Better VRAM capacity reporting correction for Intel dGPUs

Author: ProjectPhysX

src/opencl.hpp

@@ -138,19 +138,15 @@ struct Device_Info {
 		const float arm = (float)(contains(to_lower(vendor), "arm"))*(is_gpu?8.0f:1.0f); // ARM GPUs usually have 8 cores/CU, ARM CPUs have 1 core/CU
 		cores = to_uint((float)compute_units*(nvidia+amd+intel+apple+arm)); // for CPUs, compute_units is the number of threads (twice the number of cores with hyperthreading)
 		tflops = 1E-6f*(float)cores*(float)ipc*(float)clock_frequency; // estimated device floating point performance in TeraFLOPs/s
-		if(intel==8.0f) { // fix wrong global memory reporting for Intel Arc GPUs
-			if((contains_any(name, {"A770", "0x56a0"})&&memory>=11739u&&memory<14168u)||(contains_any(name, {"A770", "A750", "A580", "0x56a0", "0x56a1", "0x56a2"})&&memory>=5869u&&memory<7084u)||(contains_any(name, {"A380", "0x56a5"})&&memory>=4402u&&memory<5313u)) { // 72.5%-87.5% reporting -> /0.8
-				memory = (uint)((cl_device.getInfo<CL_DEVICE_GLOBAL_MEM_SIZE>()*5ull/4ull)/1048576ull); // fix wrong (80% on Windows) memory reporting on Intel Arc
-			}
-			if((contains_any(name, {"A770", "0x56a0"})&&memory>=14168u&&memory<15625u)||(contains_any(name, {"A770", "A750", "A580", "0x56a0", "0x56a1", "0x56a2"})&&memory>=7084u&&memory<7812u)||(contains_any(name, {"A380", "0x56a5"})&&memory>=5313u&&memory<5859u)) { // 87.5%-96.5% reporting -> /0.95
-				memory = (uint)((cl_device.getInfo<CL_DEVICE_GLOBAL_MEM_SIZE>()*20ull/19ull)/1048576ull); // fix wrong (95% on Linux) memory reporting on Intel Arc
-			}
-			if((contains_any(name, {"A770", "0x56a0"})&&memory>=15625u&&memory<16030u)||(contains_any(name, {"A770", "A750", "A580", "0x56a0", "0x56a1", "0x56a2"})&&memory>=7812u&&memory<8015u)||(contains_any(name, {"A380", "0x56a5"})&&memory>=5859u&&memory<6011u)) { // 96.5%-99.0% reporting -> /0.98
-				memory = (uint)((cl_device.getInfo<CL_DEVICE_GLOBAL_MEM_SIZE>()*50ull/49ull)/1048576ull); // fix wrong (98% on Windows) memory reporting on Intel Arc
-			}
+		if(intel>=8.0f&&!uses_ram) { // fix wrong global memory capacity reporting for Intel dGPUs
+#if defined(_WIN32)
+			memory = (uint)((cl_device.getInfo<CL_DEVICE_GLOBAL_MEM_SIZE>()*50ull/49ull)/1048576ull); // 98% on Windows https://github.com/intel/compute-runtime/blob/master/shared/source/os_interface/windows/wddm_memory_manager.cpp#L964
+#elif defined(__linux__)
+			memory = (uint)((cl_device.getInfo<CL_DEVICE_GLOBAL_MEM_SIZE>()*20ull/19ull)/1048576ull); // 95% on Linux   https://github.com/intel/compute-runtime/blob/master/shared/source/os_interface/linux/drm_memory_manager.cpp#L1424
+#endif // Linux
 		}
 		patch_nvidia_fp16 = patch_nvidia_fp16||(nvidia>0.0f&&atof(driver_version.substr(0, 6).c_str())>=520.00&&!nvidia_192_cores_per_cu&&!contains_any(to_lower(name), {"gtx 8", "gtx 9", "quadro m", "tesla m", "gtx titan"})); // enable for all Nvidia GPUs with driver>=520.00 except Kepler and Maxwell
-		patch_intel_gpu_above_4gb = patch_intel_gpu_above_4gb||((intel>=8.0f)&&(memory>4096u)); // enable memory allocations greater than 4GB for Intel GPUs with >4GB VRAM
+		patch_intel_gpu_above_4gb = patch_intel_gpu_above_4gb||(intel>=8.0f&&memory>4096u); // enable memory allocations greater than 4GB for Intel GPUs with >4GB VRAM
 		patch_legacy_gpu_fma = patch_legacy_gpu_fma||arm>0.0f; // enable for all ARM GPUs
 		if(patch_nvidia_fp16) is_fp16_capable = 2u;
 	}