notwa/crap
1
0
Fork 0
Code Issues Pull requests Releases Wiki Activity

use single-precision floats for ARM

Browse source 
NEON doesn't support doubles ;_;
anyway, this turns the two-times slowdown
caused by previous commits into a two-times speedup.
This commit is contained in:
Connor Olding 2015-04-07 13:55:12 -07:00
parent 95c3e58ff9
commit f1a8bcd876
2 changed files with 8 additions and 2 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

4
Makefile
View file

@ -70,8 +70,8 @@ linux: ALL_CXXFLAGS += $(OPT_FLAGS) -fpic
linux: VST_FLAGS += -D__cdecl= linux: VST_FLAGS += -D__cdecl=
linux: all linux: all
linux-arm: ALL_CFLAGS += -Ofast -mfpu=neon -fpic linux-arm: ALL_CFLAGS += -Ofast -march=native -DFORCE_SINGLE -fpic
linux-arm: ALL_CXXFLAGS += -Ofast -mfpu=neon -fpic linux-arm: ALL_CXXFLAGS += -Ofast -march=native -DFORCE_SINGLE -fpic
linux-arm: VST_FLAGS += -D__cdecl= linux-arm: VST_FLAGS += -D__cdecl=
linux-arm: all linux-arm: all

6
include/util.h
View file

@ -7,7 +7,13 @@
#define INNER static inline #define INNER static inline
#define PURE __attribute__((pure)) #define PURE __attribute__((pure))
#define CONST __attribute__((const)) #define CONST __attribute__((const))
#ifndef FORCE_SINGLE
typedef double v2df __attribute__((vector_size(16), aligned(16))); typedef double v2df __attribute__((vector_size(16), aligned(16)));
#else
typedef float v2df __attribute__((vector_size(8), aligned(8)));
#endif
typedef float v4sf __attribute__((vector_size(16), aligned(16))); typedef float v4sf __attribute__((vector_size(16), aligned(16)));
typedef unsigned long ulong; // __attribute((aligned(16))); typedef unsigned long ulong; // __attribute((aligned(16)));