x86: Rewrite SGR AVX2 asm

The previous implementation did multiple passes in the horizontal
and vertical directions, with the intermediate values being stored
in buffers on the stack. This caused bad cache thrashing.

By interleaving the all the different passes in combination with a
ring buffer for storing only a few rows at a time the performance
is improved by a significant amount.

Also slightly speed up neighbor calculations by packing the a and b
values into a single 32-bit unsigned integer which allows calculations
on both values simultaneously.

src/x86/looprestoration_init_tmpl.c

@@ -43,6 +43,24 @@ void dav1d_wiener_filter5_##ext(pixel *dst, ptrdiff_t dst_stride, \
                                 enum LrEdgeFlags edges);
 
 #define SGR_FILTER(ext) \
+void dav1d_sgr_filter_5x5_##ext(pixel *dst, ptrdiff_t dst_stride, \
+                                const pixel (*left)[4], const pixel *lpf, \
+                                ptrdiff_t lpf_stride, int w, int h, \
+                                const LooprestorationParams *params, \
+                                enum LrEdgeFlags edges); \
+void dav1d_sgr_filter_3x3_##ext(pixel *dst, ptrdiff_t dst_stride, \
+                                const pixel (*left)[4], const pixel *lpf, \
+                                ptrdiff_t lpf_stride, int w, int h, \
+                                const LooprestorationParams *params, \
+                                enum LrEdgeFlags edges); \
+void dav1d_sgr_filter_mix_##ext(pixel *dst, ptrdiff_t dst_stride, \
+                                const pixel (*left)[4], const pixel *lpf, \
+                                ptrdiff_t lpf_stride, int w, int h, \
+                                const LooprestorationParams *params, \
+                                enum LrEdgeFlags edges);
+
+/* FIXME: Replace with a port of the AVX2 code */
+#define SGR_FILTER_OLD(ext) \
 void dav1d_sgr_box3_h_##ext(int32_t *sumsq, int16_t *sum, \
                             const pixel (*left)[4], \
                             const pixel *src, const ptrdiff_t stride, \
@@ -181,7 +199,7 @@ static void sgr_filter_mix_##ext(pixel *const dst, const ptrdiff_t dst_stride, \
 #if BITDEPTH == 8
 WIENER_FILTER(sse2)
 WIENER_FILTER(ssse3)
-SGR_FILTER(ssse3)
+SGR_FILTER_OLD(ssse3)
 # if ARCH_X86_64
 WIENER_FILTER(avx2)
 SGR_FILTER(avx2)
@@ -210,8 +228,8 @@ COLD void bitfn(dav1d_loop_restoration_dsp_init_x86)(Dav1dLoopRestorationDSPCont
 #if BITDEPTH == 8 && ARCH_X86_64
     c->wiener[0] = dav1d_wiener_filter7_avx2;
     c->wiener[1] = dav1d_wiener_filter5_avx2;
-    c->sgr[0] = sgr_filter_5x5_avx2;
-    c->sgr[1] = sgr_filter_3x3_avx2;
-    c->sgr[2] = sgr_filter_mix_avx2;
+    c->sgr[0] = dav1d_sgr_filter_5x5_avx2;
+    c->sgr[1] = dav1d_sgr_filter_3x3_avx2;
+    c->sgr[2] = dav1d_sgr_filter_mix_avx2;
 #endif
 }