OpenGL & Vulkan

[vindast]

I trying to implement sparse texture array.

I think I found a bug in the Amd driver, or i am doing something wrong.

On sparse array texture i get memory leaking:

Opengl memory leak on amd - YouTube

Texture format:

rgba8

Resolution:

1024 * 1024, 11 mip levels

Sourses:

Render/GLSparseTextureArray.h at master · vindast/Render · GitHub 

Render/GLSparseTextureArray.cpp at master · vindast/Render · GitHub 

Min exe, which produces this bug:

https://yadi.sk/d/VOjT0uZ8uhRdWw

System requirements: opengl 4.5, avx2

I found two scenarios:

case 1:

press button "create sparseTexture buffer"

enable "bDrawOnAllocation"

press button "create texture"

-> memory leak occur after glTexPageCommitmentARB and acces to texture in shader

case 2:

press button "create sparseTexture buffer"

press button "create texture"

press button "load texture"

-> memory leak occur after glTexPageCommitmentARB and try to update 0 mip level by glTexSubImage3D without acces to texture in shader

important information:

bug occur for layer count more then 512. (i get 2048 for GL_MAX_SPARSE_ARRAY_TEXTURE_LAYERS_ARB and GL_MAX_SPARSE_ARRAY_TEXTURE_LAYERS_AMD)

for 512 not occur, if i have one GLSparseTextureArray object

for 513 occurs.

No gl errors get by glGetError().

Tested on:

rx 480 8gb, last driver : memory leak (win 10)

AMD Vega 8, new driver :  loaded, no memory leak, driver shutdown after a while (win 10)

nv gtx850m - ok, old driver (win 8.1)

GTX 1050 - ok, new driver

I think, my problem same like this:

OpenGL ARB_sparse_texture (Driver crash, texture issues) 

Sorry for the bad English

[dipak]

Thank you for reporting it. I've moved this post to AMD OpenGL forum. Also, I have whitelisted you for AMD Devgurus community.

Thanks.

[vindast]

udt
Min c++ code, which produces this bug:

int w = 1024;   
int h = 1024;  
int n = CL::max(w, h);   
int numLods = 1;   
while (n >>= 1) 
{     
numLods++;   
}    
int numLayers = 0;   
glGetIntegerv(GL_MAX_SPARSE_ARRAY_TEXTURE_LAYERS, &numLayers);    
GLenum target = GL_TEXTURE_2D_ARRAY;   
GLenum internalFormat = GL_RGBA8;   
GLenum format = GL_RGBA;   
GLenum type = GL_UNSIGNED_BYTE;    
char* SysMem = new char[w * h * 4];    
int numPageSizes = 0;   
glGetInternalformativ(target, internalFormat, GL_NUM_VIRTUAL_PAGE_SIZES_ARB, 1, &numPageSizes);    
assert(numPageSizes > 0);    
std::cout << "numPageSizes = " << numPageSizes << std::endl;    
int* pageSizesX = new int[numPageSizes];   
int* pageSizesY = new int[numPageSizes];    
glGetInternalformativ(target, internalFormat, GL_VIRTUAL_PAGE_SIZE_X_ARB, numPageSizes, pageSizesX);   
glGetInternalformativ(target, internalFormat, GL_VIRTUAL_PAGE_SIZE_Y_ARB, numPageSizes, pageSizesY);    
int pageSizeIndex = -1;    
for (int i = 0; i < numPageSizes; i++) 
{     
if ((w % pageSizesX[i]) == 0      && (h % pageSizesY[i]) == 0) 
{       
pageSizeIndex = i;       
break;     
}      
std::cout << "x = " << pageSizesX[i] << ", y = " << pageSizesY[i] << std::endl;   
}    
std::cout << "pageSizeIndex = " << pageSizeIndex << std::endl;    
assert(pageSizeIndex != -1);    
GLuint id = -1;   
glGenTextures(1, &id);    
assert(id > 0);       
glActiveTexture(GL_TEXTURE0);   
glBindTexture(target, id);       
glTexParameteri(target, GL_TEXTURE_SPARSE_ARB, GL_TRUE);   
glTexParameteri(target, GL_VIRTUAL_PAGE_SIZE_INDEX_ARB, pageSizeIndex);   
glTexStorage3D(target, numLods, internalFormat, w, h, numLayers);    
int testCount = 10;    
for (int i = 0; i < testCount; i++) 
{      
glTexPageCommitmentARB(      
target,       
0, // lod      
0, // x      
0, // y      
0, // layer      
w,       
h,       
1,       
GL_TRUE);   
   
glPixelStorei(GL_UNPACK_ALIGNMENT, 1);      

glTexSubImage3D(      
target,       
0, // lod      
0, // x      
0, // y      
0, // layer      
w,       
h,       
1,       
format,       
type,       
SysMem     
);    
}