griefith/test
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

Fix: Metal: GPU_finish() not syncing when there's no (current) command buffer to sync on

Browse Source 
Reported by @fclem.  On Metal GPU_finish enacts a CPU<-> GPU sync by
submitting a command buffer and waiting on the completion event.
However if there was no command buffer to submit then the call was
returning immediately, regardless of any outstanding GPU work.
This fixes that case by keeping track of all outstanding work and
blocking on that.

Authored by Apple: James McCarthy

Co-authored-by: James McCarthy <jamesmccarthy@apple.com>
Pull Request: https://projects.blender.org/blender/blender/pulls/128987
This commit is contained in:
Jason Fielder
2024-10-14 17:12:03 +02:00
committed by Clément Foucault
parent c082260b50
commit 6da42e9c95
3 changed files with 37 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
source/blender/gpu/metal/mtl_command_buffer.mm
View File

@@ -20,7 +20,7 @@ using namespace blender::gpu;
namespace blender::gpu {
/* Counter for active command buffers. */
int MTLCommandBufferManager::num_active_cmd_bufs = 0;
volatile std::atomic<int> MTLCommandBufferManager::num_active_cmd_bufs_in_system = 0;
/* -------------------------------------------------------------------- */
/** \name MTLCommandBuffer initialization and render coordination.
@@ -50,7 +50,7 @@ id<MTLCommandBuffer> MTLCommandBufferManager::ensure_begin()
*
* NOTE: We currently stall until completion of GPU work upon ::submit if we have reached the
* in-flight command buffer limit. */
BLI_assert(MTLCommandBufferManager::num_active_cmd_bufs <
BLI_assert(MTLCommandBufferManager::num_active_cmd_bufs_in_system <
GHOST_ContextCGL::max_command_buffer_count);
if (G.debug & G_DEBUG_GPU) {
@@ -68,7 +68,7 @@ id<MTLCommandBuffer> MTLCommandBufferManager::ensure_begin()
}
[active_command_buffer_ retain];
MTLCommandBufferManager::num_active_cmd_bufs++;
context_.main_command_buffer.inc_active_command_buffer_count();
/* Ensure we begin new Scratch Buffer if we are on a new frame. */
MTLScratchBufferManager &mem = context_.memory_manager;
@@ -90,6 +90,12 @@ bool MTLCommandBufferManager::submit(bool wait)
{
/* Skip submission if command buffer is empty. */
if (empty_ || active_command_buffer_ == nil) {
if (wait) {
/* Wait for any previously submitted work on this context to complete. */
while (context_.main_command_buffer.get_active_command_buffer_count()) {
std::this_thread::sleep_for(std::chrono::milliseconds(1));
}
}
return false;
}
@@ -123,7 +129,7 @@ bool MTLCommandBufferManager::submit(bool wait)
[cmd_buffer_ref release];
/* Decrement count. */
MTLCommandBufferManager::num_active_cmd_bufs--;
context_.main_command_buffer.dec_active_command_buffer_count();
}];
/* Submit command buffer to GPU. */
@@ -131,7 +137,7 @@ bool MTLCommandBufferManager::submit(bool wait)
/* If we have too many active command buffers in flight, wait until completed to avoid running
* out. We can increase */
if (MTLCommandBufferManager::num_active_cmd_bufs >=
if (MTLCommandBufferManager::num_active_cmd_bufs_in_system >=
(GHOST_ContextCGL::max_command_buffer_count - 1))
{
wait = true;

23
source/blender/gpu/metal/mtl_context.hh
View File

@@ -548,8 +548,8 @@ class MTLCommandBufferManager {
friend class MTLContext;
public:
/* Counter for active command buffers. */
static int num_active_cmd_bufs;
/* Counter for all active command buffers. */
static volatile std::atomic<int> num_active_cmd_bufs_in_system;
private:
/* Associated Context and properties. */
@@ -559,6 +559,7 @@ class MTLCommandBufferManager {
/* CommandBuffer tracking. */
id<MTLCommandBuffer> active_command_buffer_ = nil;
id<MTLCommandBuffer> last_submitted_command_buffer_ = nil;
volatile std::atomic<int> num_active_cmd_bufs = 0;
/* Active MTLCommandEncoders. */
enum {
@@ -653,6 +654,24 @@ class MTLCommandBufferManager {
void push_debug_group(const char *name, int index);
void pop_debug_group();
void inc_active_command_buffer_count()
{
num_active_cmd_bufs_in_system++;
num_active_cmd_bufs++;
}
void dec_active_command_buffer_count()
{
BLI_assert(num_active_cmd_bufs_in_system > 0 && num_active_cmd_bufs > 0);
num_active_cmd_bufs_in_system--;
num_active_cmd_bufs--;
}
int get_active_command_buffer_count()
{
return num_active_cmd_bufs;
}
private:
/* Begin new command buffer. */
id<MTLCommandBuffer> ensure_begin();

8
source/blender/gpu/metal/mtl_context.mm
View File

@@ -2706,7 +2706,7 @@ void present(MTLRenderPassDescriptor *blit_descriptor,
* possible. This command buffer is separate as it does not utilize the global state
* for rendering as the main context does. */
id<MTLCommandBuffer> cmdbuf = [ctx->queue commandBuffer];
MTLCommandBufferManager::num_active_cmd_bufs++;
ctx->main_command_buffer.inc_active_command_buffer_count();
/* Do Present Call and final Blit to MTLDrawable. */
id<MTLRenderCommandEncoder> enc = [cmdbuf renderCommandEncoderWithDescriptor:blit_descriptor];
@@ -2739,8 +2739,10 @@ void present(MTLRenderPassDescriptor *blit_descriptor,
[cmd_buffer_ref release];
/* Decrement count */
MTLCommandBufferManager::num_active_cmd_bufs--;
MTL_LOG_INFO("Active command buffers: %d", MTLCommandBufferManager::num_active_cmd_bufs);
ctx->main_command_buffer.dec_active_command_buffer_count();
MTL_LOG_INFO("Active command buffers: %d",
int(MTLCommandBufferManager::num_active_cmd_bufs_in_system));
/* Drawable count and latency management. */
MTLContext::max_drawables_in_flight--;