diff options
author | wernsaar <wernsaar@googlemail.com> | 2014-05-13 17:25:19 +0200 |
---|---|---|
committer | wernsaar <wernsaar@googlemail.com> | 2014-05-13 17:25:19 +0200 |
commit | 43fbdb7a5a5b72bac922d405a9b626c56234c767 (patch) | |
tree | 15f4b5da9cf8115c37341b4e452003308df218a2 | |
parent | aa5c73e20f7d397e578f4b47ead214d015eb75f1 (diff) | |
download | openblas-43fbdb7a5a5b72bac922d405a9b626c56234c767.tar.gz openblas-43fbdb7a5a5b72bac922d405a9b626c56234c767.tar.bz2 openblas-43fbdb7a5a5b72bac922d405a9b626c56234c767.zip |
added ARMV5 as reference platform
-rw-r--r-- | Makefile.arm | 6 | ||||
-rw-r--r-- | common.h | 2 | ||||
-rw-r--r-- | getarch.c | 16 | ||||
-rw-r--r-- | kernel/arm/KERNEL.ARMV5 | 134 | ||||
-rw-r--r-- | param.h | 40 |
5 files changed, 197 insertions, 1 deletions
diff --git a/Makefile.arm b/Makefile.arm index 8502d5286..5bdd4d151 100644 --- a/Makefile.arm +++ b/Makefile.arm @@ -10,3 +10,9 @@ FCOMMON_OPT += -marm -mfpu=vfp -mfloat-abi=hard -march=armv6 endif +ifeq ($(CORE), ARMV5) +CCOMMON_OPT += -marm -mfpu=vfp -mfloat-abi=hard -march=armv6 +FCOMMON_OPT += -marm -mfpu=vfp -mfloat-abi=hard -march=armv6 +endif + + @@ -310,7 +310,7 @@ typedef int blasint; #define YIELDING SwitchToThread() #endif -#if defined(ARMV7) || defined(ARMV6) || defined(ARMV8) +#if defined(ARMV7) || defined(ARMV6) || defined(ARMV8) || defined(ARMV5) #define YIELDING asm volatile ("nop;nop;nop;nop;nop;nop;nop;nop; \n"); #endif @@ -724,6 +724,22 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. #else #endif +#ifdef FORCE_ARMV5 +#define FORCE +#define ARCHITECTURE "ARM" +#define SUBARCHITECTURE "ARMV5" +#define SUBDIRNAME "arm" +#define ARCHCONFIG "-DARMV5 " \ + "-DL1_DATA_SIZE=65536 -DL1_DATA_LINESIZE=32 " \ + "-DL2_SIZE=512488 -DL2_LINESIZE=32 " \ + "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 -DL2_ASSOCIATIVE=4 " \ + "-DHAVE_VFP" +#define LIBNAME "armv5" +#define CORENAME "ARMV5" +#else +#endif + + #ifdef FORCE_ARMV8 #define FORCE #define ARCHITECTURE "ARM64" diff --git a/kernel/arm/KERNEL.ARMV5 b/kernel/arm/KERNEL.ARMV5 new file mode 100644 index 000000000..ecf278cf9 --- /dev/null +++ b/kernel/arm/KERNEL.ARMV5 @@ -0,0 +1,134 @@ +SAMAXKERNEL = ../arm/amax.c +DAMAXKERNEL = ../arm/amax.c +CAMAXKERNEL = ../arm/zamax.c +ZAMAXKERNEL = ../arm/zamax.c + +SAMINKERNEL = ../arm/amin.c +DAMINKERNEL = ../arm/amin.c +CAMINKERNEL = ../arm/zamin.c +ZAMINKERNEL = ../arm/zamin.c + +SMAXKERNEL = ../arm/max.c +DMAXKERNEL = ../arm/max.c + +SMINKERNEL = ../arm/min.c +DMINKERNEL = ../arm/min.c + +ISAMAXKERNEL = ../arm/iamax.c +IDAMAXKERNEL = ../arm/iamax.c +ICAMAXKERNEL = ../arm/izamax.c +IZAMAXKERNEL = ../arm/izamax.c + +ISAMINKERNEL = ../arm/iamin.c +IDAMINKERNEL = ../arm/iamin.c +ICAMINKERNEL = ../arm/izamin.c +IZAMINKERNEL = ../arm/izamin.c + +ISMAXKERNEL = ../arm/imax.c +IDMAXKERNEL = ../arm/imax.c + +ISMINKERNEL = ../arm/imin.c +IDMINKERNEL = ../arm/imin.c + +SASUMKERNEL = ../arm/asum.c +DASUMKERNEL = ../arm/asum.c +CASUMKERNEL = ../arm/zasum.c +ZASUMKERNEL = ../arm/zasum.c + +SAXPYKERNEL = ../arm/axpy.c +DAXPYKERNEL = ../arm/axpy.c +CAXPYKERNEL = ../arm/zaxpy.c +ZAXPYKERNEL = ../arm/zaxpy.c + +SCOPYKERNEL = ../arm/copy.c +DCOPYKERNEL = ../arm/copy.c +CCOPYKERNEL = ../arm/zcopy.c +ZCOPYKERNEL = ../arm/zcopy.c + +SDOTKERNEL = ../arm/dot.c +DDOTKERNEL = ../arm/dot.c +CDOTKERNEL = ../arm/zdot.c +ZDOTKERNEL = ../arm/zdot.c + +SNRM2KERNEL = ../arm/nrm2.c +DNRM2KERNEL = ../arm/nrm2.c +CNRM2KERNEL = ../arm/znrm2.c +ZNRM2KERNEL = ../arm/znrm2.c + +SROTKERNEL = ../arm/rot.c +DROTKERNEL = ../arm/rot.c +CROTKERNEL = ../arm/zrot.c +ZROTKERNEL = ../arm/zrot.c + +SSCALKERNEL = ../arm/scal.c +DSCALKERNEL = ../arm/scal.c +CSCALKERNEL = ../arm/zscal.c +ZSCALKERNEL = ../arm/zscal.c + +SSWAPKERNEL = ../arm/swap.c +DSWAPKERNEL = ../arm/swap.c +CSWAPKERNEL = ../arm/zswap.c +ZSWAPKERNEL = ../arm/zswap.c + +SGEMVNKERNEL = ../arm/gemv_n.c +DGEMVNKERNEL = ../arm/gemv_n.c +CGEMVNKERNEL = ../arm/zgemv_n.c +ZGEMVNKERNEL = ../arm/zgemv_n.c + +SGEMVTKERNEL = ../arm/gemv_t.c +DGEMVTKERNEL = ../arm/gemv_t.c +CGEMVTKERNEL = ../arm/zgemv_t.c +ZGEMVTKERNEL = ../arm/zgemv_t.c + +STRMMKERNEL = ../generic/trmmkernel_2x2.c +DTRMMKERNEL = ../generic/trmmkernel_2x2.c +CTRMMKERNEL = ../generic/ztrmmkernel_2x2.c +ZTRMMKERNEL = ../generic/ztrmmkernel_2x2.c + +SGEMMKERNEL = ../generic/gemmkernel_2x2.c +SGEMMONCOPY = ../generic/gemm_ncopy_2.c +SGEMMOTCOPY = ../generic/gemm_tcopy_2.c +SGEMMONCOPYOBJ = sgemm_oncopy.o +SGEMMOTCOPYOBJ = sgemm_otcopy.o + +DGEMMKERNEL = ../generic/gemmkernel_2x2.c +DGEMMONCOPY = ../generic/gemm_ncopy_2.c +DGEMMOTCOPY = ../generic/gemm_tcopy_2.c +DGEMMONCOPYOBJ = dgemm_oncopy.o +DGEMMOTCOPYOBJ = dgemm_otcopy.o + +CGEMMKERNEL = ../generic/zgemmkernel_2x2.c +CGEMMONCOPY = ../generic/zgemm_ncopy_2.c +CGEMMOTCOPY = ../generic/zgemm_tcopy_2.c +CGEMMONCOPYOBJ = cgemm_oncopy.o +CGEMMOTCOPYOBJ = cgemm_otcopy.o + +ZGEMMKERNEL = ../generic/zgemmkernel_2x2.c +ZGEMMONCOPY = ../generic/zgemm_ncopy_2.c +ZGEMMOTCOPY = ../generic/zgemm_tcopy_2.c +ZGEMMONCOPYOBJ = zgemm_oncopy.o +ZGEMMOTCOPYOBJ = zgemm_otcopy.o + +STRSMKERNEL_LN = ../generic/trsm_kernel_LN.c +STRSMKERNEL_LT = ../generic/trsm_kernel_LT.c +STRSMKERNEL_RN = ../generic/trsm_kernel_RN.c +STRSMKERNEL_RT = ../generic/trsm_kernel_RT.c + +DTRSMKERNEL_LN = ../generic/trsm_kernel_LN.c +DTRSMKERNEL_LT = ../generic/trsm_kernel_LT.c +DTRSMKERNEL_RN = ../generic/trsm_kernel_RN.c +DTRSMKERNEL_RT = ../generic/trsm_kernel_RT.c + +CTRSMKERNEL_LN = ../generic/trsm_kernel_LN.c +CTRSMKERNEL_LT = ../generic/trsm_kernel_LT.c +CTRSMKERNEL_RN = ../generic/trsm_kernel_RN.c +CTRSMKERNEL_RT = ../generic/trsm_kernel_RT.c + +ZTRSMKERNEL_LN = ../generic/trsm_kernel_LN.c +ZTRSMKERNEL_LT = ../generic/trsm_kernel_LT.c +ZTRSMKERNEL_RN = ../generic/trsm_kernel_RN.c +ZTRSMKERNEL_RT = ../generic/trsm_kernel_RT.c + + + + @@ -2021,6 +2021,46 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. #endif +#if defined(ARMV5) +#define SNUMOPT 2 +#define DNUMOPT 2 + +#define GEMM_DEFAULT_OFFSET_A 0 +#define GEMM_DEFAULT_OFFSET_B 0 +#define GEMM_DEFAULT_ALIGN 0x03fffUL + +#define SGEMM_DEFAULT_UNROLL_M 2 +#define SGEMM_DEFAULT_UNROLL_N 2 + +#define DGEMM_DEFAULT_UNROLL_M 2 +#define DGEMM_DEFAULT_UNROLL_N 2 + +#define CGEMM_DEFAULT_UNROLL_M 2 +#define CGEMM_DEFAULT_UNROLL_N 2 + +#define ZGEMM_DEFAULT_UNROLL_M 2 +#define ZGEMM_DEFAULT_UNROLL_N 2 + +#define SGEMM_DEFAULT_P 128 +#define DGEMM_DEFAULT_P 128 +#define CGEMM_DEFAULT_P 96 +#define ZGEMM_DEFAULT_P 64 + +#define SGEMM_DEFAULT_Q 240 +#define DGEMM_DEFAULT_Q 120 +#define CGEMM_DEFAULT_Q 120 +#define ZGEMM_DEFAULT_Q 120 + +#define SGEMM_DEFAULT_R 12288 +#define DGEMM_DEFAULT_R 8192 +#define CGEMM_DEFAULT_R 4096 +#define ZGEMM_DEFAULT_R 4096 + + +#define SYMV_P 16 +#endif + + #ifdef GENERIC |