 @@ static void ConvertAltivec32to32_noprefetch(SDL_BlitInfo *info)
SDL_assert(srcfmt->BytesPerPixel == 4);
SDL_assert(dstfmt->BytesPerPixel == 4);
+#if defined(__powerpc__) && (SDL_BYTEORDER == SDL_LIL_ENDIAN)
+ /* reorder bytes for PowerPC little endian */
+ vpermute = reorder_ppc64le_vec(vpermute);
+#endif
+
while (height--) {
vector unsigned char valigner;
vector unsigned int vbits;
@@ -755,10 +761,6 @@ static void ConvertAltivec32to32_noprefetch(SDL_BlitInfo *info)
src += 4;
width -= 4;
vbits = vec_perm(vbits, voverflow, valigner); /* src is ready. */
-#if defined(__powerpc__) && (SDL_BYTEORDER == SDL_LIL_ENDIAN)
- /* reorder bytes for PowerPC little endian */
- vpermute = reorder_ppc64le_vec(vpermute);
-#endif
vbits = vec_perm(vbits, vzero, vpermute); /* swizzle it. */
vec_st(vbits, 0, dst); /* store it back out. */
dst += 4;
@@ -809,6 +811,11 @@ static void ConvertAltivec32to32_prefetch(SDL_BlitInfo *info)
SDL_assert(srcfmt->BytesPerPixel == 4);
SDL_assert(dstfmt->BytesPerPixel == 4);
+#if defined(__powerpc__) && (SDL_BYTEORDER == SDL_LIL_ENDIAN)
+ /* reorder bytes for PowerPC little endian */
+ vpermute = reorder_ppc64le_vec(vpermute);
+#endif
+
while (height--) {
vector unsigned char valigner;
vector unsigned int vbits;
@@ -848,10 +855,6 @@ static void ConvertAltivec32to32_prefetch(SDL_BlitInfo *info)
src += 4;
width -= 4;
vbits = vec_perm(vbits, voverflow, valigner); /* src is ready. */
-#if defined(__powerpc__) && (SDL_BYTEORDER == SDL_LIL_ENDIAN)
- /* reorder bytes for PowerPC little endian */
- vpermute = reorder_ppc64le_vec(vpermute);
-#endif
vbits = vec_perm(vbits, vzero, vpermute); /* swizzle it. */
vec_st(vbits, 0, dst); /* store it back out. */
dst += 4;