use GL_MAX_COMPUTE_WORK_GROUP_INVOCATIONS instead of 1792 for portabi…

…lity
archibate · Mar 27, 2020 · 24bbc4a · 24bbc4a
1 parent 76ff830
commit 24bbc4a
Show file tree

Hide file tree

Showing 3 changed files with 16 additions and 4 deletions.
diff --git a/taichi/codegen/codegen_opengl.cpp b/taichi/codegen/codegen_opengl.cpp
@@ -305,10 +305,10 @@ int _rand_i32()\n\
 
     line_appender_header_.append_raw(kernel_header);
 
-    int threads_per_group = 1792;
-    if (num_threads_ < 1792)
+    int threads_per_group = opengl_get_threads_per_group();
+    if (num_threads_ < threads_per_group)
       threads_per_group = num_threads_;
-    num_groups_ = (num_threads_ + 1791) / 1792;
+    num_groups_ = (num_threads_ + threads_per_group - 1) / threads_per_group;
     if (threads_per_group == 0)
       threads_per_group = 1;
     emit("layout(local_size_x = {}, local_size_y = 1, local_size_z = 1) in;",

diff --git a/taichi/platform/opengl/opengl_api.cpp b/taichi/platform/opengl/opengl_api.cpp
@@ -286,6 +286,12 @@ bool is_opengl_api_available() {
   return true;
 }
 
+int opengl_get_threads_per_group() {
+  int ret = 1792;
+  glGetIntegerv(GL_MAX_COMPUTE_WORK_GROUP_INVOCATIONS, &ret);
+  return ret;
+}
+
 #else
 void create_glsl_root_buffer(size_t size) {
   TI_NOT_IMPLEMENTED
@@ -308,10 +314,15 @@ bool is_opengl_api_available() {
 }
 
 void initialize_opengl() {
+  TI_NOT_IMPLEMENTED
 }
 
 GLProgram *compile_glsl_program(std::string source) {
-  return nullptr;
+  TI_NOT_IMPLEMENTED
+}
+
+int opengl_get_threads_per_group() {
+  TI_NOT_IMPLEMENTED
 }
 #endif
 

diff --git a/taichi/platform/opengl/opengl_api.h b/taichi/platform/opengl/opengl_api.h
@@ -19,6 +19,7 @@ void begin_glsl_kernels(const std::vector<IOV> &iov);
 void launch_glsl_kernel(GLProgram *program, int num_groups);
 void end_glsl_kernels(const std::vector<IOV> &iov);
 GLProgram *compile_glsl_program(std::string source);
+int opengl_get_threads_per_group();
 
 }  // namespace opengl