[FFmpeg-devel] [PATCH V2 3/3] avfilter/x86/vf_eq: add SSE2 version

[Ting Fu]

Signed-off-by: Ting Fu <ting.fu at intel.com>
---
 libavfilter/x86/vf_eq.asm    | 12 ++++++++++--
 libavfilter/x86/vf_eq_init.c | 19 +++++++++++++++++++
 2 files changed, 29 insertions(+), 2 deletions(-)

diff --git a/libavfilter/x86/vf_eq.asm b/libavfilter/x86/vf_eq.asm
index 8460342896..a30a287029 100644
--- a/libavfilter/x86/vf_eq.asm
+++ b/libavfilter/x86/vf_eq.asm
@@ -24,7 +24,7 @@
 
 SECTION .text
 
-INIT_MMX mmxext
+%macro PROCESS_ONE_LINE 1
 cglobal process_one_line, 5, 7, 5, src, dst, contrast, brightness, w
     movd m3, contrastd
     movd m4, brightnessd
@@ -39,7 +39,7 @@ cglobal process_one_line, 5, 7, 5, src, dst, contrast, brightness, w
     pxor m1, m1
     mov scalard, wd
     and scalard, mmsize-1
-    sar wd, 3
+    sar wd, %1
     cmp wd, 1
     jl .loop1
 
@@ -80,3 +80,11 @@ cglobal process_one_line, 5, 7, 5, src, dst, contrast, brightness, w
 
     .end:
         RET
+
+%endmacro
+
+INIT_MMX mmxext
+PROCESS_ONE_LINE 3
+
+INIT_XMM sse2
+PROCESS_ONE_LINE 4
diff --git a/libavfilter/x86/vf_eq_init.c b/libavfilter/x86/vf_eq_init.c
index c18db75545..8ad9f4bcaf 100644
--- a/libavfilter/x86/vf_eq_init.c
+++ b/libavfilter/x86/vf_eq_init.c
@@ -28,6 +28,8 @@
 
 extern void ff_process_one_line_mmxext(const uint8_t *src, uint8_t *dst, short contrast,
                                        short brightness, int w);
+extern void ff_process_one_line_sse2(const uint8_t *src, uint8_t *dst, short contrast,
+                                     short brightness, int w);
 
 static void process_mmxext(EQParameters *param, uint8_t *dst, int dst_stride,
                            const uint8_t *src, int src_stride, int w, int h)
@@ -44,6 +46,20 @@ static void process_mmxext(EQParameters *param, uint8_t *dst, int dst_stride,
     emms_c();
 }
 
+static void process_sse2(EQParameters *param, uint8_t *dst, int dst_stride,
+                         const uint8_t *src, int src_stride, int w, int h)
+{
+    short contrast = (short) (param->contrast * 256 * 16);
+    short brightness = ((short) (100.0 * param->brightness + 100.0) * 511)
+                       / 200 - 128 - contrast / 32;
+
+    while (h--) {
+        ff_process_one_line_sse2(src, dst, contrast, brightness, w);
+        src += src_stride;
+        dst += dst_stride;
+    }
+}
+
 av_cold void ff_eq_init_x86(EQContext *eq)
 {
     int cpu_flags = av_get_cpu_flags();
@@ -51,5 +67,8 @@ av_cold void ff_eq_init_x86(EQContext *eq)
     if (cpu_flags & AV_CPU_FLAG_MMXEXT) {
         eq->process = process_mmxext;
     }
+    if (cpu_flags & AV_CPU_FLAG_SSE2) {
+        eq->process = process_sse2;
+    }
 }
 
-- 
2.17.1