vo_opengl: use glBufferSubData instead of glMapBufferRange

Performance seems pretty much unchanged but I no longer get nasty spikes
on NUMA systems, probably because glBufferSubData runs in the driver or
something.

As a simplification of the code, we also just size the PBO to always
have the full size, even for cropped textures. This seems slower but not
by relevant amounts, and only affects e.g. --vf=crop. It also slightly
increases VRAM usage for textures with big strides.

This new code path is especially nice because it no longer depends on
GL_ARB_map_buffer_range, and no longer uses any functions that can
possibly fail, thus simplifying control flow and seemingly deprecating
the manpage's claim about possible image corruption.

In theory we could also reduce NUM_PBO_BUFFERS since it doesn't seem
like we're streaming uploads anyway, but leave it in there just in
case some drivers disagree...

3 files changed, 10 insertions, 24 deletions

diff --git a/video/out/opengl/common.c b/video/out/opengl/common.c
index 6913b77433..203c14b7ef 100644
--- a/video/out/opengl/common.c
+++ b/video/out/opengl/common.c
@@ -95,6 +95,7 @@ static const struct gl_functions gl_functions[] = {
             DEF_FN(BindTexture),
             DEF_FN(BlendFuncSeparate),
             DEF_FN(BufferData),
+            DEF_FN(BufferSubData),
             DEF_FN(Clear),
             DEF_FN(ClearColor),
             DEF_FN(CompileShader),
diff --git a/video/out/opengl/common.h b/video/out/opengl/common.h
index 351624051f..c9162f2479 100644
--- a/video/out/opengl/common.h
+++ b/video/out/opengl/common.h
@@ -128,6 +128,7 @@ struct GL {
                                           GLbitfield);
     GLboolean (GLAPIENTRY *UnmapBuffer)(GLenum);
     void (GLAPIENTRY *BufferData)(GLenum, intptr_t, const GLvoid *, GLenum);
+    void (GLAPIENTRY *BufferSubData)(GLenum, GLintptr, GLsizeiptr, const GLvoid *);
     void (GLAPIENTRY *ActiveTexture)(GLenum);
     void (GLAPIENTRY *BindTexture)(GLenum, GLuint);
     int (GLAPIENTRY *SwapInterval)(int);
diff --git a/video/out/opengl/utils.c b/video/out/opengl/utils.c
index 9870936bc5..878c468175 100644
--- a/video/out/opengl/utils.c
+++ b/video/out/opengl/utils.c
@@ -1310,15 +1310,16 @@ void gl_pbo_upload_tex(struct gl_pbo_upload *pbo, GL *gl, bool use_pbo,
     assert(x >= 0 && y >= 0 && w >= 0 && h >= 0);
     assert(x + w <= tex_w && y + h <= tex_h);
 
-    if (!use_pbo || !gl->MapBufferRange)
-        goto no_pbo;
+    if (!use_pbo) {
+        gl_upload_tex(gl, target, format, type, dataptr, stride, x, y, w, h);
+        return;
+    }
 
     // We align the buffer size to 4096 to avoid possible subregion
     // dependencies. This is not a strict requirement (the spec requires no
     // alignment), but a good precaution for performance reasons
-    size_t pix_stride = gl_bytes_per_pixel(format, type);
-    size_t buffer_size = FFALIGN(pix_stride * tex_w * tex_h, 4096);
-    size_t needed_size = pix_stride * w * h;
+    size_t needed_size = stride * h;
+    size_t buffer_size = FFALIGN(needed_size, 4096);
 
     if (buffer_size != pbo->buffer_size)
         gl_pbo_upload_uninit(pbo);
@@ -1345,26 +1346,9 @@ void gl_pbo_upload_tex(struct gl_pbo_upload *pbo, GL *gl, bool use_pbo,
     pbo->index = (pbo->index + 1) % NUM_PBO_BUFFERS;
 
     gl->BindBuffer(GL_PIXEL_UNPACK_BUFFER, pbo->buffer);
-    void *data = gl->MapBufferRange(GL_PIXEL_UNPACK_BUFFER, offset, needed_size,
-                                    GL_MAP_WRITE_BIT | GL_MAP_INVALIDATE_RANGE_BIT);
-    if (!data)
-        goto no_pbo;
-
-    memcpy_pic(data, dataptr, pix_stride * w,  h, pix_stride * w, stride);
-
-    if (!gl->UnmapBuffer(GL_PIXEL_UNPACK_BUFFER)) {
-        gl->BindBuffer(GL_PIXEL_UNPACK_BUFFER, 0);
-        goto no_pbo;
-    }
-
-    gl_upload_tex(gl, target, format, type, (void *)offset, pix_stride * w, x, y, w, h);
-
+    gl->BufferSubData(GL_PIXEL_UNPACK_BUFFER, offset, needed_size, dataptr);
+    gl_upload_tex(gl, target, format, type, (void *)offset, stride, x, y, w, h);
     gl->BindBuffer(GL_PIXEL_UNPACK_BUFFER, 0);
-
-    return;
-
-no_pbo:
-    gl_upload_tex(gl, target, format, type, dataptr, stride, x, y, w, h);
 }
 
 void gl_pbo_upload_uninit(struct gl_pbo_upload *pbo)