diff options
Diffstat (limited to 'intern/cycles/CMakeLists.txt')
| -rw-r--r-- | intern/cycles/CMakeLists.txt | 86 |
1 files changed, 45 insertions, 41 deletions
diff --git a/intern/cycles/CMakeLists.txt b/intern/cycles/CMakeLists.txt index c53a9f91cc0..5844c2480d6 100644 --- a/intern/cycles/CMakeLists.txt +++ b/intern/cycles/CMakeLists.txt @@ -41,61 +41,65 @@ elseif(WIN32 AND MSVC) set(CYCLES_AVX2_ARCH_FLAGS "/arch:SSE2") endif() + # Unlike GCC/clang we still use fast math, because there is no fine + # grained control and the speedup we get here is too big to ignore. + set(CYCLES_KERNEL_FLAGS "/fp:fast -D_CRT_SECURE_NO_WARNINGS /GS-") + # there is no /arch:SSE3, but intrinsics are available anyway if(CMAKE_CL_64) - set(CYCLES_SSE2_KERNEL_FLAGS "/fp:fast -D_CRT_SECURE_NO_WARNINGS /GS-") - set(CYCLES_SSE3_KERNEL_FLAGS "/fp:fast -D_CRT_SECURE_NO_WARNINGS /GS-") - set(CYCLES_SSE41_KERNEL_FLAGS "/fp:fast -D_CRT_SECURE_NO_WARNINGS /GS-") - set(CYCLES_AVX_KERNEL_FLAGS "${CYCLES_AVX_ARCH_FLAGS} /fp:fast -D_CRT_SECURE_NO_WARNINGS /GS-") - set(CYCLES_AVX2_KERNEL_FLAGS "${CYCLES_AVX2_ARCH_FLAGS} /fp:fast -D_CRT_SECURE_NO_WARNINGS /GS-") + set(CYCLES_SSE2_KERNEL_FLAGS "${CYCLES_KERNEL_FLAGS}") + set(CYCLES_SSE3_KERNEL_FLAGS "${CYCLES_KERNEL_FLAGS}") + set(CYCLES_SSE41_KERNEL_FLAGS "${CYCLES_KERNEL_FLAGS}") + set(CYCLES_AVX_KERNEL_FLAGS "${CYCLES_AVX_ARCH_FLAGS} ${CYCLES_KERNEL_FLAGS}") + set(CYCLES_AVX2_KERNEL_FLAGS "${CYCLES_AVX2_ARCH_FLAGS} ${CYCLES_KERNEL_FLAGS}") else() - set(CYCLES_SSE2_KERNEL_FLAGS "/arch:SSE2 /fp:fast -D_CRT_SECURE_NO_WARNINGS /GS-") - set(CYCLES_SSE3_KERNEL_FLAGS "/arch:SSE2 /fp:fast -D_CRT_SECURE_NO_WARNINGS /GS-") - set(CYCLES_SSE41_KERNEL_FLAGS "/arch:SSE2 /fp:fast -D_CRT_SECURE_NO_WARNINGS /GS-") - set(CYCLES_AVX_KERNEL_FLAGS "${CYCLES_AVX_ARCH_FLAGS} /fp:fast -D_CRT_SECURE_NO_WARNINGS /GS-") - set(CYCLES_AVX2_KERNEL_FLAGS "${CYCLES_AVX2_ARCH_FLAGS} /fp:fast -D_CRT_SECURE_NO_WARNINGS /GS-") + set(CYCLES_SSE2_KERNEL_FLAGS "/arch:SSE2 ${CYCLES_KERNEL_FLAGS}") + set(CYCLES_SSE3_KERNEL_FLAGS "/arch:SSE2 ${CYCLES_KERNEL_FLAGS}") + set(CYCLES_SSE41_KERNEL_FLAGS "/arch:SSE2 ${CYCLES_KERNEL_FLAGS}") + set(CYCLES_AVX_KERNEL_FLAGS "${CYCLES_AVX_ARCH_FLAGS} ${CYCLES_KERNEL_FLAGS}") + set(CYCLES_AVX2_KERNEL_FLAGS "${CYCLES_AVX2_ARCH_FLAGS} ${CYCLES_KERNEL_FLAGS}") endif() - set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /fp:fast -D_CRT_SECURE_NO_WARNINGS /GS-") + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${CYCLES_KERNEL_FLAGS}") set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} /Ox") set(CMAKE_CXX_FLAGS_RELWITHDEBINFO "${CMAKE_CXX_FLAGS_RELWITHDEBINFO} /Ox") set(CMAKE_CXX_FLAGS_MINSIZEREL "${CMAKE_CXX_FLAGS_MINSIZEREL} /Ox") - - set(CYCLES_KERNEL_FLAGS "/fp:fast -D_CRT_SECURE_NO_WARNINGS /GS-") -elseif(CMAKE_COMPILER_IS_GNUCC) +elseif(CMAKE_COMPILER_IS_GNUCC OR (CMAKE_CXX_COMPILER_ID MATCHES "Clang")) check_cxx_compiler_flag(-msse CXX_HAS_SSE) check_cxx_compiler_flag(-mavx CXX_HAS_AVX) check_cxx_compiler_flag(-mavx2 CXX_HAS_AVX2) - set(CYCLES_KERNEL_FLAGS "-ffast-math") - if(CXX_HAS_SSE) - set(CYCLES_SSE2_KERNEL_FLAGS "-ffast-math -msse -msse2 -mfpmath=sse") - set(CYCLES_SSE3_KERNEL_FLAGS "-ffast-math -msse -msse2 -msse3 -mssse3 -mfpmath=sse") - set(CYCLES_SSE41_KERNEL_FLAGS "-ffast-math -msse -msse2 -msse3 -mssse3 -msse4.1 -mfpmath=sse") - endif() - if(CXX_HAS_AVX) - set(CYCLES_AVX_KERNEL_FLAGS "-ffast-math -msse -msse2 -msse3 -mssse3 -msse4.1 -mavx -mfpmath=sse") - endif() - if(CXX_HAS_AVX2) - set(CYCLES_AVX2_KERNEL_FLAGS "-ffast-math -msse -msse2 -msse3 -mssse3 -msse4.1 -mavx -mavx2 -mfma -mlzcnt -mbmi -mbmi2 -mf16c -mfpmath=sse") + + # Assume no signal trapping for better code generation. + set(CYCLES_KERNEL_FLAGS "-fno-trapping-math") + # Avoid overhead of setting errno for NaNs. + set(CYCLES_KERNEL_FLAGS "${CYCLES_KERNEL_FLAGS} -fno-math-errno") + # Let compiler optimize 0.0 - x without worrying about signed zeros. + set(CYCLES_KERNEL_FLAGS "${CYCLES_KERNEL_FLAGS} -fno-signed-zeros") + + if(CMAKE_COMPILER_IS_GNUCC) + # Assume no signal trapping for better code generation. + set(CYCLES_KERNEL_FLAGS "${CYCLES_KERNEL_FLAGS} -fno-signaling-nans") + # Assume a fixed rounding mode for better constant folding. + set(CYCLES_KERNEL_FLAGS "${CYCLES_KERNEL_FLAGS} -fno-rounding-math") endif() - set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -ffast-math -fno-finite-math-only") -elseif(CMAKE_CXX_COMPILER_ID MATCHES "Clang") - check_cxx_compiler_flag(-msse CXX_HAS_SSE) - check_cxx_compiler_flag(-mavx CXX_HAS_AVX) - check_cxx_compiler_flag(-mavx2 CXX_HAS_AVX2) - set(CYCLES_KERNEL_FLAGS "-ffast-math") + if(CXX_HAS_SSE) - set(CYCLES_SSE2_KERNEL_FLAGS "-ffast-math -msse -msse2") - set(CYCLES_SSE3_KERNEL_FLAGS "-ffast-math -msse -msse2 -msse3 -mssse3") - set(CYCLES_SSE41_KERNEL_FLAGS "-ffast-math -msse -msse2 -msse3 -mssse3 -msse4.1") - endif() - if(CXX_HAS_AVX) - set(CYCLES_AVX_KERNEL_FLAGS "-ffast-math -msse -msse2 -msse3 -mssse3 -msse4.1 -mavx") - endif() - if(CXX_HAS_AVX2) - set(CYCLES_AVX2_KERNEL_FLAGS "-ffast-math -msse -msse2 -msse3 -mssse3 -msse4.1 -mavx -mavx2 -mfma -mlzcnt -mbmi -mbmi2 -mf16c") + if(CMAKE_COMPILER_IS_GNUCC) + set(CYCLES_KERNEL_FLAGS "${CYCLES_KERNEL_FLAGS} -mfpmath=sse") + endif() + + set(CYCLES_SSE2_KERNEL_FLAGS "${CYCLES_KERNEL_FLAGS} -msse -msse2") + set(CYCLES_SSE3_KERNEL_FLAGS "${CYCLES_SSE2_KERNEL_FLAGS} -msse3 -mssse3") + set(CYCLES_SSE41_KERNEL_FLAGS "${CYCLES_SSE3_KERNEL_FLAGS} -msse4.1") + if(CXX_HAS_AVX) + set(CYCLES_AVX_KERNEL_FLAGS "${CYCLES_SSE41_KERNEL_FLAGS} -mavx") + endif() + if(CXX_HAS_AVX2) + set(CYCLES_AVX2_KERNEL_FLAGS "${CYCLES_SSE41_KERNEL_FLAGS} -mavx -mavx2 -mfma -mlzcnt -mbmi -mbmi2 -mf16c") + endif() endif() - set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -ffast-math -fno-finite-math-only") + + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${CYCLES_KERNEL_FLAGS}") endif() if(CXX_HAS_SSE) |