<div dir="ltr"><div><div class="gmail_extra"><div class="gmail_quote">On Wed, Mar 4, 2015 at 4:56 PM, Matt Turner <<a href="mailto:mattst88@gmail.com" target="_blank">mattst88@gmail.com</a>> wrote: <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">The SSSE3 swizzling code was written for fast uploads to the GPU and assumed the destination was always 16-byte aligned. When we began using this code for fast downloads as well we didn't do anything to account for the fact that the destination pointer given by glReadPixels() or glGetTexImage() is not guaranteed to be suitably aligned. With SSSE3 enabled (at compile-time), some applications would crash when an SSE aligned-store instruction tried to store to an unaligned destination (or an assertion that the destination is aligned would trigger). To remedy this, tell intel_get_memcpy() whether we're uploading or downloading so that it can select whether to assume the destination or source is aligned, respectively. Cc: 10.5 <<a href="mailto:mesa-stable@lists.freedesktop.org">mesa-stable@lists.freedesktop.org</a>> Bugzilla: <a href="https://bugs.freedesktop.org/show_bug.cgi?id=89416" target="_blank">https://bugs.freedesktop.org/show_bug.cgi?id=89416</a> Tested-by: Uriy Zhuravlev <<a href="mailto:stalkerg@gmail.com">stalkerg@gmail.com</a>> [v1] Reviewed-by: Jason Ekstrand <<a href="mailto:jason.ekstrand@intel.com">jason.ekstrand@intel.com</a>> --- Benchmarks I should run? </blockquote><div> </div><div>I'm not worried about it. Maybe changing from aligned to unaligned reads makes more or less of a difference than changing writes. </div><div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> src/mesa/drivers/dri/i965/intel_pixel_read.c | 3 +- src/mesa/drivers/dri/i965/intel_tex_image.c | 3 +- src/mesa/drivers/dri/i965/intel_tex_subimage.c | 3 +- src/mesa/drivers/dri/i965/intel_tiled_memcpy.c | 140 ++++++++++++++++++------- src/mesa/drivers/dri/i965/intel_tiled_memcpy.h | 15 ++- 5 files changed, 122 insertions(+), 42 deletions(-) diff --git a/src/mesa/drivers/dri/i965/intel_pixel_read.c b/src/mesa/drivers/dri/i965/intel_pixel_read.c index df22a63..0972121 100644 --- a/src/mesa/drivers/dri/i965/intel_pixel_read.c +++ b/src/mesa/drivers/dri/i965/intel_pixel_read.c @@ -139,7 +139,8 @@ intel_readpixels_tiled_memcpy(struct gl_context * ctx, rb->Format == MESA_FORMAT_R8G8B8X8_UNORM) return false; - if (!intel_get_memcpy(rb->Format, format, type, &mem_copy, &cpp)) + if (!intel_get_memcpy(rb->Format, format, type, &mem_copy, &cpp, + INTEL_DOWNLOAD)) return false; if (!irb->mt || diff --git a/src/mesa/drivers/dri/i965/intel_tex_image.c b/src/mesa/drivers/dri/i965/intel_tex_image.c index da42fdd..00bf9ce 100644 --- a/src/mesa/drivers/dri/i965/intel_tex_image.c +++ b/src/mesa/drivers/dri/i965/intel_tex_image.c @@ -408,7 +408,8 @@ intel_gettexsubimage_tiled_memcpy(struct gl_context *ctx, texImage->TexFormat == MESA_FORMAT_R8G8B8X8_UNORM) return false; - if (!intel_get_memcpy(texImage->TexFormat, format, type, &mem_copy, &cpp)) + if (!intel_get_memcpy(texImage->TexFormat, format, type, &mem_copy, &cpp, + INTEL_DOWNLOAD)) return false; /* If this is a nontrivial texture view, let another path handle it instead. */ diff --git a/src/mesa/drivers/dri/i965/intel_tex_subimage.c b/src/mesa/drivers/dri/i965/intel_tex_subimage.c index 4262f71..909ff25 100644 --- a/src/mesa/drivers/dri/i965/intel_tex_subimage.c +++ b/src/mesa/drivers/dri/i965/intel_tex_subimage.c @@ -118,7 +118,8 @@ intel_texsubimage_tiled_memcpy(struct gl_context * ctx, packing->Invert) return false; - if (!intel_get_memcpy(texImage->TexFormat, format, type, &mem_copy, &cpp)) + if (!intel_get_memcpy(texImage->TexFormat, format, type, &mem_copy, &cpp, + INTEL_UPLOAD)) return false; /* If this is a nontrivial texture view, let another path handle it instead. */ diff --git a/src/mesa/drivers/dri/i965/intel_tiled_memcpy.c b/src/mesa/drivers/dri/i965/intel_tiled_memcpy.c index f2b35cb..dcf0462 100644 <div><div class="h5">--- a/src/mesa/drivers/dri/i965/intel_tiled_memcpy.c +++ b/src/mesa/drivers/dri/i965/intel_tiled_memcpy.c @@ -60,42 +60,79 @@ static const uint32_t ytile_span = 16; static const uint8_t rgba8_permutation[16] = { 2,1,0,3, 6,5,4,7, 10,9,8,11, 14,13,12,15 }; -/* NOTE: dst must be 16 byte aligned */ -#define rgba8_copy_16(dst, src) \ - *(__m128i *)(dst) = _mm_shuffle_epi8( \ - (__m128i) _mm_loadu_ps((float *)(src)), \ - *(__m128i *) rgba8_permutation \ - ) +/* NOTE: dst must be 16-byte aligned. src may be unaligned. */ +#define rgba8_copy_16_aligned_dst(dst, src) \ + _mm_store_si128((__m128i *)(dst), \ + _mm_shuffle_epi8(_mm_loadu_si128((__m128i *)(src)), \ + *(__m128i *) rgba8_permutation)) + +/* NOTE: src must be 16-byte aligned. dst may be unaligned. */ +#define rgba8_copy_16_aligned_src(dst, src) \ + _mm_storeu_si128((__m128i *)(dst), \ + _mm_shuffle_epi8(_mm_load_si128((__m128i *)(src)), \ + *(__m128i *) rgba8_permutation)) #endif /** - * Copy RGBA to BGRA - swap R and B. + * Copy RGBA to BGRA - swap R and B, with the destination 16-byte aligned. */ static inline void * -rgba8_copy(void *dst, const void *src, size_t bytes) +rgba8_copy_aligned_dst(void *dst, const void *src, size_t bytes) { uint8_t *d = dst; uint8_t const *s = src; #ifdef __SSSE3__ - /* Fast copying for tile spans. - * - * As long as the destination texture is 16 aligned, - * any 16 or 64 spans we get here should also be 16 aligned. - */ - if (bytes == 16) { assert(!(((uintptr_t)dst) & 0xf)); - rgba8_copy_16(d+ 0, s+ 0); + rgba8_copy_16_aligned_dst(d+ 0, s+ 0); return dst; } if (bytes == 64) { assert(!(((uintptr_t)dst) & 0xf)); - rgba8_copy_16(d+ 0, s+ 0); - rgba8_copy_16(d+16, s+16); - rgba8_copy_16(d+32, s+32); - rgba8_copy_16(d+48, s+48); + rgba8_copy_16_aligned_dst(d+ 0, s+ 0); + rgba8_copy_16_aligned_dst(d+16, s+16); + rgba8_copy_16_aligned_dst(d+32, s+32); + rgba8_copy_16_aligned_dst(d+48, s+48); + return dst; + } +#endif + + while (bytes >= 4) { + d[0] = s[2]; + d[1] = s[1]; + d[2] = s[0]; + d[3] = s[3]; + d += 4; + s += 4; + bytes -= 4; + } + return dst; +} + +/** + * Copy RGBA to BGRA - swap R and B, with the source 16-byte aligned. + */ +static inline void * +rgba8_copy_aligned_src(void *dst, const void *src, size_t bytes) +{ + uint8_t *d = dst; + uint8_t const *s = src; + +#ifdef __SSSE3__ + if (bytes == 16) { + assert(!(((uintptr_t)src) & 0xf)); + rgba8_copy_16_aligned_src(d+ 0, s+ 0); + return dst; + } + + if (bytes == 64) { + assert(!(((uintptr_t)src) & 0xf)); + rgba8_copy_16_aligned_src(d+ 0, s+ 0); + rgba8_copy_16_aligned_src(d+16, s+16); + rgba8_copy_16_aligned_src(d+32, s+32); + rgba8_copy_16_aligned_src(d+48, s+48); return dst; } #endif </div></div>@@ -357,16 +394,22 @@ linear_to_xtiled_faster(uint32_t x0, uint32_t x1, uint32_t x2, uint32_t x3, if (mem_copy == memcpy) return linear_to_xtiled(0, 0, xtile_width, xtile_width, 0, xtile_height, dst, src, src_pitch, swizzle_bit, memcpy); - else if (mem_copy == rgba8_copy) + else if (mem_copy == rgba8_copy_aligned_dst) return linear_to_xtiled(0, 0, xtile_width, xtile_width, 0, xtile_height, - dst, src, src_pitch, swizzle_bit, rgba8_copy); + dst, src, src_pitch, swizzle_bit, + rgba8_copy_aligned_dst); + else + unreachable("not reached"); </blockquote><div> </div><div>I'm not sure what I think about the unreachable here. The original versions of the *_faster functions would work regardless of what memcpy you passed in. The *_faster was only to force the compiler to inline the universe in the case where we knew some parameters. Adding unreachable changes that. That said, we never call them with anything other than memcpy or rgba8_copy_aligned_* so it's probably ok. I think I'm ok with it with or without the unreachable()'s. </div><div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"> } else { if (mem_copy == memcpy) return linear_to_xtiled(x0, x1, x2, x3, y0, y1, dst, src, src_pitch, swizzle_bit, memcpy); - else if (mem_copy == rgba8_copy) + else if (mem_copy == rgba8_copy_aligned_dst) return linear_to_xtiled(x0, x1, x2, x3, y0, y1, - dst, src, src_pitch, swizzle_bit, rgba8_copy); + dst, src, src_pitch, swizzle_bit, + rgba8_copy_aligned_dst); + else + unreachable("not reached"); } linear_to_xtiled(x0, x1, x2, x3, y0, y1, dst, src, src_pitch, swizzle_bit, mem_copy); @@ -393,16 +436,22 @@ linear_to_ytiled_faster(uint32_t x0, uint32_t x1, uint32_t x2, uint32_t x3, if (mem_copy == memcpy) return linear_to_ytiled(0, 0, ytile_width, ytile_width, 0, ytile_height, dst, src, src_pitch, swizzle_bit, memcpy); - else if (mem_copy == rgba8_copy) + else if (mem_copy == rgba8_copy_aligned_dst) return linear_to_ytiled(0, 0, ytile_width, ytile_width, 0, ytile_height, - dst, src, src_pitch, swizzle_bit, rgba8_copy); + dst, src, src_pitch, swizzle_bit, + rgba8_copy_aligned_dst); + else + unreachable("not reached"); } else { if (mem_copy == memcpy) return linear_to_ytiled(x0, x1, x2, x3, y0, y1, dst, src, src_pitch, swizzle_bit, memcpy); - else if (mem_copy == rgba8_copy) + else if (mem_copy == rgba8_copy_aligned_dst) return linear_to_ytiled(x0, x1, x2, x3, y0, y1, - dst, src, src_pitch, swizzle_bit, rgba8_copy); + dst, src, src_pitch, swizzle_bit, + rgba8_copy_aligned_dst); + else + unreachable("not reached"); } linear_to_ytiled(x0, x1, x2, x3, y0, y1, dst, src, src_pitch, swizzle_bit, mem_copy); @@ -429,16 +478,22 @@ xtiled_to_linear_faster(uint32_t x0, uint32_t x1, uint32_t x2, uint32_t x3, if (mem_copy == memcpy) return xtiled_to_linear(0, 0, xtile_width, xtile_width, 0, xtile_height, dst, src, dst_pitch, swizzle_bit, memcpy); - else if (mem_copy == rgba8_copy) + else if (mem_copy == rgba8_copy_aligned_src) return xtiled_to_linear(0, 0, xtile_width, xtile_width, 0, xtile_height, - dst, src, dst_pitch, swizzle_bit, rgba8_copy); + dst, src, dst_pitch, swizzle_bit, + rgba8_copy_aligned_src); + else + unreachable("not reached"); } else { if (mem_copy == memcpy) return xtiled_to_linear(x0, x1, x2, x3, y0, y1, dst, src, dst_pitch, swizzle_bit, memcpy); - else if (mem_copy == rgba8_copy) + else if (mem_copy == rgba8_copy_aligned_src) return xtiled_to_linear(x0, x1, x2, x3, y0, y1, - dst, src, dst_pitch, swizzle_bit, rgba8_copy); + dst, src, dst_pitch, swizzle_bit, + rgba8_copy_aligned_src); + else + unreachable("not reached"); } xtiled_to_linear(x0, x1, x2, x3, y0, y1, dst, src, dst_pitch, swizzle_bit, mem_copy); @@ -465,16 +520,22 @@ ytiled_to_linear_faster(uint32_t x0, uint32_t x1, uint32_t x2, uint32_t x3, if (mem_copy == memcpy) return ytiled_to_linear(0, 0, ytile_width, ytile_width, 0, ytile_height, dst, src, dst_pitch, swizzle_bit, memcpy); - else if (mem_copy == rgba8_copy) + else if (mem_copy == rgba8_copy_aligned_src) return ytiled_to_linear(0, 0, ytile_width, ytile_width, 0, ytile_height, - dst, src, dst_pitch, swizzle_bit, rgba8_copy); + dst, src, dst_pitch, swizzle_bit, + rgba8_copy_aligned_src); + else + unreachable("not reached"); } else { if (mem_copy == memcpy) return ytiled_to_linear(x0, x1, x2, x3, y0, y1, dst, src, dst_pitch, swizzle_bit, memcpy); - else if (mem_copy == rgba8_copy) + else if (mem_copy == rgba8_copy_aligned_src) return ytiled_to_linear(x0, x1, x2, x3, y0, y1, - dst, src, dst_pitch, swizzle_bit, rgba8_copy); + dst, src, dst_pitch, swizzle_bit, + rgba8_copy_aligned_src); + else + unreachable("not reached"); } ytiled_to_linear(x0, x1, x2, x3, y0, y1, dst, src, dst_pitch, swizzle_bit, mem_copy); @@ -684,7 +745,8 @@ tiled_to_linear(uint32_t xt1, uint32_t xt2, * \return true if the format and type combination are valid */ bool intel_get_memcpy(mesa_format tiledFormat, GLenum format, - GLenum type, mem_copy_fn* mem_copy, uint32_t* cpp) + GLenum type, mem_copy_fn *mem_copy, uint32_t *cpp, + enum intel_memcpy_direction direction) { if (type == GL_UNSIGNED_INT_8_8_8_8_REV && !(format == GL_RGBA || format == GL_BGRA)) @@ -700,7 +762,8 @@ bool intel_get_memcpy(mesa_format tiledFormat, GLenum format, if (format == GL_BGRA) { *mem_copy = memcpy; } else if (format == GL_RGBA) { - *mem_copy = rgba8_copy; + *mem_copy = direction == INTEL_UPLOAD ? rgba8_copy_aligned_dst + : rgba8_copy_aligned_src; } } else if ((tiledFormat == MESA_FORMAT_R8G8B8A8_UNORM) || (tiledFormat == MESA_FORMAT_R8G8B8X8_UNORM)) { @@ -709,7 +772,8 @@ bool intel_get_memcpy(mesa_format tiledFormat, GLenum format, <div class="HOEnZb"><div class="h5"> /* Copying from RGBA to BGRA is the same as BGRA to RGBA so we can * use the same function. */ - *mem_copy = rgba8_copy; + *mem_copy = direction == INTEL_UPLOAD ? rgba8_copy_aligned_dst + : rgba8_copy_aligned_src; } else if (format == GL_RGBA) { *mem_copy = memcpy; } diff --git a/src/mesa/drivers/dri/i965/intel_tiled_memcpy.h b/src/mesa/drivers/dri/i965/intel_tiled_memcpy.h index 3ff0d71..9dc1088 100644 --- a/src/mesa/drivers/dri/i965/intel_tiled_memcpy.h +++ b/src/mesa/drivers/dri/i965/intel_tiled_memcpy.h @@ -55,7 +55,20 @@ tiled_to_linear(uint32_t xt1, uint32_t xt2, uint32_t tiling, mem_copy_fn mem_copy); +/* Tells intel_get_memcpy() whether the memcpy() is + * + * - an upload to the GPU with an aligned destination and a potentially + * unaligned source; or + * - a download from the GPU with an aligned source and a potentially + * unaligned destination. + */ +enum intel_memcpy_direction { + INTEL_UPLOAD, + INTEL_DOWNLOAD +}; + bool intel_get_memcpy(mesa_format tiledFormat, GLenum format, - GLenum type, mem_copy_fn* mem_copy, uint32_t* cpp); + GLenum type, mem_copy_fn *mem_copy, uint32_t *cpp, + enum intel_memcpy_direction direction); #endif /* INTEL_TILED_MEMCPY */ -- 2.0.5 </div></div></blockquote></div> </div></div></div>