| @@ -18,3 +18,8 @@ ifeq ($(CORE), THUNDERX) | |||||
| CCOMMON_OPT += -mtune=thunderx -mcpu=thunderx | CCOMMON_OPT += -mtune=thunderx -mcpu=thunderx | ||||
| FCOMMON_OPT += -mtune=thunderx -mcpu=thunderx | FCOMMON_OPT += -mtune=thunderx -mcpu=thunderx | ||||
| endif | endif | ||||
| ifeq ($(CORE), THUNDERX2T99) | |||||
| CCOMMON_OPT += -mtune=vulcan -mcpu=vulcan | |||||
| FCOMMON_OPT += -mtune=vulcan -mcpu=vulcan | |||||
| endif | |||||
| @@ -82,4 +82,5 @@ ARMV8 | |||||
| CORTEXA57 | CORTEXA57 | ||||
| VULCAN | VULCAN | ||||
| THUNDERX | THUNDERX | ||||
| THUNDERX2T99 | |||||
| @@ -32,13 +32,15 @@ | |||||
| #define CPU_CORTEXA57 2 | #define CPU_CORTEXA57 2 | ||||
| #define CPU_VULCAN 3 | #define CPU_VULCAN 3 | ||||
| #define CPU_THUNDERX 4 | #define CPU_THUNDERX 4 | ||||
| #define CPU_THUNDERX2T99 5 | |||||
| static char *cpuname[] = { | static char *cpuname[] = { | ||||
| "UNKNOWN", | "UNKNOWN", | ||||
| "ARMV8" , | "ARMV8" , | ||||
| "CORTEXA57", | "CORTEXA57", | ||||
| "VULCAN", | "VULCAN", | ||||
| "THUNDERX" | |||||
| "THUNDERX", | |||||
| "THUNDERX2T99" | |||||
| }; | }; | ||||
| static char *cpuname_lower[] = { | static char *cpuname_lower[] = { | ||||
| @@ -46,7 +48,8 @@ static char *cpuname_lower[] = { | |||||
| "armv8" , | "armv8" , | ||||
| "cortexa57", | "cortexa57", | ||||
| "vulcan", | "vulcan", | ||||
| "thunderx" | |||||
| "thunderx", | |||||
| "thunderx2t99" | |||||
| }; | }; | ||||
| int get_feature(char *search) | int get_feature(char *search) | ||||
| @@ -117,6 +120,8 @@ int detect(void) | |||||
| return CPU_VULCAN; | return CPU_VULCAN; | ||||
| else if (strstr(cpu_part, "0x0a1") && strstr(cpu_implementer, "0x43")) | else if (strstr(cpu_part, "0x0a1") && strstr(cpu_implementer, "0x43")) | ||||
| return CPU_THUNDERX; | return CPU_THUNDERX; | ||||
| else if (strstr(cpu_part, "0xFFF") && strstr(cpu_implementer, "0x43")) /* TODO */ | |||||
| return CPU_THUNDERX2T99; | |||||
| } | } | ||||
| p = (char *) NULL ; | p = (char *) NULL ; | ||||
| @@ -241,6 +246,28 @@ void get_cpuconfig(void) | |||||
| printf("#define DTB_SIZE 4096\n"); | printf("#define DTB_SIZE 4096\n"); | ||||
| printf("#define L2_ASSOCIATIVE 16\n"); | printf("#define L2_ASSOCIATIVE 16\n"); | ||||
| break; | break; | ||||
| case CPU_THUNDERX2T99: | |||||
| printf("#define VULCAN \n"); | |||||
| printf("#define HAVE_VFP \n"); | |||||
| printf("#define HAVE_VFPV3 \n"); | |||||
| printf("#define HAVE_NEON \n"); | |||||
| printf("#define HAVE_VFPV4 \n"); | |||||
| printf("#define L1_CODE_SIZE 32768 \n"); | |||||
| printf("#define L1_CODE_LINESIZE 64 \n"); | |||||
| printf("#define L1_CODE_ASSOCIATIVE 8 \n"); | |||||
| printf("#define L1_DATA_SIZE 32768 \n"); | |||||
| printf("#define L1_DATA_LINESIZE 64 \n"); | |||||
| printf("#define L1_DATA_ASSOCIATIVE 8 \n"); | |||||
| printf("#define L2_SIZE 262144 \n"); | |||||
| printf("#define L2_LINESIZE 64 \n"); | |||||
| printf("#define L2_ASSOCIATIVE 8 \n"); | |||||
| printf("#define L3_SIZE 33554432 \n"); | |||||
| printf("#define L3_LINESIZE 64 \n"); | |||||
| printf("#define L3_ASSOCIATIVE 32 \n"); | |||||
| printf("#define DTB_DEFAULT_ENTRIES 64 \n"); | |||||
| printf("#define DTB_SIZE 4096 \n"); | |||||
| break; | |||||
| } | } | ||||
| } | } | ||||
| @@ -730,22 +730,22 @@ void blas_set_parameter(void){ | |||||
| #if defined(ARCH_ARM64) | #if defined(ARCH_ARM64) | ||||
| #if defined(VULCAN) | |||||
| unsigned long vulcan_pre_a; | |||||
| unsigned long vulcan_pre_b; | |||||
| unsigned long vulcan_pre_c; | |||||
| #if defined(VULCAN) || defined(THUNDERX2T99) | |||||
| unsigned long dgemm_prefetch_size_a; | |||||
| unsigned long dgemm_prefetch_size_b; | |||||
| unsigned long dgemm_prefetch_size_c; | |||||
| #endif | #endif | ||||
| void blas_set_parameter(void) | void blas_set_parameter(void) | ||||
| { | { | ||||
| #if defined(VULCAN) | |||||
| #if defined(VULCAN) || defined(THUNDERX2T99) | |||||
| dgemm_p = 160; | dgemm_p = 160; | ||||
| dgemm_q = 128; | dgemm_q = 128; | ||||
| dgemm_r = 4096; | dgemm_r = 4096; | ||||
| vulcan_pre_a = 3584; | |||||
| vulcan_pre_b = 512; | |||||
| vulcan_pre_c = 128; | |||||
| dgemm_prefetch_size_a = 3584; | |||||
| dgemm_prefetch_size_b = 512; | |||||
| dgemm_prefetch_size_c = 128; | |||||
| #endif | #endif | ||||
| } | } | ||||
| @@ -928,6 +928,23 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. | |||||
| #else | #else | ||||
| #endif | #endif | ||||
| #ifdef FORCE_THUNDERX2T99 | |||||
| #define FORCE | |||||
| #define ARCHITECTURE "ARM64" | |||||
| #define SUBARCHITECTURE "THUNDERX2T99" | |||||
| #define SUBDIRNAME "arm64" | |||||
| #define ARCHCONFIG "-DTHUNDERX2T99 " \ | |||||
| "-DL1_CODE_SIZE=32768 -DL1_CODE_LINESIZE=64 -DL1_CODE_ASSOCIATIVE=8 " \ | |||||
| "-DL1_DATA_SIZE=32768 -DL1_DATA_LINESIZE=64 -DL1_DATA_ASSOCIATIVE=8 " \ | |||||
| "-DL2_SIZE=262144 -DL2_LINESIZE=64 -DL2_ASSOCIATIVE=8 " \ | |||||
| "-DL3_SIZE=33554432 -DL3_LINESIZE=64 -DL3_ASSOCIATIVE=32 " \ | |||||
| "-DDTB_DEFAULT_ENTRIES=64 -DDTB_SIZE=4096 " \ | |||||
| "-DHAVE_VFPV4 -DHAVE_VFPV3 -DHAVE_VFP -DHAVE_NEON" | |||||
| #define LIBNAME "thunderx2t99" | |||||
| #define CORENAME "THUNDERX2T99" | |||||
| #else | |||||
| #endif | |||||
| #ifndef FORCE | #ifndef FORCE | ||||
| #if defined(__powerpc__) || defined(__powerpc) || defined(powerpc) || \ | #if defined(__powerpc__) || defined(__powerpc) || defined(powerpc) || \ | ||||
| @@ -0,0 +1,2 @@ | |||||
| include $(KERNELDIR)/KERNEL.VULCAN | |||||
| @@ -973,11 +973,11 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. | |||||
| prfm PLDL1KEEP, [origPA] | prfm PLDL1KEEP, [origPA] | ||||
| ldr A_PRE_SIZE, =vulcan_pre_a | |||||
| ldr A_PRE_SIZE, =dgemm_prefetch_size_a | |||||
| ldr A_PRE_SIZE, [A_PRE_SIZE] | ldr A_PRE_SIZE, [A_PRE_SIZE] | ||||
| ldr B_PRE_SIZE, =vulcan_pre_b | |||||
| ldr B_PRE_SIZE, =dgemm_prefetch_size_b | |||||
| ldr B_PRE_SIZE, [B_PRE_SIZE] | ldr B_PRE_SIZE, [B_PRE_SIZE] | ||||
| ldr C_PRE_SIZE, =vulcan_pre_c | |||||
| ldr C_PRE_SIZE, =dgemm_prefetch_size_c | |||||
| ldr C_PRE_SIZE, [C_PRE_SIZE] | ldr C_PRE_SIZE, [C_PRE_SIZE] | ||||
| add A_PRE_SIZE_64, A_PRE_SIZE, #64 | add A_PRE_SIZE_64, A_PRE_SIZE, #64 | ||||
| add B_PRE_SIZE_64, B_PRE_SIZE, #64 | add B_PRE_SIZE_64, B_PRE_SIZE, #64 | ||||
| @@ -2459,6 +2459,44 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. | |||||
| #define ZGEMM_DEFAULT_R 4096 | #define ZGEMM_DEFAULT_R 4096 | ||||
| #define SYMV_P 16 | |||||
| #endif | |||||
| #if defined(THUNDERX2T99) | |||||
| #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 16 | |||||
| #define SGEMM_DEFAULT_UNROLL_N 4 | |||||
| #define DGEMM_DEFAULT_UNROLL_M 8 | |||||
| #define DGEMM_DEFAULT_UNROLL_N 4 | |||||
| #define CGEMM_DEFAULT_UNROLL_M 8 | |||||
| #define CGEMM_DEFAULT_UNROLL_N 4 | |||||
| #define ZGEMM_DEFAULT_UNROLL_M 4 | |||||
| #define ZGEMM_DEFAULT_UNROLL_N 4 | |||||
| #define SGEMM_DEFAULT_P 512 | |||||
| #define DGEMM_DEFAULT_P dgemm_p | |||||
| #define CGEMM_DEFAULT_P 256 | |||||
| #define ZGEMM_DEFAULT_P 128 | |||||
| #define SGEMM_DEFAULT_Q 1024 | |||||
| #define DGEMM_DEFAULT_Q dgemm_q | |||||
| #define CGEMM_DEFAULT_Q 512 | |||||
| #define ZGEMM_DEFAULT_Q 512 | |||||
| #define SGEMM_DEFAULT_R 4096 | |||||
| #define DGEMM_DEFAULT_R dgemm_r | |||||
| #define CGEMM_DEFAULT_R 4096 | |||||
| #define ZGEMM_DEFAULT_R 2048 | |||||
| #define SYMV_P 16 | #define SYMV_P 16 | ||||
| #endif | #endif | ||||