From 1622beeac4823110cbb1819ad06b8501097b71d9 Mon Sep 17 00:00:00 2001
From: nick black <dankamongmen@gmail.com>
Date: Sat, 22 May 2021 17:28:15 -0400
Subject: [PATCH] ncvisual_from_rgba: align suitably for ffmpeg (64B) #1675

---
 include/notcurses/notcurses.h |  4 ++--
 src/lib/visual.c              | 13 +++++++++++--
 src/media/ffmpeg.c            | 11 ++++++-----
 3 files changed, 19 insertions(+), 9 deletions(-)

diff --git a/include/notcurses/notcurses.h b/include/notcurses/notcurses.h
index 74196767d..2e402c987 100644
--- a/include/notcurses/notcurses.h
+++ b/include/notcurses/notcurses.h
@@ -2643,8 +2643,8 @@ API int ncblit_rgb_loose(const void* data, int linesize,
 // Per libav, we "store as BGRA on little-endian, and ARGB on big-endian".
 // This is an RGBA *byte-order* scheme. libav emits bytes, not words. Those
 // bytes are R-G-B-A. When read as words, on little endian this will be ABGR,
-// and on big-endian this will be RGBA. force everything to LE ABGR, a no-op on
-// and thus favoring little-endian. Take that, big-endian mafia!
+// and on big-endian this will be RGBA. force everything to LE ABGR, a no-op
+// on (and thus favoring) little-endian. Take that, big-endian mafia!
 
 // Extract the 8-bit alpha component from a pixel
 static inline unsigned
diff --git a/src/lib/visual.c b/src/lib/visual.c
index 3c6dac188..fd1732ba7 100644
--- a/src/lib/visual.c
+++ b/src/lib/visual.c
@@ -526,15 +526,24 @@ ncvisual* ncvisual_from_rgba(const void* rgba, int rows, int rowstride, int cols
   }
   ncvisual* ncv = ncvisual_create();
   if(ncv){
+    // ffmpeg needs inputs with rows aligned on 192-byte boundaries
     ncv->rowstride = rowstride;
+    #define IMGALIGN 64
+    if(ncv->rowstride % IMGALIGN){
+      ncv->rowstride = (ncv->rowstride + IMGALIGN) / IMGALIGN * IMGALIGN;
+    }
+    #undef IMGALIGN
     ncv->pixx = cols;
     ncv->pixy = rows;
-    uint32_t* data = memdup(rgba, rowstride * ncv->pixy);
-//fprintf(stderr, "COPY US %d (%d)\n", rowstride * ncv->pixy, ncv->pixy);
+    uint32_t* data = malloc(ncv->rowstride * ncv->pixy);
     if(data == NULL){
       ncvisual_destroy(ncv);
       return NULL;
     }
+    for(int y = 0 ; y < rows ; ++y){
+      memcpy(data + (ncv->rowstride * y) / 4, rgba + rowstride * y, rowstride);
+    }
+//fprintf(stderr, "COPY US %d (%d)\n", rowstride * ncv->pixy, ncv->pixy);
 //fprintf(stderr, "ROWS: %d STRIDE: %d (%d) COLS: %d\n", rows, rowstride, rowstride / 4, cols);
     ncvisual_set_data(ncv, data, true);
     ncvisual_details_seed(ncv);
diff --git a/src/media/ffmpeg.c b/src/media/ffmpeg.c
index 034509170..663bc3c5c 100644
--- a/src/media/ffmpeg.c
+++ b/src/media/ffmpeg.c
@@ -42,8 +42,8 @@ typedef struct ncvisual_details {
 print_frame_summary(const AVCodecContext* cctx, const AVFrame* f){
   char pfmt[128];
   av_get_pix_fmt_string(pfmt, sizeof(pfmt), f->format);
-  fprintf(stderr, "Frame %05d (%d? %d?) %dx%d pfmt %d (%s)\n",
-          cctx->frame_number,
+  fprintf(stderr, "Frame %05d %p (%d? %d?) %dx%d pfmt %d (%s)\n",
+          cctx ? cctx->frame_number : 0, f,
           f->coded_picture_number,
           f->display_picture_number,
           f->width, f->height,
@@ -455,12 +455,12 @@ int ffmpeg_decode_loop(ncvisual* ncv){
 int ffmpeg_blit(ncvisual* ncv, int rows, int cols, ncplane* n,
                 const struct blitset* bset, const blitterargs* bargs){
   const AVFrame* inframe = ncv->details->frame;
-//fprintf(stderr, "inframe: %p frame: %p\n", inframe, ncv->details->frame);
+//print_frame_summary(NULL, inframe);
   void* data = NULL;
   int stride = 0;
   AVFrame* sframe = NULL;
   const int targformat = AV_PIX_FMT_RGBA;
-//fprintf(stderr, "got format: %d want format: %d\n", inframe->format, targformat);
+//fprintf(stderr, "got format: %d (%d/%d) want format: %d\n", inframe->format, inframe->height, inframe->width, targformat);
   if(inframe && (cols != inframe->width || rows != inframe->height || inframe->format != targformat)){
 //fprintf(stderr, "resize+render: %d/%d->%d/%d\n", inframe->height, inframe->width, rows, cols);
     sframe = av_frame_alloc();
@@ -471,6 +471,7 @@ int ffmpeg_blit(ncvisual* ncv, int rows, int cols, ncplane* n,
 //fprintf(stderr, "WHN NCV: %d/%d bargslen: %d/%d\n", inframe->width, inframe->height, bargs->leny, bargs->lenx);
     const int srclenx = bargs->lenx ? bargs->lenx : inframe->width;
     const int srcleny = bargs->leny ? bargs->leny : inframe->height;
+//fprintf(stderr, "src %d/%d -> targ %d/%d ctx: %p\n", srcleny, srclenx, rows, cols, ncv->details->swsctx);
     ncv->details->swsctx = sws_getCachedContext(ncv->details->swsctx,
                                                 srclenx, srcleny,
                                                 inframe->format,
@@ -492,7 +493,7 @@ int ffmpeg_blit(ncvisual* ncv, int rows, int cols, ncplane* n,
 //fprintf(stderr, "Error allocating visual data (%d X %d)\n", sframe->height, sframe->width);
       return -1;
     }
-//fprintf(stderr, "INFRAME DAA: %p SDATA: %p FDATA: %p\n", inframe->data[0], sframe->data[0], ncv->details->frame->data[0]);
+//fprintf(stderr, "INFRAME DAA: %p SDATA: %p FDATA: %p to %d/%d\n", inframe->data[0], sframe->data[0], ncv->details->frame->data[0], sframe->height, sframe->width);
     int height = sws_scale(ncv->details->swsctx, (const uint8_t* const*)inframe->data,
                            inframe->linesize, 0, srcleny, sframe->data,
                            sframe->linesize);