|
Title: Have the engine always used SSE2? Post by: sago007 on June 21, 2016, 11:20:05 AM Yesterday I worked on creating an automatic build of the Windows version.
Apart from libcurl and libxmp I got the version compiled with a MXE installation. However there was one thing that hit me. I had to add the following flags: Code: BASE_CFLAGS += -mmmx -msse2 Otherwise errors from the inline assembly: Code: LD build/release-mingw32-x86/oa_ded.x86.exe code/asm/ftola.c: In function 'qvmftolsse': code/asm/ftola.c:49:3: error: unknown register name '%xmm0' in 'asm' __asm__ volatile ^ code/asm/snapvector.c: In function 'qsnapvectorsse': code/asm/snapvector.c:38:2: error: unknown register name '%xmm2' in 'asm' __asm__ volatile ^ code/asm/snapvector.c:38:2: error: unknown register name '%xmm1' in 'asm' code/asm/snapvector.c:38:2: error: unknown register name '%xmm0' in 'asm' lto-wrapper: i686-w64-mingw32.static-gcc returned 1 exit status The code in question seems to be original idtech3 code. I wonder why I never had this problem before... Anybody knows? Title: Re: Have the engine always used SSE2? Post by: fromhell on June 21, 2016, 02:39:43 PM The original engine does use SSE/SIMD but in a non-restrictive way that doesn't leave out non-SSE targets. I can still run the latest OA SDL1 engine build on a non-MMX, non-SSE Pentium target :)
However the (engine-master) makefile does need sse2 exposed to the inline assembler, which is why it specifies the AMD64 architecture (the first on the AMD side to feature SSE2) Code: # k8 so inline assembler knows about SSE $(B)/ded/%.o: $(ASMDIR)/%.c $(DO_CC) -march=k8 |