Commit 15d4947850 for ffmpeg

commit 15d4947850b0fb0630a295deb02e9424f22aebdb
Author: Zuxy Meng <zuxy.meng@gmail.com>
Date:   Mon Nov 2 21:09:34 2026 -0800

    avcodec/x86/vp9: convert h_4x4_16 predictor MMXEXT to SSE2

    Convert the 16-bit horizontal 4x4 predictor from MMXEXT to
    XMM.

    Benchmarks (checkasm, Zen 4, pinned, median of 3
    interleaved pairs, cycles):
     vp9_hor_4x4_10bpp: 0.8 -> 0.4
     vp9_hor_4x4_12bpp: 0.8 -> 0.4

    checkasm --test=vp9dsp passes (1272/1272 over 3 seeds).

    Signed-off-by: Zuxy Meng <zuxy.meng@gmail.com>

diff --git a/libavcodec/x86/vp9dsp_init_16bpp.c b/libavcodec/x86/vp9dsp_init_16bpp.c
index 0c290e0f4b..18c7006eb7 100644
--- a/libavcodec/x86/vp9dsp_init_16bpp.c
+++ b/libavcodec/x86/vp9dsp_init_16bpp.c
@@ -44,7 +44,7 @@ decl_fpel_func(avg,  64, _16, avx2);
 decl_fpel_func(avg, 128, _16, avx2);

 decl_ipred_fns(v,       16, sse,    sse);
-decl_ipred_fns(h,       16, mmxext, sse2);
+decl_ipred_fns(h,       16, sse2,   sse2);
 decl_ipred_fns(dc,      16, mmxext, sse2);
 decl_ipred_fns(dc_top,  16, mmxext, sse2);
 decl_ipred_fns(dc_left, 16, mmxext, sse2);
@@ -72,7 +72,6 @@ av_cold void ff_vp9dsp_init_16bpp_x86(VP9DSPContext *dsp)
     int cpu_flags = av_get_cpu_flags();

     if (EXTERNAL_MMXEXT(cpu_flags)) {
-        init_ipred_func(h, HOR, 4, 16, mmxext);
         init_ipred_func(dc, DC, 4, 16, mmxext);
         init_ipred_func(dc_top,  TOP_DC,  4, 16, mmxext);
         init_ipred_func(dc_left, LEFT_DC, 4, 16, mmxext);
@@ -94,7 +93,7 @@ av_cold void ff_vp9dsp_init_16bpp_x86(VP9DSPContext *dsp)
         init_fpel_func(2, 1,  32, avg, _16, sse2);
         init_fpel_func(1, 1,  64, avg, _16, sse2);
         init_fpel_func(0, 1, 128, avg, _16, sse2);
-        init_8_16_32_ipred_funcs(h, HOR, 16, sse2);
+        init_ipred_funcs(h, HOR, 16, sse2);
         init_8_16_32_ipred_funcs(dc, DC, 16, sse2);
         init_8_16_32_ipred_funcs(dc_top,  TOP_DC,  16, sse2);
         init_8_16_32_ipred_funcs(dc_left, LEFT_DC, 16, sse2);
diff --git a/libavcodec/x86/vp9intrapred_16bpp.asm b/libavcodec/x86/vp9intrapred_16bpp.asm
index 9ba20a986a..91eb181aa9 100644
--- a/libavcodec/x86/vp9intrapred_16bpp.asm
+++ b/libavcodec/x86/vp9intrapred_16bpp.asm
@@ -157,19 +157,19 @@ cglobal vp9_ipred_v_32x32_16, 2, 4, 4, dst, stride, l, a
     jg .loop
     RET

-INIT_MMX mmxext
+INIT_XMM sse2
 cglobal vp9_ipred_h_4x4_16, 3, 3, 4, dst, stride, l, a
-    mova                    m3, [lq]
+    movq                    m3, [lq]
     DEFINE_ARGS dst, stride, stride3
     lea               stride3q, [strideq*3]
-    pshufw                  m0, m3, q3333
-    pshufw                  m1, m3, q2222
-    pshufw                  m2, m3, q1111
-    pshufw                  m3, m3, q0000
-    mova      [dstq+strideq*0], m0
-    mova      [dstq+strideq*1], m1
-    mova      [dstq+strideq*2], m2
-    mova      [dstq+stride3q ], m3
+    pshuflw                 m0, m3, q3333
+    pshuflw                 m1, m3, q2222
+    pshuflw                 m2, m3, q1111
+    pshuflw                 m3, m3, q0000
+    movq      [dstq+strideq*0], m0
+    movq      [dstq+strideq*1], m1
+    movq      [dstq+strideq*2], m2
+    movq      [dstq+stride3q ], m3
     RET

 INIT_XMM sse2