GPU: Enable printf only between render boundaries and use a stack

Printf buffer read needs to be inside render boundaries
to work. Since render boundaries can be nested, use a stack.

Fixes assert when quitting blender.

source/blender/gpu/intern/gpu_context.cc

@@ -120,7 +120,6 @@ GPUContext *GPU_context_create(void *ghost_window, void *ghost_context)
 void GPU_context_discard(GPUContext *ctx_)
 {
   Context *ctx = unwrap(ctx_);
-  printf_end(ctx);
   delete ctx;
   active_ctx = nullptr;
 
@@ -140,7 +139,6 @@ void GPU_context_active_set(GPUContext *ctx_)
   Context *ctx = unwrap(ctx_);
 
   if (active_ctx) {
-    printf_end(active_ctx);
     active_ctx->deactivate();
   }
 
@@ -148,7 +146,6 @@ void GPU_context_active_set(GPUContext *ctx_)
 
   if (ctx) {
     ctx->activate();
-    printf_begin(ctx);
   }
 }
 
@@ -208,7 +205,6 @@ void GPU_render_begin()
    * but should be fixed for Metal. */
   if (backend) {
     backend->render_begin();
-    printf_end(active_ctx);
     printf_begin(active_ctx);
   }
 }
@@ -218,7 +214,6 @@ void GPU_render_end()
   BLI_assert(backend);
   if (backend) {
     printf_end(active_ctx);
-    printf_begin(active_ctx);
     backend->render_end();
   }
 }

source/blender/gpu/intern/gpu_context_private.hh

@@ -60,7 +60,8 @@ class Context {
   static int context_counter;
   int context_id = 0;
 
-  GPUStorageBuf *printf_buf = nullptr;
+  /* Used as a stack. Each render_begin/end pair will push pop from the stack. */
+  Vector<GPUStorageBuf *> printf_buf;
 
  protected:
   /** Thread on which this context is active. */

source/blender/gpu/intern/gpu_shader.cc

@@ -451,8 +451,8 @@ void GPU_shader_bind(GPUShader *gpu_shader)
     }
   }
 #if GPU_SHADER_PRINTF_ENABLE
-  if (ctx->printf_buf) {
-    GPU_storagebuf_bind(ctx->printf_buf, GPU_SHADER_PRINTF_SLOT);
+  if (!ctx->printf_buf.is_empty()) {
+    GPU_storagebuf_bind(ctx->printf_buf.last(), GPU_SHADER_PRINTF_SLOT);
   }
 #endif
 }

source/blender/gpu/intern/gpu_shader_log.cc

@@ -333,9 +333,10 @@ void printf_begin(Context *ctx)
   if (!shader::gpu_shader_dependency_has_printf()) {
     return;
   }
-  BLI_assert(ctx->printf_buf == nullptr);
-  ctx->printf_buf = GPU_storagebuf_create(GPU_SHADER_PRINTF_MAX_CAPACITY * sizeof(uint32_t));
-  GPU_storagebuf_clear_to_zero(ctx->printf_buf);
+  GPUStorageBuf *printf_buf = GPU_storagebuf_create(GPU_SHADER_PRINTF_MAX_CAPACITY *
+                                                    sizeof(uint32_t));
+  GPU_storagebuf_clear_to_zero(printf_buf);
+  ctx->printf_buf.append(printf_buf);
 }
 
 void printf_end(Context *ctx)
@@ -343,14 +344,14 @@ void printf_end(Context *ctx)
   if (ctx == nullptr) {
     return;
   }
-  if (ctx->printf_buf == nullptr) {
+  if (ctx->printf_buf.is_empty()) {
     return;
   }
+  GPUStorageBuf *printf_buf = ctx->printf_buf.pop_last();
 
   Vector<uint32_t> data(GPU_SHADER_PRINTF_MAX_CAPACITY);
-  GPU_storagebuf_read(ctx->printf_buf, data.data());
-  GPU_storagebuf_free(ctx->printf_buf);
-  ctx->printf_buf = nullptr;
+  GPU_storagebuf_read(printf_buf, data.data());
+  GPU_storagebuf_free(printf_buf);
 
   uint32_t data_len = data[0];
   if (data_len == 0) {