diff --git a/ps2xRuntime/CMakeLists.txt b/ps2xRuntime/CMakeLists.txt index e4dc195..aad5fb8 100644 --- a/ps2xRuntime/CMakeLists.txt +++ b/ps2xRuntime/CMakeLists.txt @@ -552,6 +552,11 @@ if(PS2X_IS_VITA) endif() endif() +# Every configuration, not just Release: SSE4.1 is required to compile, not an +# optimisation. PUBLIC so recompiled game code linking against ps2_runtime +# inherits it -- that code is where most of the SSE4.1 intrinsics actually are. +EnableX86SimdBaseline(ps2_runtime) + if(CMAKE_BUILD_TYPE STREQUAL "Release" OR CMAKE_BUILD_TYPE STREQUAL "RelWithDebInfo") EnableFastReleaseMode(ps2_runtime) EnableFastReleaseMode(ps2EntryRunner) diff --git a/ps2xRuntime/cmake/ReleaseMode.cmake b/ps2xRuntime/cmake/ReleaseMode.cmake index 92ef0d5..0b97e90 100644 --- a/ps2xRuntime/cmake/ReleaseMode.cmake +++ b/ps2xRuntime/cmake/ReleaseMode.cmake @@ -2,6 +2,26 @@ include(CheckIPOSupported) check_ipo_supported(RESULT IPO_SUPPORTED OUTPUT IPO_ERROR) +# ps2_runtime.h unconditionally includes and the recompiler emits +# SSE4.1-only intrinsics (_mm_blendv_ps and friends) for the COP2/FPU select +# idioms, so SSE4.1 is a hard requirement of the codebase rather than a tuning +# knob. MSVC enables it implicitly (its intrinsics are not gated by a target +# feature), but GCC and Clang default to the plain x86-64 baseline, which is +# SSE2 -- so on any stock Linux or macOS toolchain those translation units fail +# to compile with "always_inline function ... requires target feature 'sse4.1'". +# +# This must not live in EnableFastReleaseMode: that is only applied for Release +# and RelWithDebInfo, whereas the requirement applies to every configuration. +function(EnableX86SimdBaseline TargetName) + if(MSVC) + return() + endif() + if(NOT CMAKE_SYSTEM_PROCESSOR MATCHES "^(x86_64|AMD64|amd64|i[3-6]86|x86)$") + return() + endif() + target_compile_options(${TargetName} PUBLIC -msse4.1) +endfunction() + function(EnableFastReleaseMode TargetName) message("> Enabling optimization for: ${TargetName}") if(MSVC)