From 0871470ca8bedb899f0f142e8f515e4dd2080f38 Mon Sep 17 00:00:00 2001
From: Thomas Mundt <loudmax@yahoo.de>
Date: Mon, 13 Mar 2017 16:48:44 +0100
Subject: [PATCH 1/2] avfilter/interlace: change lowpass_line function
 prototype

Signed-off-by: Thomas Mundt <loudmax@yahoo.de>
---
 libavfilter/interlace.h              |  2 +-
 libavfilter/tinterlace.h             |  2 +-
 libavfilter/vf_interlace.c           | 17 +++++++++--------
 libavfilter/vf_tinterlace.c          | 14 ++++++++------
 libavfilter/x86/vf_interlace.asm     | 31 +++++++++++++------------------
 libavfilter/x86/vf_interlace_init.c  |  6 ++----
 libavfilter/x86/vf_tinterlace_init.c |  6 ++----
 7 files changed, 36 insertions(+), 42 deletions(-)

diff --git a/libavfilter/interlace.h b/libavfilter/interlace.h
index da073ae..107c94f 100644
--- a/libavfilter/interlace.h
+++ b/libavfilter/interlace.h
@@ -50,7 +50,7 @@ typedef struct InterlaceContext {
     int lowpass;           // enable or disable low pass filtering
     AVFrame *cur, *next;   // the two frames from which the new one is obtained
     void (*lowpass_line)(uint8_t *dstp, ptrdiff_t linesize, const uint8_t *srcp,
-                         const uint8_t *srcp_above, const uint8_t *srcp_below);
+                         ptrdiff_t mref, ptrdiff_t pref);
 } InterlaceContext;
 
 void ff_interlace_init_x86(InterlaceContext *interlace);
diff --git a/libavfilter/tinterlace.h b/libavfilter/tinterlace.h
index 3b703e7..f52af13 100644
--- a/libavfilter/tinterlace.h
+++ b/libavfilter/tinterlace.h
@@ -54,7 +54,7 @@ typedef struct {
     uint8_t *black_data[4];     ///< buffer used to fill padded lines
     int black_linesize[4];
     void (*lowpass_line)(uint8_t *dstp, ptrdiff_t width, const uint8_t *srcp,
-                         const uint8_t *srcp_above, const uint8_t *srcp_below);
+                         ptrdiff_t mref, ptrdiff_t pref);
 } TInterlaceContext;
 
 void ff_tinterlace_init_x86(TInterlaceContext *interlace);
diff --git a/libavfilter/vf_interlace.c b/libavfilter/vf_interlace.c
index efa3128..8da8326 100644
--- a/libavfilter/vf_interlace.c
+++ b/libavfilter/vf_interlace.c
@@ -55,9 +55,10 @@ AVFILTER_DEFINE_CLASS(interlace);
 
 static void lowpass_line_c(uint8_t *dstp, ptrdiff_t linesize,
                            const uint8_t *srcp,
-                           const uint8_t *srcp_above,
-                           const uint8_t *srcp_below)
+                           ptrdiff_t mref, ptrdiff_t pref)
 {
+    const uint8_t *srcp_above = srcp + mref;
+    const uint8_t *srcp_below = srcp + pref;
     int i;
     for (i = 0; i < linesize; i++) {
         // this calculation is an integer representation of
@@ -154,13 +155,13 @@ static void copy_picture_field(InterlaceContext *s,
             int srcp_linesize = src_frame->linesize[plane] * 2;
             int dstp_linesize = dst_frame->linesize[plane] * 2;
             for (j = lines; j > 0; j--) {
-                const uint8_t *srcp_above = srcp - src_frame->linesize[plane];
-                const uint8_t *srcp_below = srcp + src_frame->linesize[plane];
+                ptrdiff_t pref = src_frame->linesize[plane];
+                ptrdiff_t mref = -pref;
                 if (j == lines)
-                    srcp_above = srcp; // there is no line above
-                if (j == 1)
-                    srcp_below = srcp; // there is no line below
-                s->lowpass_line(dstp, cols, srcp, srcp_above, srcp_below);
+                    mref = 0;    // there is no line above
+                else if (j == 1)
+                    pref = 0;    // there is no line below
+                s->lowpass_line(dstp, cols, srcp, mref, pref);
                 dstp += dstp_linesize;
                 srcp += srcp_linesize;
             }
diff --git a/libavfilter/vf_tinterlace.c b/libavfilter/vf_tinterlace.c
index 80146a9..09ca4d3 100644
--- a/libavfilter/vf_tinterlace.c
+++ b/libavfilter/vf_tinterlace.c
@@ -89,8 +89,10 @@ static int query_formats(AVFilterContext *ctx)
 }
 
 static void lowpass_line_c(uint8_t *dstp, ptrdiff_t width, const uint8_t *srcp,
-                           const uint8_t *srcp_above, const uint8_t *srcp_below)
+                           ptrdiff_t mref, ptrdiff_t pref)
 {
+    const uint8_t *srcp_above = srcp + mref;
+    const uint8_t *srcp_below = srcp + pref;
     int i;
     for (i = 0; i < width; i++) {
         // this calculation is an integer representation of
@@ -228,12 +230,12 @@ void copy_picture_field(TInterlaceContext *tinterlace,
             int srcp_linesize = src_linesize[plane] * k;
             int dstp_linesize = dst_linesize[plane] * (interleave ? 2 : 1);
             for (h = lines; h > 0; h--) {
-                const uint8_t *srcp_above = srcp - src_linesize[plane];
-                const uint8_t *srcp_below = srcp + src_linesize[plane];
-                if (h == lines) srcp_above = srcp; // there is no line above
-                if (h == 1) srcp_below = srcp;     // there is no line below
+                ptrdiff_t pref = src_linesize[plane];
+                ptrdiff_t mref = -pref;
+                if (h == lines)  mref = 0; // there is no line above
+                else if (h == 1) pref = 0; // there is no line below
 
-                tinterlace->lowpass_line(dstp, cols, srcp, srcp_above, srcp_below);
+                tinterlace->lowpass_line(dstp, cols, srcp, mref, pref);
                 dstp += dstp_linesize;
                 srcp += srcp_linesize;
             }
diff --git a/libavfilter/x86/vf_interlace.asm b/libavfilter/x86/vf_interlace.asm
index f70c700..ab11519 100644
--- a/libavfilter/x86/vf_interlace.asm
+++ b/libavfilter/x86/vf_interlace.asm
@@ -28,33 +28,28 @@ SECTION_RODATA
 SECTION .text
 
 %macro LOWPASS_LINE 0
-cglobal lowpass_line, 5, 5, 7
-    add r0, r1
-    add r2, r1
-    add r3, r1
-    add r4, r1
-    neg r1
-
+cglobal lowpass_line, 5, 5, 7, dst, h, src, mref, pref
     pcmpeqb m6, m6
-
 .loop:
-    mova m0, [r3+r1]
-    mova m1, [r3+r1+mmsize]
-    pavgb m0, [r4+r1]
-    pavgb m1, [r4+r1+mmsize]
+    mova m0, [srcq+mrefq]
+    mova m1, [srcq+mrefq+mmsize]
+    pavgb m0, [srcq+prefq]
+    pavgb m1, [srcq+prefq+mmsize]
     pxor m0, m6
     pxor m1, m6
-    pxor m2, m6, [r2+r1]
-    pxor m3, m6, [r2+r1+mmsize]
+    pxor m2, m6, [srcq]
+    pxor m3, m6, [srcq+mmsize]
     pavgb m0, m2
     pavgb m1, m3
     pxor m0, m6
     pxor m1, m6
-    mova [r0+r1], m0
-    mova [r0+r1+mmsize], m1
+    mova [dstq], m0
+    mova [dstq+mmsize], m1
 
-    add r1, 2*mmsize
-    jl .loop
+    add dstq, 2*mmsize
+    add srcq, 2*mmsize
+    sub hd, 2*mmsize
+    jg .loop
 REP_RET
 %endmacro
 
diff --git a/libavfilter/x86/vf_interlace_init.c b/libavfilter/x86/vf_interlace_init.c
index 52a22f8..7d8acd6 100644
--- a/libavfilter/x86/vf_interlace_init.c
+++ b/libavfilter/x86/vf_interlace_init.c
@@ -28,12 +28,10 @@
 
 void ff_lowpass_line_sse2(uint8_t *dstp, ptrdiff_t linesize,
                           const uint8_t *srcp,
-                          const uint8_t *srcp_above,
-                          const uint8_t *srcp_below);
+                          ptrdiff_t mref, ptrdiff_t pref);
 void ff_lowpass_line_avx (uint8_t *dstp, ptrdiff_t linesize,
                           const uint8_t *srcp,
-                          const uint8_t *srcp_above,
-                          const uint8_t *srcp_below);
+                          ptrdiff_t mref, ptrdiff_t pref);
 
 av_cold void ff_interlace_init_x86(InterlaceContext *s)
 {
diff --git a/libavfilter/x86/vf_tinterlace_init.c b/libavfilter/x86/vf_tinterlace_init.c
index ddb0cce..175b5cf 100644
--- a/libavfilter/x86/vf_tinterlace_init.c
+++ b/libavfilter/x86/vf_tinterlace_init.c
@@ -29,12 +29,10 @@
 
 void ff_lowpass_line_sse2(uint8_t *dstp, ptrdiff_t linesize,
                           const uint8_t *srcp,
-                          const uint8_t *srcp_above,
-                          const uint8_t *srcp_below);
+                          ptrdiff_t mref, ptrdiff_t pref);
 void ff_lowpass_line_avx (uint8_t *dstp, ptrdiff_t linesize,
                           const uint8_t *srcp,
-                          const uint8_t *srcp_above,
-                          const uint8_t *srcp_below);
+                          ptrdiff_t mref, ptrdiff_t pref);
 
 av_cold void ff_tinterlace_init_x86(TInterlaceContext *s)
 {
-- 
2.7.4.windows.1

