Eric wing
Hi, I think I found a bug when using SDL_WINDOW_ALLOW_HIGHDPI with SDL_RenderSetLogicalSize on iOS. I use SDL_RenderSetLogicalSize for all my stuff. I just tried turning on SDL_WINDOW_ALLOW_HIGHDPI on iOS and suddenly all my touch/mouse positions are really broken/far-off-the-mark.
I actually don't have a real retina device (still) so I'm seeing this using the iOS simulator with a 6plus template.
Attached is a simple test program that can reproduce the problem. It uses RenderSetLogicalSize and draws some moving happy faces (to show the boundaries/space of the LogicalSize and that it is working correctly for that part).
When you click/touch, it will draw one more happy face where your button point is.
If you comment out SDL_WINDOW_ALLOW_HIGHDPI, everything works as expected. But if you compile with it in, the mouse coordinates seem really far off the mark. (Face appears far up and to the left.)
Alex Szpakowski on the mailing list suggests the problem is
"I believe this is a bug in SDL_Render?s platform-agnostic mouse coordinate scaling code. It assumes the units of the mouse coordinates are always in pixels, which isn?t the case where high-DPI is involved (regardless of whether iOS is used) ? they?re actually in ?DPI independent? coordinates (which matches the window size, but not the renderer output size)."
Additionally, if this is correct, the Mac under Retina is also probably affected too and "as well as any other platform SDL adds high-dpi support for in the future".
felix
The functions in src/render/SDL_yuv_mmx.c contain the following inline assembly snippet:
/* tap dance to workaround the inability to use %%ebx at will... */
/* move one thing to the stack... */
"pushl $0\n" /* save a slot on the stack. */
"pushl %%ebx\n" /* save %%ebx. */
"movl %0, %%ebx\n" /* put the thing in ebx. */
"movl %%ebx,4(%%esp)\n" /* put the thing in the stack slot. */
"popl %%ebx\n" /* get back %%ebx (the PIC register). */
Here's how it ended up in a binary on my old laptop:
0xb5c17dbd <ColorRGBDitherYV12MMX1X+93>: push $0x0
0xb5c17dbf <ColorRGBDitherYV12MMX1X+95>: push %ebx
0xb5c17dc0 <ColorRGBDitherYV12MMX1X+96>: mov 0xc(%esp),%ebx
0xb5c17dc4 <ColorRGBDitherYV12MMX1X+100>: mov %ebx,0x4(%esp)
0xb5c17dc8 <ColorRGBDitherYV12MMX1X+104>: pop %ebx
Apparently the compiler, oblivious to the fact that the assembly snippet manipulates the %esp register, decided to refer to the operand via that same register instead of via %ebp (I believe -fomit-frame-pointer enables this). This causes %ebx to be loaded with the wrong value, which later leads to a null pointer dereference.
Recent GCC can use the %ebx register normally: <https://gcc.gnu.org/bugzilla/show_bug.cgi?id=47602#c16>. There is even an explicit constraint "b" for allocating it.
Tom Seddon
GL_ActivateRenderer may call GL_UpdateViewport, which leaves the GL_PROJECTION matrix selected. But after GL_ResetState, the GL_MODELVIEW matrix is selected, suggesting that's the intended default state.
It seems at least like these should be consistent. Presumably GL_UpdateViewport should be doing a glMatrixMode(GL_MODELVIEW) before it finishes.
Richard Russell
Resuming from a suspended state results in a black screen. This only happens when using GLES 1.1 (GLES 2 resumes correctly) and when the render target has been changed using SDL_SetRenderTarget. This problem is new in 2.0.4.
The attached test case demonstrates the issue.
Sylvain Becker has apparently found a fix as follows:
"In the opengles leaf function (in 'src/render/opengles/SDL_render_gles.c'), it appears there is a call to 'GLES_ActivateRenderer' in 'GLES_SetRenderTarget', which is not present in opengles2. When commenting out this 'GLES_ActivateRenderer', it seems to resume fine".
This appears to fix the testcase perfectly, but I don't know whether it could have any undesirable side-effects.
felix
Here's a snippet of SDL_DestroyRenderer from hg revision 10746:7540ff5d0e0e:
SDL_Texture *texture = NULL;
SDL_Texture *nexttexture = NULL;
/* ... */
for (texture = renderer->textures; texture; texture = nexttexture) {
nexttexture = texture->next;
SDL_DestroyTexture(texture);
}
SDL_DestroyTexture removes the texture from the linked list pointed to by the renderer and ends up calling SDL_DestroyTextureInternal, which contains this:
if (texture->native) {
SDL_DestroyTexture(texture->native);
}
If it happens that texture->native is an alias of nexttexture two stack frames up, SDL_DestroyRenderer will end up trying to destroy an already freed texture. I've had this very situation happen in dosemu2.
Bug introduced in revision 10650:a8253d439914, which has a somewhat ironic description of "Fixed all known static analysis bugs"...
Simon Hug
The software renderer produces incorrect results when blending textures at an angle with certain blend modes. It seems that there were some edge cases that weren't considered when the SW_RenderCopyEx function was last changed. Or another bug possibly covered up the problem. (More on that in another bug report.)
Most of the issues come from the fact that the rotating function sets a black colorkey. This is problematic because black is most likely appearing in the surface and the final blit will ignore these pixels. Unless a colorkey is already set (the software renderer currently never sets one), it's very hard to find a free color. Of course it could scan over the whole image until one is found, but that seems inefficient.
The following blend modes have issues when drawn at an angle.
NONE: The black pixels get ignored, making them essentially transparent. This breaks the 'dstRGBA = srcRGBA' definition of the NONE blend mode.
MOD: Again, the black pixels get ignored. This also breaks the 'dstRGB = dstRGB * srcRGB' definition of the MOD blend mode, where black pixels would make the destination black as well. A white colorkey will work though, with some preparations.
BLEND: There are some issues when blending a texture with a translucent RGBA target texture. I - uh - forgot what the problem here exactly is.
This patch fixes the issues mentioned above. It mainly changes the code so it tries to do things without the colorkey and removes the automatic format conversion part from the SDLgfx_rotateSurface function. Getting the format right is something the caller has to do now and the required code has been added to the SW_RenderCopyEx function.
There's a small change to the SW_CreateTexture function. RLE encoding a surface with an alpha mask can be a lossy process. Depending on how the user uses the RGBA channels, this may be undesired. The change that surfaces with an alpha mask don't get encoded makes the software renderer consistent with the other renderers.
The SW_RenderCopyEx function now does these steps: Lock the source surface if necessary. Create a clone of the source by using the pixel buffer directly. Check the format and set a flag if a conversion is necessary. Check if scaling or cropping is necessary and set the flag for that as well. Check if color and alpha modulation has to be done before the rotate. Check if the source is an opaque surface. If not, it creates a mask surface that is necessary for the NONE blend mode. If any of the flags were set, a new surface is created and the source will be converted, scaled, cropped, and modulated. The rest of the function stays somewhat the same. The mask also needs to be rotated of course and then there is the NONE blend mode...
It's surprisingly hard to get the pixel from a rotated surface to the destination buffer without affecting the pixel outside the rotated area. I found a way to do this with three blits which is pretty hard on the performance. Perhaps someone has an idea how to do this faster?
As mentioned above, the SDLgfx_rotateSurface now only takes 8-bit paletted or 32-bit with alpha mask surfaces. It additionally sets the new surfaces up for the MOD blend mode.
I shortly tested the 8-bit path of SDLgfx_rotateSurface and it seemed to work so far. This path is not used by the software renderer anyway.
Sylvain 2016-11-07 08:49:34 UTC
when rotated +90 or -90, some transparent lines appears, though there is no Alpha or ColorKey.
if you set a dummy colorkey, it will remove the line ...
if you set a some alpha mod, the +90/-90 get transparent but not the 0/180 ...
Call SDL_GL_GetDrawableSize() directly because we may be in the initialization path and SDL_GetRendererOutputSize() will fail because the renderer magic isn't set up yet.
Daniel Gibson
Ok, I followed the simple approach of just making SDL_PIXELFORMAT_RGBA32 an alias of SDL_PIXELFORMAT_RGBA8888/SDL_PIXELFORMAT_ABGR8888, depending on endianess. And I did the same for SDL_PIXELFORMAT_ARGB32, .._BGRA, .._ABGR.
SDL_GetPixelFormatName() will of course return SDL_PIXELFORMAT_RGBA8888 (or SDL_PIXELFORMAT_ABGR8888) instead of SDL_PIXELFORMAT_RGBA32, but as long as that's mentioned in the docs it shouldn't be a problem.
D39 does not support negative viewport values which the current implementation relies on.
D3D11 does support negative viewport values so that will continue working.
Refer to Bug 2799.
Adam M.
When doing a rotated texture copy with the software renderer, where the angle is a multiple of 90 degrees, one or two edges of the image get cut off. This is because of the following line in sw_rotate.c:
if ((unsigned)dx < (unsigned)sw && (unsigned)dy < (unsigned)sh) {
which is effectively saying:
if (dx >= 0 && dx < src->w-1 && dy >= 0 && dy < src->h-1) {
As a result, it doesn't process pixels in the right column or bottom row of the source image (except when they're accessed as part of the bilinear filtering for nearby pixels). This causes it to look like the edges are cut off, and it's especially obvious with an exact multiple of 90 degrees.
Nitz
In function SDLgfx_rotateSurface:
rz_dst =
SDL_CreateRGBSurface(SDL_SWSURFACE, dstwidth, dstheight + GUARD_ROWS,
rz_src->format->Rmask, rz_src->format->Gmask,
rz_src->format->Bmask, rz_src->format->Amask);
Here rz_src get De-referenced without NULL check, which is risky.
Daniel
Seems like check of the visibility of renderer (renderer->hidden) is missing in SDL_RenderCopyEx.
In SDL_RenderCopy it should be done much earlier (after checking support for RenderCopyEx, line 1750).
Yann Dirson
When attempting to force use of opengles2 renderer with:
int wanted_renderer = -1;
for (int i = 0; i < numrenderers; i++) {
SDL_RendererInfo renderer_info;
if (SDL_GetRenderDriverInfo(i, &renderer_info) != 0) {
SDL_LogError(SDL_LOG_CATEGORY_APPLICATION, "Couldn't get renderer driver info: %s\n",
SDL_GetError());
quit(2);
}
std::cerr << "Renderer " << i << " '" << renderer_info.name << "': flags=0x"
<< std::hex << renderer_info.flags << std::dec
<< ", " << renderer_info.num_texture_formats << " texture formats, max="
<< renderer_info.max_texture_width << "x"
<< renderer_info.max_texture_height << "\n";
if (!strcmp(renderer_info.name, "opengles2")) {
std::cerr << " selecting!\n";
wanted_renderer = i;
}
}
renderer = SDL_CreateRenderer(window, wanted_renderer, 0);
... on banana pi or raspberry pi I get an error like the following (the actual
context profile value varies, being used uninitialized)
ERROR: Couldn't create renderer: Unknown OpenGL context profile 900
With this patch I get the following, which should help more pointing to a real problem:
ERROR: Couldn't create renderer: Failed getting OpenGL glGetString entry point
I pushed a patch (based on master branch of unofficial git mirror):
550389c89f
I'll be opening a different bug for the underlying issue.
ny00
Using the OpenGL ES 1.1 renderer, after updating a texture with SDL_UpdateTexture (or SDL_UnlockTexture), a following call to SDL_RenderFillRect draws a rectangle with the wrong color (which appears to be the same as the texture's top-left pixel).
Comparing SDL_render_gles.c:GLES_UpdateTexture to SDL_render_gl.c:GL_UpdateTexture, a missing call to glDisable appears to be the cause. After adding it back, the bug is resolved.
Simon Hug
The description of the SDL_RenderClear function in the SDL_render.h header says the following:
"This function clears the entire rendering target, ignoring the viewport."
The word "entire" implies that the clipping rectangle set with SDL_RenderSetClipRect also gets ignored. This is left somewhat ambiguous if only the viewport is mentioned. Minor thing, but let's see what the implementations actually do.
The software renderer ignores the clipping rectangle when clearing. It even has a comment on this: /* By definition the clear ignores the clip rect */
Most other render drivers (opengl, opengles, opengles2, direct3d, and psp [I assume. Can't test it.]) use the scissor test for the ClipRect and don't disable it when clearing. Clearing will only happen within the clipping rectangle for these drivers.
An exception is direct3d11 which uses a clear function that ignores the scissor test.
