# HG changeset patch

# User Andrew Osmond <aosmond@mozilla.com>

# Date 1579706360 0

# Node ID b3d8b08265b800165d684281d19ac845a8ff9a66

# Parent  50c371b37a9fcd994a5866db73bd0d078e19f95d

Bug 1610814 - Fix NEON compile error with gcc and RGB unpacking. r=lsalzman

This patch makes us use the correct intrinsic for loading a uint8x16

register. It is not entirely clear why clang accepts this without

complaint but beyond the types, it should be equivalent.

Differential Revision: https://phabricator.services.mozilla.com/D60667

diff --git a/gfx/2d/SwizzleNEON.cpp b/gfx/2d/SwizzleNEON.cpp

--- a/gfx/2d/SwizzleNEON.cpp

+++ b/gfx/2d/SwizzleNEON.cpp

@@ -407,25 +407,25 @@ void UnpackRowRGB24_NEON(const uint8_t* 

   }

   uint8x16_t alpha = vreinterpretq_u8_u32(vdupq_n_u32(0xFF000000));

   // Process all 4-pixel chunks as one vector.

   src -= 4 * 3;

   dst -= 4 * 4;

   while (src >= aSrc) {

-    uint8x16_t px = vld1q_u16(reinterpret_cast<const uint16_t*>(src));

+    uint8x16_t px = vld1q_u8(src);

     // G2R2B1G1 R1B0G0R0 -> X1R1G1B1 X0R0G0B0

     uint8x8_t pxlo = vtbl1_u8(vget_low_u8(px), masklo);

     // B3G3R3B2 G2R2B1G1 -> X3R3G3B3 X2R2G2B2

     uint8x8_t pxhi =

         vtbl1_u8(vext_u8(vget_low_u8(px), vget_high_u8(px), 4), maskhi);

     px = vcombine_u8(pxlo, pxhi);

     px = vorrq_u8(px, alpha);

-    vst1q_u16(reinterpret_cast<uint16_t*>(dst), px);

+    vst1q_u8(dst, px);

     src -= 4 * 3;

     dst -= 4 * 4;

   }

 }

 // Force instantiation of swizzle variants here.

 template void UnpackRowRGB24_NEON<false>(const uint8_t*, uint8_t*, int32_t);

 template void UnpackRowRGB24_NEON<true>(const uint8_t*, uint8_t*, int32_t);