[Bf-blender-cvs] [2eaf90b305f] master: Cycles: Remove Fermi support from CMake and update runtime checks in device_cuda.cpp.

[Thomas Dinges]

Commit: 2eaf90b305fc6fb37f775f288c3d5cd34cc66c81
Author: Thomas Dinges
Date:   Sat Feb 17 16:15:07 2018 +0100
Branches: master
https://developer.blender.org/rB2eaf90b305fc6fb37f775f288c3d5cd34cc66c81

Cycles: Remove Fermi support from CMake and update runtime checks in device_cuda.cpp.

Fermi code in Cycles kernel and texture system are coming next.

===================================================================

M	CMakeLists.txt
M	build_files/cmake/config/blender_release.cmake
M	intern/cycles/device/device_cuda.cpp
M	intern/cycles/kernel/CMakeLists.txt

===================================================================

diff --git a/CMakeLists.txt b/CMakeLists.txt
index dce622d4778..025b76cd22b 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -409,7 +409,7 @@ option(WITH_CYCLES_OSL				"Build Cycles with OSL support" ${_init_CYCLES_OSL})
 option(WITH_CYCLES_OPENSUBDIV		"Build Cycles with OpenSubdiv support" ${_init_CYCLES_OPENSUBDIV})
 option(WITH_CYCLES_CUDA_BINARIES	"Build Cycles CUDA binaries" OFF)
 option(WITH_CYCLES_CUBIN_COMPILER	"Build cubins with nvrtc based compiler instead of nvcc" OFF)
-set(CYCLES_CUDA_BINARIES_ARCH sm_20 sm_21 sm_30 sm_35 sm_37 sm_50 sm_52 sm_60 sm_61 CACHE STRING "CUDA architectures to build binaries for")
+set(CYCLES_CUDA_BINARIES_ARCH sm_30 sm_35 sm_37 sm_50 sm_52 sm_60 sm_61 CACHE STRING "CUDA architectures to build binaries for")
 mark_as_advanced(CYCLES_CUDA_BINARIES_ARCH)
 unset(PLATFORM_DEFAULT)
 option(WITH_CYCLES_LOGGING	"Build Cycles with logging support" ON)
diff --git a/build_files/cmake/config/blender_release.cmake b/build_files/cmake/config/blender_release.cmake
index 1d1793e0aba..26267fd314a 100644
--- a/build_files/cmake/config/blender_release.cmake
+++ b/build_files/cmake/config/blender_release.cmake
@@ -54,7 +54,7 @@ set(WITH_X11_XF86VMODE       ON  CACHE BOOL "" FORCE)
 set(WITH_PLAYER              ON  CACHE BOOL "" FORCE)
 set(WITH_MEM_JEMALLOC        ON  CACHE BOOL "" FORCE)
 set(WITH_CYCLES_CUDA_BINARIES ON  CACHE BOOL "" FORCE)
-set(CYCLES_CUDA_BINARIES_ARCH sm_20;sm_21;sm_30;sm_35;sm_37;sm_50;sm_52;sm_60;sm_61 CACHE STRING "" FORCE)
+set(CYCLES_CUDA_BINARIES_ARCH sm_30;sm_35;sm_37;sm_50;sm_52;sm_60;sm_61 CACHE STRING "" FORCE)
 
 # platform dependent options
 if(UNIX AND NOT APPLE)
diff --git a/intern/cycles/device/device_cuda.cpp b/intern/cycles/device/device_cuda.cpp
index dfedf922ca9..9644937d906 100644
--- a/intern/cycles/device/device_cuda.cpp
+++ b/intern/cycles/device/device_cuda.cpp
@@ -322,9 +322,9 @@ public:
 		cuDeviceGetAttribute(&major, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR, cuDevId);
 		cuDeviceGetAttribute(&minor, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR, cuDevId);
 
-		/* We only support sm_20 and above */
-		if(major < 2) {
-			cuda_error_message(string_printf("CUDA device supported only with compute capability 2.0 or up, found %d.%d.", major, minor));
+		/* We only support sm_30 and above */
+		if(major < 3) {
+			cuda_error_message(string_printf("CUDA device supported only with compute capability 3.0 or up, found %d.%d.", major, minor));
 			return false;
 		}
 
@@ -462,9 +462,9 @@ public:
 
 #ifdef _WIN32
 		if(have_precompiled_kernels()) {
-			if(major < 2) {
+			if(major < 3) {
 				cuda_error_message(string_printf(
-				        "CUDA device requires compute capability 2.0 or up, "
+				        "CUDA device requires compute capability 3.0 or up, "
 				        "found %d.%d. Your GPU is not supported.",
 				        major, minor));
 			}
@@ -2532,7 +2532,7 @@ void device_cuda_info(vector<DeviceInfo>& devices)
 
 		int major;
 		cuDeviceGetAttribute(&major, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR, num);
-		if(major < 2) {
+		if(major < 3) {
 			VLOG(1) << "Ignoring device \"" << name
 			        << "\", compute capability is too low.";
 			continue;
@@ -2544,7 +2544,7 @@ void device_cuda_info(vector<DeviceInfo>& devices)
 		info.description = string(name);
 		info.num = num;
 
-		info.advanced_shading = (major >= 2);
+		info.advanced_shading = (major >= 3);
 		info.has_fermi_limits = !(major >= 3);
 		info.has_half_images = (major >= 3);
 		info.has_volume_decoupled = false;
diff --git a/intern/cycles/kernel/CMakeLists.txt b/intern/cycles/kernel/CMakeLists.txt
index d981b67559e..4d4b8602c78 100644
--- a/intern/cycles/kernel/CMakeLists.txt
+++ b/intern/cycles/kernel/CMakeLists.txt
@@ -316,12 +316,8 @@ set(SRC_SPLIT_HEADERS
 # CUDA module
 
 if(WITH_CYCLES_CUDA_BINARIES)
-	# 32 bit or 64 bit
-	if(CUDA_64_BIT_DEVICE_CODE)
-		set(CUDA_BITS 64)
-	else()
-		set(CUDA_BITS 32)
-	endif()
+	# 64 bit only
+	set(CUDA_BITS 64)
 
 	# CUDA version
 	execute_process(COMMAND ${CUDA_NVCC_EXECUTABLE} "--version" OUTPUT_VARIABLE NVCC_OUT)
@@ -420,17 +416,13 @@ if(WITH_CYCLES_CUDA_BINARIES)
 	endmacro()
 
 	foreach(arch ${CYCLES_CUDA_BINARIES_ARCH})
-		if(CUDA_VERSION GREATER "89" AND ${arch} MATCHES "sm_2.")
-			message(STATUS "CUDA binaries for ${arch} disabled, not supported by CUDA 9.")
-		else()
-			# Compile regular kernel
-			CYCLES_CUDA_KERNEL_ADD(${arch} filter "" "${cuda_filter_sources}" FALSE)
-			CYCLES_CUDA_KERNEL_ADD(${arch} kernel "" "${cuda_sources}" FALSE)
+		# Compile regular kernel
+		CYCLES_CUDA_KERNEL_ADD(${arch} filter "" "${cuda_filter_sources}" FALSE)
+		CYCLES_CUDA_KERNEL_ADD(${arch} kernel "" "${cuda_sources}" FALSE)
 
-			if(WITH_CYCLES_CUDA_SPLIT_KERNEL_BINARIES)
-				# Compile split kernel
-				CYCLES_CUDA_KERNEL_ADD(${arch} kernel_split "-D __SPLIT__" ${cuda_sources} FALSE)
-			endif()
+		if(WITH_CYCLES_CUDA_SPLIT_KERNEL_BINARIES)
+			# Compile split kernel
+			CYCLES_CUDA_KERNEL_ADD(${arch} kernel_split "-D __SPLIT__" ${cuda_sources} FALSE)
 		endif()
 	endforeach()