Compare commits

...

139 Commits
riscv ... i280

Author SHA1 Message Date
Dr-Noob
51dd89b005 WIP 2024-09-17 08:52:01 +01:00
Dr-Noob
ef1bfa5cc9 WIP 2024-09-17 08:49:51 +01:00
Dr-Noob
2e2e660b97 [v1.06][ARM] Add more NVIDIA SoCs 2024-09-12 08:08:53 +01:00
Dr-Noob
fbd50822cf [v1.06][ARM] Add more Marvell SoCs 2024-09-12 07:51:12 +01:00
Dr-Noob
bc8a779de6 [v1.06][ARM] Add more Amlogic SoCs 2024-09-11 18:38:56 +01:00
Dr-Noob
995ebc6736 [v1.06][RISCV] Add zicbop multi-letter extension (fixes #285) 2024-09-11 07:58:12 +01:00
Dr-Noob
ab43a11ef2 [v1.06][X86] Fix accurate-pp in hybrid architectures (fixes #169)
Overview of changes:
- Adds field max_pp in frequency struct to hold the max freq for peak-performance estimation.
- Instead of getting the max frequency in get_peak_performance, we get it in get_cpu_info (more natural).
- Adds fill_frequency_info_pp which fills the max_pp of the passed cpu by calling measure_frequency.

The approach is to call measure_frequency with a vector where the max frequencies are stored. Then,
the first time measure_frequency is called, the frequency is measured while running all the cores,
and the max frequency is computed per module (e.g., in the case of 2 modules, we would compute
the freq for the first and for the second module), and saved into this vector. Subsequent calls to
measure_frequency will just read the corresponding value for the vector. In other words, the frequency
is only measured once for the whole CPU.
2024-09-10 22:43:23 +01:00
Wunk
edbfc9722e [v1.06][ARM] Add Windows on Arm support (#273) 2024-09-10 09:40:46 +02:00
Dr-Noob
57bbe2de4f [v1.06][X86] Add Sapphire Rapids uarch (#281) 2024-09-10 08:32:18 +01:00
Dr-Noob
278efb75c9 [v1.06][ARM] Add support for Marvell SoC (#279) 2024-09-10 07:41:11 +01:00
Dr-Noob
343150e516 [v1.06][ARM] Add Tegra Orin (#275) 2024-09-09 08:19:18 +01:00
Dr-Noob
1e2c7e565c [v1.06][ARM] Add SC8280XP (on device tree) (#272) 2024-09-09 07:11:57 +01:00
Dr-Noob
977c35a9af [v1.06][X86] Add Zen5 uarch 2024-09-05 20:14:52 +01:00
Dr-Noob
eb8fad2843 [v1.06][ARM] Simplify is_ARMv8_or_newer 2024-09-02 08:27:58 +01:00
Dr-Noob
bd38951439 [v1.06][ARM] Update get_vpus_width to match SVE detection 2024-09-02 08:26:58 +01:00
Dr-Noob
057a36efd5 [v1.06][ARM] Add new microarchitectures 2024-09-02 08:20:40 +01:00
Dr-Noob
56901d70ab [v1.06][ARM] Add more NXP SoCs 2024-08-31 18:40:35 +01:00
Dr-Noob
5bd507e4b6 [v1.06][ARM] Add support for Amlogic A311D (#268) 2024-08-31 09:37:50 +01:00
Dr-Noob
9192ba3eb8 [v1.06][ARM] Add (another) cpufamily for M3 and set the ARM version, fixing peak performance wrong result (#230) 2024-08-29 08:13:05 +01:00
Dr-Noob
807a13d29e [v1.06] Fix manpage file extension (fixes #269) 2024-08-28 21:33:48 +01:00
Dr-Noob
15a803dae5 [v1.06][ARM] Check if HWCAP2_SVE2 is defined before using it (fixes #270) 2024-08-28 21:26:27 +01:00
Dr-Noob
0507355372 [v1.06][X86] Bugfix: set affinity in --accurate-pp
There are cases where measure_frequency is called after binding the
process to a specific core via bind_to_cpu (e.g., when iterating over
modules in hybrid architectures). Thus, in measure_frequency we must
set the affinity of the newly created threads, ensuring they are
binded to the right core.
2024-08-26 12:31:18 +01:00
Dr-Noob
9f66a137c5 [v1.06][ARM] Add support for NXP i.MX 8M Plus (#261) 2024-08-25 17:44:29 +01:00
Dr-Noob
324d0fbe94 [v1.06][PPC] Add support for Espresso (#231) 2024-08-25 16:50:58 +01:00
Dr-Noob
af8bb16302 [v1.06][ARM] Fix for commit 025e28c 2024-08-24 15:44:06 +01:00
Dr-Noob
40374121b8 [v1.06] Replace emalloc+memset with ecalloc when possible. Refactor some memory allocation code 2024-08-24 15:32:33 +01:00
Dr-Noob
025e28c516 [v1.06][ARM] Set Android properties as defines 2024-08-24 14:47:34 +01:00
Dr-Noob
7ad19d113c [v1.06] Remove soc_ prefix from field names in system_on_chip struct 2024-08-24 12:44:24 +01:00
Dr-Noob
13605ed0ce [v1.06][ARM] Remove BCM prefix from Broadcom vendor string 2024-08-24 12:34:57 +01:00
Dr-Noob
7689355a72 [v1.06][ARM] Add support for Apple SoCs in Asahi Linux (#263) 2024-08-23 08:37:13 +01:00
Timothy Warren
321a1ec375 [v1.06][X86] Add old Intel and AMD CPUs (#260)
* Add Intel and AMD 486, Intel Pentium Pro, and AMD K5 chips
* Update infer_cpu_name_from_uarch to support added CPU uarches
* Add sources for more of the legacy chip data
* Add the more specific codename for Intel Pentium MMX Mobile chips
* Document source of AMD 486/K5 variants
* Add link to documentation for Intel 486

---------

Co-authored-by: Timothy Warren <tim@timshomepage.net>
2024-08-20 09:25:01 +02:00
Dr-Noob
48c598cf3b [v1.06][ARM] Add support for Ampere Altra (#262) 2024-08-19 08:46:07 +01:00
Dr-Noob
aa94389bbe [v1.06][ARM] Fix two off-by-one bugs (#264) 2024-08-19 08:39:52 +01:00
Er2
8d10a03adc [v1.06][FreeBSD][Apple] Implement fallback frequency calculation (#251) 2024-08-19 09:21:41 +02:00
Dr-Noob
2410fd16d3 [v1.06] Bump version 2024-08-18 15:12:27 +01:00
Dr-Noob
f63178b41c [v1.05][ARM] Fix compile warning in SVE cntb 2024-08-10 11:23:13 +01:00
Dr-Noob
2788e6831e [v1.05][ARM] Fix SVE cntb datatype 2024-08-10 11:16:22 +01:00
Dr-Noob
146f2a13aa [v1.05][ARM] Show SVE cntb in debug 2024-08-10 11:11:46 +01:00
Dr-Noob
90624b9aaa [v1.05][ARM] Preeliminary support for SVE detection (#259) 2024-08-10 11:08:39 +01:00
Dr-Noob
e42f04cca8 [v1.05][ARM] Add support for Kirin 9000S and TSV120 (#259) 2024-08-10 11:03:22 +01:00
Dr-Noob
3aacaf5f9e [v1.05] Add RISC-V picture to README (thanks #241!) 2024-08-03 16:31:56 +01:00
Dr-Noob
50f66ec571 [v1.05][X86] Fix compilation issue in clang and fallback to get_topology_from_udev also in AMD 2024-08-03 15:45:10 +01:00
Dr-Noob
cb186a2f97 [v1.05][X86] Do not show empty SSE if not supported (#260) 2024-08-02 10:09:51 +01:00
Dr-Noob
6164884415 [v1.05][ARM] Add latest Snapdragon SoCs (closes #256) 2024-08-02 08:07:47 +01:00
Dr-Noob
260f9ec3b8 [v1.05] Add new contributor to acknowledgements 2024-07-25 08:29:05 +01:00
Dr-Noob
79013d0ec9 [v1.05][X86] Improve robustness in case of old CPUs (verified by #220)
- Use udev to get topo when apic failed or is not available.
- Assume single core in udev when total cores is 1.
- Print core/cores accordingly to the number of cores.
2024-07-25 08:23:56 +01:00
Dr-Noob
e4227388b9 [v1.05][ARM] Add new SnapD SoCs and use ro.soc.mode in Android to improve SoC detection (#253) 2024-07-19 08:21:11 +01:00
Dr-Noob
8fca4cb250 [v1.05][X86] Fix frequency fetching from udev and measurement in hybrid architectures 2024-07-11 22:27:42 +01:00
Dr-Noob
7c947bdf64 [v1.05] Use UNKNOWN_DATA instead of -1 in frequency measurement 2024-07-11 22:10:00 +01:00
Dr-Noob
1ed3a0f2bf [v1.05] Adapt frequency measurement iterations depending on CPU speed
This is achieved by running a first measurement which gets a taste of
CPU speed, then estimate a reasonable value for the iterations of the
real measurement and then running the actual measurement. This is very
helpful to reduce the runtime of the measurement, especially for slow
CPUs
2024-07-11 21:55:01 +01:00
Dr-Noob
0fe6fc3f4d [v1.05][ARM] Fix mistake in Makefile 2024-07-11 08:17:26 +01:00
Dr-Noob
96c784026b [v1.05] Fix formatting issues in Makefile 2024-07-09 08:47:34 +01:00
Dr-Noob
59cd2dd128 [v1.05][X86] Add support for Hygon CPUs (#244) 2024-07-09 08:34:44 +01:00
Dr-Noob
da1981b97c [v1.05] Check read return value in frequency measurement 2024-07-09 08:32:24 +01:00
Dr-Noob
8506c91e00 [v1.05] Add --measure-max-freq (available on x86 and ARM) 2024-07-08 09:07:57 +01:00
Dr-Noob
ece28cbdee [v1.05] Small fix in help message 2024-07-08 08:31:00 +01:00
Dr-Noob
7b46c78249 [v1.05] Replace printf with proper printErr 2024-07-08 08:28:48 +01:00
Dr-Noob
e0095c303d [v1.05] Print a tilde in case the freq was measured (indicating that this value is an approximation) 2024-07-08 08:23:56 +01:00
Dr-Noob
65378aaed9 [v1.05] Move sysctl from ARM-specific to common (#251) 2024-07-07 12:43:03 +01:00
Dr-Noob
946729dd06 [v1.05] Compile measure freq only in Linux (avoids compiler warning in non-linux builds) 2024-07-06 11:28:46 +01:00
Dr-Noob
9212f19de1 [v1.05] Add support for frequency measurement (both x86 and ARM) (branch measure-freq #220) 2024-07-05 08:44:34 +01:00
Dr-Noob
b019256515 [v1.05] Continue merging measure-freq #220
- [v1.05][X86] Show SSE if AVX/FMA is not supported
- [v1.05][X86] Do not stop if cach is NULL and check for non-NULL cache in get_topology_info functions
- [v1.05][X86] Fix bug where the number of cpus were not set if NULL was returned inside the loop. Ensure topo is not NULL in get_peak_performance. Fallback to UNKNOWN_DATA when we have no information about topology
2024-07-05 08:37:54 +01:00
Dr-Noob
d4cadbd807 [v1.05] Move bind_to_cpu from x86-specific to global (merging measure-freq #220) 2024-07-05 08:32:11 +01:00
Dr-Noob
4f081ef1a2 [v1.05] Add newline to fix dummy warning with clang 2024-07-03 09:09:37 +02:00
Dr-Noob
1b746bc67d [v1.05][ARM] Add support to detect SoC from PCI (#245) with initial support for NVIDIA Tegra 2024-07-03 08:01:53 +01:00
Dr-Noob
dfa2b773d1 [v1.05][X86] Add new Zen4 uarch (#237) 2024-05-18 22:36:23 +01:00
Dr-Noob
59efbf4e08 [v1.05] Add VirtualBox hypervisor (noted in #235) 2024-05-18 22:27:17 +01:00
Dr-Noob
29768e841d [v1.05][ARM] Update TaiShan vector units 2024-02-20 09:35:30 +01:00
Dr-Noob
cc16bc56ef [v1.05][X86] Improve support for old x86 CPUs (#220) by improving Pentium uarch detection and adding support to infer the CPU name from uarch when the corresponding CPUID level is not available 2024-02-13 21:29:08 +00:00
Dr-Noob
1504c5d0ef [v1.05][X86] Report x86 / x86_64 in ARCH_X86 builds 2024-02-13 21:26:24 +00:00
Dr-Noob
08919916dc [v1.05] Always print manufacturing process in nanometers (this indirectly fixes a bug of processes in micrometers being incorrectly reported by 10x 2024-02-13 21:24:16 +00:00
Dr-Noob
4a8a7567f0 [v1.05] Add support for NO_COLOR (#227) 2024-02-13 08:41:54 +00:00
Dr-Noob
c01f60fa6c [v1.05] Implement new approach to infer SoC from the uarch. Add support for Kunpeng SoCs 2024-02-08 09:04:51 +00:00
Dr-Noob
b610dc8c7d [v1.05] Bump version 2024-02-05 18:33:15 +00:00
Dr-Noob
0967e3597b [v1.04] Fix error in previous commit 2024-02-05 09:48:46 +01:00
Dr-Noob
14df701707 [v1.04] Check release when printing error also in Apple chip failures 2024-02-05 08:46:17 +00:00
Dr-Noob
cbcce9c2ed [v1.04] Improve error reporting in the case of unkown microarchitectures. As pointed out in #224 this was already solved, but this commit tries to be an improvement over the previous implementation 2024-02-05 08:37:59 +00:00
Dr-Noob
b072e5c331 [v1.04] Update acknowledgements 2024-02-01 22:40:58 +00:00
Dr-Noob
4b42d2c89d [v1.04][ARM] Extend Allwinner SoC list (#207) 2024-02-01 23:26:23 +01:00
Dr-Noob
d221f578b7 [v1.04][ARM] Use sysctl to fetch topology in Apple SoCs 2024-01-25 09:33:22 +01:00
Dr-Noob
7b4b94bb93 [v1.04] Fix typo in Rockchip SoC detection 2024-01-22 07:18:15 +00:00
Dr-Noob
5e00d554aa [v1.04][ARM] Fix number of cores per cluster in M2 and M3 SoCs (#183) 2024-01-17 08:48:58 +00:00
Dr-Noob
5ffac60622 [v1.04][ARM] Support for reverse ordering in Rockchip SoCs 2024-01-17 09:37:15 +01:00
Dr-Noob
6566ac3ccf [v1.04] Add Apple VZ hypervisor (#218) 2024-01-15 08:43:28 +00:00
Dr-Noob
0a048717bf [v1.04][X86] Fix bug in AMD CPUs (#212) 2023-12-24 23:03:20 +00:00
Dr-Noob
ce6e6f9cd8 [v1.04][ARM] Preeliminary M3 support 2023-12-24 22:42:01 +00:00
Dr-Noob
2feee25659 [v1.04] Clarify how the unknown microarchitectures should be reported 2023-12-03 20:23:15 +00:00
Dr-Noob
8335bd004d [v1.04][X86] Add AMD uarchs 2023-12-02 18:18:43 +00:00
Dr-Noob
a172b4b281 [v1.04][ARM] Add more Mediatek SoCs 2023-11-18 20:25:57 +01:00
Dr-Noob
d88a6bf763 [v1.04][X86] Fixed bhyve matcher (#193) 2023-11-11 09:36:12 +00:00
Dr-Noob
8870a0ca0c [v1.04][RISCV] Add Sipeed ASCII logo 2023-11-01 20:50:05 +00:00
Dr-Noob
a43b700540 [v1.04][RISCV] Add Lichee PI 4a SoC (#200) 2023-11-01 20:48:25 +00:00
Dr-Noob
9a578d3670 [v1.04][RISCV] Make single-letter extension detection more robust. s and u are invalid but might appear (#200), so just skip them 2023-10-29 17:14:47 +00:00
Dr-Noob
bf890965a3 [v1.04] Add colors to Google logo 2023-10-17 09:19:05 +02:00
Dr-Noob
530207aee0 [v1.04] Added Google ascii logo 2023-10-17 08:56:20 +02:00
Dr-Noob
f64ff0d4cf [v1.04] Added Google Tensor SoCs 2023-10-17 08:55:38 +02:00
Dr-Noob
4b7600d49e [v1.04][ARM] Add Cortex-A715/X3 uarchs 2023-10-16 09:04:59 +02:00
Dr-Noob
d8ce2de287 [v1.04] Testing bhyve matcher 2023-10-03 21:00:43 +01:00
Dr-Noob
bc9b9c1e99 [1.04][ARM] Add support for Raspberry Pi 5 2023-10-01 11:56:36 +02:00
Dr-Noob
b17df01958 [1.04][ARM] Improve robustness in Raspberry Pi detection 2023-10-01 11:49:20 +02:00
Dr-Noob
44e5e1b6dc [v1.04][RISCV] Add support for multi-letter extensions 2023-09-25 07:27:13 +01:00
Dr-Noob
fe4bcde898 [v1.04][X86] Added new uarchs 2023-09-24 12:58:56 +01:00
Dr-Noob
45854a00d6 [v1.04] Add bhyve hypervisor 2023-09-24 12:10:39 +01:00
Dr-Noob
6c69ab0a00 [v1.04][RISCV] Extend extension mask to 64 bit. Fix SET_ISA_EXT_MAP else condition. Fix print_ascii_riscv iters computation 2023-09-24 11:59:43 +01:00
Dr-Noob
bd3c66395d [v1.04][RISCV] Add multi-letter extensions to list 2023-09-22 21:39:17 +01:00
Dr-Noob
a12defaf4b [v1.04][RISCV] Define to enum 2023-09-19 08:18:13 +01:00
Dr-Noob
4fa3dc2076 [v1.04][RISCV] First support for parsing multi-letter extensions 2023-09-18 08:20:32 +01:00
Dr-Noob
bcf8b8171d [v1.04][RISCV] Temporary fix to parse multi-letter extensions 2023-09-17 23:14:49 +01:00
Dr-Noob
7d81d895b0 [v1.04][ARM] Add support for M2 Max/Ultra 2023-09-02 21:08:43 +01:00
Dr-Noob
6a5b4c3da4 [v1.04][ARM] Fix bug in socs.h for Apple M2 2023-09-02 20:53:53 +01:00
Dr-Noob
5f5561f177 [v1.04][ARM] Add Rockchip 3588S SoC (thanks #188!) 2023-09-02 19:02:00 +01:00
Dr-Noob
9a46e310e4 [v1.04][X86] Add additional Zen4 cpuid 2023-09-02 18:48:22 +01:00
Dr-Noob
36071b8eb8 [v1.04][ARM] Add support for M2 Pro 2023-06-09 17:47:05 +02:00
Dr-Noob
6349f4435e [v1.04][X86] Add mobile rocket lake chip 2023-06-06 21:44:27 +02:00
Dr-Noob
bddf6cfe31 [v1.04][PPC] Small fix for jump or move depends on uninitialised value (noted in #178) 2023-05-09 10:58:16 +02:00
Dr-Noob
53e5dd19f7 [v1.04][ARM] Experimental new backend for computing peak performance, considering the number of VPUs and width (previously this was highly simplified) 2023-05-06 15:36:00 +02:00
Dr-Noob
b3719dc216 [v1.04] Bump version 2023-05-06 11:37:32 +02:00
Dr-Noob
7a3fd539c3 [v1.03][X86] Fix compilation warning under Windows 2023-05-06 11:30:14 +02:00
Dr-Noob
f6bf00e33d [v1.03][X86] Refactor --raw. Add support for printing HV level 2023-05-06 11:14:46 +02:00
Dr-Noob
985bd6f25f [v1.03][X86] Do not consider an empty HV as a bug 2023-05-06 11:02:37 +02:00
Dr-Noob
f599613f5e [v1.03] Reuse same _PATH_CPUINFO definition 2023-05-06 10:19:39 +02:00
Dr-Noob
8a8b1dffa8 [v1.03] Apply refactoring in src/common/udev.c 2023-05-06 10:06:16 +02:00
Dr-Noob
8e95828fb1 [v1.03][PPC] Read frequency from cpuinfo if the default method fails 2023-05-04 19:01:55 +02:00
Dr-Noob
93f733cb12 [v1.03][PPC] Small update in PowerPC uarch table 2023-05-04 18:59:03 +02:00
Dr-Noob
0397db2fd8 [v1.03][PPC] Implement hypervisor detection in PowerPC 2023-05-04 18:56:31 +02:00
Dr-Noob
ebe7312c9d [v1.03][PPC] Fix bug where /sys/devices/system/cpu/cpu*/topology/physical_package_id contains -1 by adding a new way of finding the number of sockets. This happened in #178 and most probably in #153 too 2023-05-03 17:51:44 +02:00
Dr-Noob
8c72295f47 [v1.03] Add PPC integrity check 2023-04-27 22:35:56 +02:00
Dr-Noob
bec47a44a7 [v1.03] Update README (PPC/macOS support was proved in #167) 2023-04-24 17:58:58 +02:00
Dr-Noob
de638aba50 [v1.03][x86] Fix Zen4 peak performance calculation 2023-04-24 17:47:42 +02:00
Dr-Noob
5be6f8bba0 [v1.03][x86] Add AMD 9004 Series CPUID 2023-04-22 08:20:35 +02:00
Dr-Noob
10d4f67cdb [v1.03] Fixing commit 99fd108 2023-04-13 17:50:49 +02:00
Dr-Noob
5dd6ab60cb [v1.03] Lets follow the function prototype rules 2023-04-13 17:45:32 +02:00
Dr-Noob
99fd108d0f [v1.03] Try detecting SoC with nvmem in last place (avoids usless work in Android) 2023-04-13 17:41:32 +02:00
Dr-Noob
f28f3ef70d [v1.03] Avoid segmentation fault when a Rockchip SoC is not found in socs.h 2023-04-13 17:37:14 +02:00
Dr-Noob
89aadeead8 [v1.03] Add common soc.c/.h for both ARM and RISC-V 2023-04-13 17:15:32 +02:00
Dr-Noob
f58d76ccac [v1.03] Update help option and perform some formatting adjustments 2023-04-13 16:24:47 +02:00
Dr-Noob
889fbf2d67 [v1.03][RISCV] Fix bug in Makefile and implement --debug option 2023-04-13 15:49:06 +02:00
Dr-Noob
91cc04c653 [v1.03][RISCV] Merge RISC-V branch into main 2023-04-13 15:28:42 +02:00
Dr-Noob
2b4e675800 [v1.03][ARM] Add more Rockchip SoCs (fix) 2023-04-03 15:10:58 +02:00
51 changed files with 3668 additions and 1099 deletions

View File

@@ -1,7 +1,7 @@
CC ?= gcc
CFLAGS+=-Wall -Wextra -pedantic
SANITY_FLAGS=-Wfloat-equal -Wshadow -Wpointer-arith
SANITY_FLAGS=-Wfloat-equal -Wshadow -Wpointer-arith -Wstrict-prototypes
PREFIX ?= /usr
@@ -13,17 +13,27 @@ COMMON_HDR = $(SRC_COMMON)ascii.h $(SRC_COMMON)cpu.h $(SRC_COMMON)udev.h $(SRC_C
ifneq ($(OS),Windows_NT)
GIT_VERSION := "$(shell git describe --abbrev=4 --dirty --always --tags)"
arch := $(shell uname -m)
os := $(shell uname -s)
ifeq ($(os), Linux)
COMMON_SRC += $(SRC_COMMON)freq.c
COMMON_HDR += $(SRC_COMMON)freq.h
endif
ifeq ($(arch), $(filter $(arch), x86_64 amd64 i386 i486 i586 i686))
SRC_DIR=src/x86/
SOURCE += $(COMMON_SRC) $(SRC_DIR)cpuid.c $(SRC_DIR)apic.c $(SRC_DIR)cpuid_asm.c $(SRC_DIR)uarch.c
HEADERS += $(COMMON_HDR) $(SRC_DIR)cpuid.h $(SRC_DIR)apic.h $(SRC_DIR)cpuid_asm.h $(SRC_DIR)uarch.h $(SRC_DIR)freq/freq.h
os := $(shell uname -s)
ifeq ($(os), Linux)
SOURCE += $(SRC_DIR)freq/freq.c freq_nov.o freq_avx.o freq_avx512.o
HEADERS += $(SRC_DIR)freq/freq.h
CFLAGS += -pthread
endif
ifeq ($(os), FreeBSD)
SOURCE += $(SRC_COMMON)sysctl.c
HEADERS += $(SRC_COMMON)sysctl.h
endif
CFLAGS += -DARCH_X86 -std=c99 -fstack-protector-all
else ifeq ($(arch), $(filter $(arch), ppc64le ppc64 ppcle ppc))
SRC_DIR=src/ppc/
@@ -32,19 +42,24 @@ ifneq ($(OS),Windows_NT)
CFLAGS += -DARCH_PPC -std=gnu99 -fstack-protector-all -Wno-language-extension-token
else ifeq ($(arch), $(filter $(arch), arm aarch64_be aarch64 arm64 armv8b armv8l armv7l armv6l))
SRC_DIR=src/arm/
SOURCE += $(COMMON_SRC) $(SRC_DIR)midr.c $(SRC_DIR)uarch.c $(SRC_DIR)soc.c $(SRC_DIR)udev.c
HEADERS += $(COMMON_HDR) $(SRC_DIR)midr.h $(SRC_DIR)uarch.h $(SRC_DIR)soc.h $(SRC_DIR)udev.c $(SRC_DIR)socs.h
SOURCE += $(COMMON_SRC) $(SRC_DIR)midr.c $(SRC_DIR)uarch.c $(SRC_COMMON)soc.c $(SRC_DIR)soc.c $(SRC_COMMON)pci.c $(SRC_DIR)udev.c sve.o
HEADERS += $(COMMON_HDR) $(SRC_DIR)midr.h $(SRC_DIR)uarch.h $(SRC_COMMON)soc.h $(SRC_DIR)soc.h $(SRC_COMMON)pci.h $(SRC_DIR)udev.c $(SRC_DIR)socs.h
CFLAGS += -DARCH_ARM -Wno-unused-parameter -std=c99 -fstack-protector-all
os := $(shell uname -s)
# Check if the compiler supports -march=armv8-a+sve. We will use it (if supported) to compile SVE detection code later
is_sve_flag_supported := $(shell $(CC) -march=armv8-a+sve -c $(SRC_DIR)sve.c -o sve_test.o 2> /dev/null && echo 'yes'; rm -f sve_test.o)
ifeq ($(is_sve_flag_supported), yes)
SVE_FLAGS += -march=armv8-a+sve
endif
ifeq ($(os), Darwin)
SOURCE += $(SRC_DIR)sysctl.c
HEADERS += $(SRC_DIR)sysctl.h
SOURCE += $(SRC_COMMON)sysctl.c
HEADERS += $(SRC_COMMON)sysctl.h
endif
else ifeq ($(arch), $(filter $(arch), riscv64 riscv32))
SRC_DIR=src/riscv/
SOURCE += $(COMMON_SRC) $(SRC_DIR)riscv.c $(SRC_DIR)uarch.c $(SRC_DIR)soc.c $(SRC_DIR)udev.c
HEADERS += $(COMMON_SRC) $(SRC_DIR)riscv.h $(SRC_DIR)uarch.h $(SRC_DIR)soc.h $(SRC_DIR)udev.h $(SRC_DIR)socs.h
SOURCE += $(COMMON_SRC) $(SRC_DIR)riscv.c $(SRC_DIR)uarch.c $(SRC_COMMON)soc.c $(SRC_DIR)soc.c $(SRC_DIR)udev.c
HEADERS += $(COMMON_HDR) $(SRC_DIR)riscv.h $(SRC_DIR)uarch.h $(SRC_COMMON)soc.h $(SRC_DIR)soc.h $(SRC_DIR)udev.h $(SRC_DIR)socs.h
CFLAGS += -DARCH_RISCV -Wno-unused-parameter -std=c99 -fstack-protector-all
else
# Error lines should not be tabulated because Makefile complains about it
@@ -55,12 +70,27 @@ $(error Aborting compilation)
OUTPUT=cpufetch
else
# Assume x86_64
GIT_VERSION := ""
arch := $(shell cc -dumpmachine)
arch := $(firstword $(subst -, ,$(arch)))
ifeq ($(arch), $(filter $(arch), x86_64 amd64 i386 i486 i586 i686))
SRC_DIR=src/x86/
SOURCE += $(COMMON_SRC) $(SRC_DIR)cpuid.c $(SRC_DIR)apic.c $(SRC_DIR)cpuid_asm.c $(SRC_DIR)uarch.c
HEADERS += $(COMMON_HDR) $(SRC_DIR)cpuid.h $(SRC_DIR)apic.h $(SRC_DIR)cpuid_asm.h $(SRC_DIR)uarch.h
CFLAGS += -DARCH_X86 -std=c99
else ifeq ($(arch), $(filter $(arch), arm aarch64_be aarch64 arm64 armv8b armv8l armv7l armv6l))
SRC_DIR=src/arm/
SOURCE += $(COMMON_SRC) $(SRC_DIR)midr.c $(SRC_DIR)uarch.c $(SRC_COMMON)soc.c $(SRC_DIR)soc.c $(SRC_COMMON)pci.c $(SRC_DIR)udev.c sve.o
HEADERS += $(COMMON_HDR) $(SRC_DIR)midr.h $(SRC_DIR)uarch.h $(SRC_COMMON)soc.h $(SRC_DIR)soc.h $(SRC_COMMON)pci.h $(SRC_DIR)udev.c $(SRC_DIR)socs.h
CFLAGS += -DARCH_ARM -std=c99
else
# Error lines should not be tabulated because Makefile complains about it
$(warning Unsupported arch detected: $(arch). See https://github.com/Dr-Noob/cpufetch#1-support)
$(warning If your architecture is supported but the compilation fails, please open an issue in https://github.com/Dr-Noob/cpufetch/issues)
$(error Aborting compilation)
endif
GIT_VERSION := ""
SANITY_FLAGS += -Wno-pedantic-ms-format
OUTPUT=cpufetch.exe
endif
@@ -86,6 +116,9 @@ freq_avx.o: Makefile $(SRC_DIR)freq/freq_avx.c $(SRC_DIR)freq/freq_avx.h $(SRC_D
freq_avx512.o: Makefile $(SRC_DIR)freq/freq_avx512.c $(SRC_DIR)freq/freq_avx512.h $(SRC_DIR)freq/freq.h
$(CC) $(CFLAGS) $(SANITY_FLAGS) -c -mavx512f -pthread $(SRC_DIR)freq/freq_avx512.c -o $@
sve.o: Makefile $(SRC_DIR)sve.c $(SRC_DIR)sve.h
$(CC) $(CFLAGS) $(SANITY_FLAGS) $(SVE_FLAGS) -c $(SRC_DIR)sve.c -o $@
$(OUTPUT): Makefile $(SOURCE) $(HEADERS)
ifeq ($(GIT_VERSION),"")
$(CC) $(CFLAGS) $(SANITY_FLAGS) $(SOURCE) -o $(OUTPUT)
@@ -102,9 +135,9 @@ clean:
install: $(OUTPUT)
install -Dm755 "cpufetch" "$(DESTDIR)$(PREFIX)/bin/cpufetch"
install -Dm644 "LICENSE" "$(DESTDIR)$(PREFIX)/share/licenses/cpufetch-git/LICENSE"
install -Dm644 "cpufetch.1" "$(DESTDIR)$(PREFIX)/share/man/man1/cpufetch.1.gz"
install -Dm644 "cpufetch.1" "$(DESTDIR)$(PREFIX)/share/man/man1/cpufetch.1"
uninstall:
rm -f "$(DESTDIR)$(PREFIX)/bin/cpufetch"
rm -f "$(DESTDIR)$(PREFIX)/share/licenses/cpufetch-git/LICENSE"
rm -f "$(DESTDIR)$(PREFIX)/share/man/man1/cpufetch.1.gz"
rm -f "$(DESTDIR)$(PREFIX)/share/man/man1/cpufetch.1"

View File

@@ -45,11 +45,14 @@ cpufetch is a command-line tool written in C that displays the CPU information i
- [3.1 x86_64](#31-x86_64)
- [3.2 ARM](#32-arm)
- [3.3 PowerPC](#33-powerpc)
- [3.4 RISC-V](#34-risc-v)
- [4. Colors](#4-colors)
- [4.1 Specifying a name](#41-specifying-a-name)
- [4.2 Specifying the colors in RGB format](#42-specifying-the-colors-in-rgb-format)
- [5. Implementation](#5-implementation)
- [6. Bugs or improvements](#6-bugs-or-improvements)
- [6.1 Unknown microarchitecture error](#61-unknown-microarchitecture-error)
- [6.2 Other situations](#62-other-situations)
- [7. Acknowledgements](#7-acknowledgements)
- [8. cpufetch for GPUs (gpufetch)](#8-cpufetch-for-gpus-gpufetch)
@@ -57,17 +60,17 @@ cpufetch is a command-line tool written in C that displays the CPU information i
## 1. Support
| OS | x86_64 / x86 | ARM | PowerPC |
|:-----------:|:------------------:|:------------------:|:------------------:|
| GNU / Linux | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: |
| Windows | :heavy_check_mark: | :x: | :x: |
| Android | :heavy_check_mark: | :heavy_check_mark: | :x: |
| macOS | :heavy_check_mark: | :heavy_check_mark: | :x: |
| FreeBSD | :heavy_check_mark: | :x: | :x: |
| OS | x86_64 / x86 | ARM | RISC-V | PowerPC |
|:-----------:|:------------------:|:------------------:|:------------------:|:------------------:|
| GNU / Linux | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: | :heavy_check_mark: |
| Windows | :heavy_check_mark: | :heavy_check_mark: | :x: | :x: |
| Android | :heavy_check_mark: | :heavy_check_mark: | :x: | :x: |
| macOS | :heavy_check_mark: | :heavy_check_mark: | :x: | :heavy_check_mark: |
| FreeBSD | :heavy_check_mark: | :x: | :x: | :x: |
**NOTES:**
- Colors will be used in Windows only if the terminal supports it.
- Support in macOS ARM is limited to Apple M1 only
- Support in macOS ARM is limited to Apple chips only
## 2. Installation
### 2.1 Installing from a package
@@ -118,6 +121,11 @@ make
<p align="center"><img width=90% src="pictures/ibm.png"></p>
<p align="center">Talos II</p>
## 3.4 RISC-V
<p align="center"><img width=80% src="pictures/starfive.png"></p>
<p align="center">StarFive VisionFive 2</p>
## 4. Colors
By default, `cpufetch` will print the CPU logo with the system colorscheme. However, you can set a custom color scheme in two different ways:
@@ -147,15 +155,33 @@ By default, `cpufetch` will print the CPU logo with the system colorscheme. Howe
See [cpufetch programming documentation](https://github.com/Dr-Noob/cpufetch/tree/master/doc).
## 6. Bugs or improvements
### 6.1 Unknown microarchitecture error
If you get the `Unknown microarchitecture detected` error when running cpufetch, it might be caused by two possible reasons:
1. You are running an old release of cpufetch (most likely)
2. Your microarchitecture is not yet supported
Download and compile the latest version (see https://github.com/Dr-Noob/cpufetch#22-building-from-source for instructions)
and verify if the error persists.
* __If the error dissapears__: It means that this is the first situation. In this case, just use the
latest version of cpufetch which already has support for your hardware.
* __If the error does not dissapear__: It means that this is the
second situation. In this case, please create a new issue with the error message and the output of 'cpufetch --debug' on https://github.com/Dr-Noob/cpufetch/issues
### 6.2 Other situations
See [cpufetch contributing guidelines](https://github.com/Dr-Noob/cpufetch/blob/master/CONTRIBUTING.md).
## 7. Acknowledgements
Thanks to the fellow contributors and interested people in the project. Special thanks to:
- [Gonzalocl](https://github.com/Gonzalocl) and [OdnetninI](https://github.com/OdnetninI): Tested cpufetch in the earlier versions of the project in many different CPUs.
- [Kyngo](https://github.com/Kyngo): Tested cpufetch in the Apple M1 CPU.
- [avollmerhaus](https://github.com/avollmerhaus): Gave me ssh acess to a PowerPC machine, allowing me to develop the PowerPC port.
- [avollmerhaus](https://github.com/avollmerhaus): Helped with PowerPC port giving ssh access to a PowerPC machine.
- [bbonev](https://github.com/bbonev) and [stephan-cr](https://github.com/stephan-cr): Reviewed the source code.
- [mdoksa76](https://github.com/mdoksa76) and [exkc](https://github.com/exkc): Excellent ideas and impeccable feedback for supporting Allwinner SoCs.
- [mdoksa76](https://github.com/mdoksa76) and [exkc](https://github.com/exkc): Excellent ideas and feedback for supporting Allwinner SoCs.
- [Sakura286](https://github.com/Sakura286), [exkc](https://github.com/exkc) and [Patola](https://github.com/Patola): Helped with RISC-V port with ssh access, ideas, testing, etc.
- [ThomasKaiser](https://github.com/ThomasKaiser): Very valuable feedback on improving ARM SoC detection (Apple, Allwinner, Rockchip).
- [zerkerX](https://github.com/zerkerX): Helped with feedback for supporting old (e.g., Pentium III) Intel CPUs.
## 8. cpufetch for GPUs (gpufetch)
See [gpufetch](https://github.com/Dr-Noob/gpufetch) project!

BIN
pictures/starfive.png Normal file

Binary file not shown.

After

Width:  |  Height:  |  Size: 42 KiB

View File

@@ -8,15 +8,76 @@
#ifdef __linux__
#include <sys/auxv.h>
#include <asm/hwcap.h>
#include "../common/freq.h"
#elif defined __APPLE__ || __MACH__
#include "sysctl.h"
#include "../common/sysctl.h"
#elif defined _WIN32
#define WIN32_LEAN_AND_MEAN
#define NOMINMAX
#include <windows.h>
#endif
#include "../common/global.h"
#include "../common/soc.h"
#include "../common/args.h"
#include "udev.h"
#include "midr.h"
#include "uarch.h"
#include "soc.h"
#include "sve.h"
#if defined _WIN32
// Windows stores processor information in registery at:
// "HKEY_LOCAL_MACHINE\HARDWARE\DESCRIPTION\System\CentralProcessor"
// Within this directory, each core will get its own folder with
// registery entries named `CP ####` that map to ARM system registers.
// Ex. the MIDR register for core 0 is the `REG_QWORD` at:
// "HKEY_LOCAL_MACHINE\HARDWARE\DESCRIPTION\System\CentralProcessor\0\CP 4000"
// The name of these `CP ####`-registers follow their register ID encoding in hexadecimal
// (op0&1):op1:crn:crm:op2.
// More registers can be found here:
// https://developer.arm.com/documentation/ddi0601/2024-06/AArch64-Registers
// Some important ones:
// CP 4000: MIDR_EL1
// CP 4020: ID_AA64PFR0_EL1
// CP 4021: ID_AA64PFR1_EL1
// CP 4028: ID_AA64DFR0_EL1
// CP 4029: ID_AA64DFR1_EL1
// CP 402C: ID_AA64AFR0_EL1
// CP 402D: ID_AA64AFR1_EL1
// CP 4030: ID_AA64ISAR0_EL1
// CP 4031: ID_AA64ISAR1_EL1
// CP 4038: ID_AA64MMFR0_EL1
// CP 4039: ID_AA64MMFR1_EL1
// CP 403A: ID_AA64MMFR2_EL1
bool read_registry_hklm_int(char* path, char* name, void* value, bool is64) {
DWORD value_len;
int reg_type;
if (is64) {
value_len = sizeof(int64_t);
reg_type = RRF_RT_REG_QWORD;
}
else {
value_len = sizeof(int32_t);
reg_type = RRF_RT_REG_DWORD;
}
if(RegGetValueA(HKEY_LOCAL_MACHINE, path, name, reg_type, NULL, value, &value_len) != ERROR_SUCCESS) {
printBug("Error reading registry entry \"%s\\%s\"", path, name);
return false;
}
return true;
}
bool get_win32_core_info_int(uint32_t core_index, char* name, void* value, bool is64) {
// path + digits
uint32_t max_path_size = 45+3+1;
char* path = ecalloc(sizeof(char) * max_path_size, sizeof(char));
snprintf(path, max_path_size, "HARDWARE\\DESCRIPTION\\System\\CentralProcessor\\%u", core_index);
return read_registry_hklm_int(path, name, value, is64);
}
#endif
bool cores_are_equal(int c1pos, int c2pos, uint32_t* midr_array, int32_t* freq_array) {
return midr_array[c1pos] == midr_array[c2pos] && freq_array[c1pos] == freq_array[c2pos];
@@ -39,8 +100,17 @@ struct cache* get_cache_info(struct cpuInfo* cpu) {
struct frequency* get_frequency_info(uint32_t core) {
struct frequency* freq = emalloc(sizeof(struct frequency));
freq->measured = false;
freq->base = UNKNOWN_DATA;
freq->max = get_max_freq_from_file(core);
#ifdef __linux__
if (freq->max == UNKNOWN_DATA || measure_max_frequency_flag()) {
if (freq->max == UNKNOWN_DATA)
printWarn("Unable to find max frequency from udev, measuring CPU frequency");
freq->max = measure_max_frequency(core);
freq->measured = true;
}
#endif
return freq;
}
@@ -80,26 +150,21 @@ int64_t get_peak_performance(struct cpuInfo* cpu) {
}
}
int64_t flops = 0;
int64_t total_flops = 0;
ptr = cpu;
if(cpu->soc->soc_vendor == SOC_VENDOR_APPLE) {
// Special case for M1/M2
// First we find the E cores, then the P
// M1 have 2 (E cores) or 4 (P cores) FMA units
// Source: https://dougallj.github.io/applecpu/firestorm-simd.html
flops += ptr->topo->total_cores * (get_freq(ptr->freq) * 1000000) * 2 * 4 * 2;
ptr = ptr->next_cpu;
flops += ptr->topo->total_cores * (get_freq(ptr->freq) * 1000000) * 2 * 4 * 4;
}
else {
for(int i=0; i < cpu->num_cpus; ptr = ptr->next_cpu, i++) {
flops += ptr->topo->total_cores * (get_freq(ptr->freq) * 1000000);
}
if(cpu->feat->NEON) flops = flops * 4;
int vpus = get_number_of_vpus(ptr);
int vpus_width = get_vpus_width(ptr);
bool has_fma = has_fma_support(ptr);
int64_t flops = ptr->topo->total_cores * get_freq(ptr->freq) * 1000000 * vpus * (vpus_width/32);
if(has_fma) flops = flops * 2;
total_flops += flops;
}
return flops;
return total_flops;
}
uint32_t fill_ids_from_midr(uint32_t* midr_array, int32_t* freq_array, uint32_t* ids_array, int len) {
@@ -162,6 +227,17 @@ struct features* get_features_info(void) {
feat->SHA1 = hwcaps & HWCAP_SHA1;
feat->SHA2 = hwcaps & HWCAP_SHA2;
feat->NEON = hwcaps & HWCAP_ASIMD;
feat->SVE = hwcaps & HWCAP_SVE;
hwcaps = getauxval(AT_HWCAP2);
if (errno == ENOENT) {
printWarn("Unable to retrieve AT_HWCAP2 using getauxval");
}
else {
#ifdef HWCAP2_SVE2
feat->SVE2 = hwcaps & HWCAP2_SVE2;
#endif
}
}
#else
else {
@@ -177,6 +253,8 @@ struct features* get_features_info(void) {
feat->CRC32 = hwcaps & HWCAP2_CRC32;
feat->SHA1 = hwcaps & HWCAP2_SHA1;
feat->SHA2 = hwcaps & HWCAP2_SHA2;
feat->SVE = false;
feat->SVE2 = false;
}
#endif // ifdef __aarch64__
#elif defined __APPLE__ || __MACH__
@@ -186,8 +264,54 @@ struct features* get_features_info(void) {
feat->SHA1 = true;
feat->SHA2 = true;
feat->NEON = true;
feat->SVE = false;
feat->SVE2 = false;
#elif defined _WIN32
// CP 4020 maps to the ID_AA64PFR0_EL1 register on Windows
// https://developer.arm.com/documentation/ddi0601/2024-06/AArch64-Registers/ID-AA64PFR0-EL1--AArch64-Processor-Feature-Register-0
int64_t pfr0 = 0;
if(!get_win32_core_info_int(0, "CP 4020", &pfr0, true)) {
printWarn("Unable to retrieve PFR0 via registry");
}
else {
// AdvSimd[23:20]
// -1: Not available
// 0: AdvSimd support
// 1: AdvSimd support + FP16
int8_t adv_simd = ((int64_t)(pfr0 << (60 - 20)) >> 60);
feat->NEON = (adv_simd >= 0);
// SVE[35:32]
feat->SVE = (pfr0 >> 32) & 0xF ? true : false;
}
// Windoes does not expose a registry entry for the ID_AA64ZFR0_EL1 register
// this would have mapped to "CP 4024".
feat->SVE2 = false;
// CP 4030 maps to the ID_AA64ISAR0_EL1 register on Windows
// https://developer.arm.com/documentation/ddi0601/2024-06/AArch64-Registers/ID-AA64ISAR0-EL1--AArch64-Instruction-Set-Attribute-Register-0
int64_t isar0 = 0;
if(!get_win32_core_info_int(0, "CP 4030", &isar0, true)) {
printWarn("Unable to retrieve ISAR0 via registry");
}
else {
// AES[7:4]
feat->AES = (isar0 >> 4) & 0xF ? true : false;
// SHA1[11:8]
feat->SHA1 = (isar0 >> 8) & 0xF ? true : false;
// SHA2[15:12]
feat->SHA2 = (isar0 >> 12) & 0xF ? true : false;
// CRC32[19:16]
feat->CRC32 = (isar0 >> 16) & 0xF ? true : false;
}
#endif // ifdef __linux__
if (feat->SVE || feat->SVE2) {
feat->cntb = sve_cntb();
}
return feat;
}
@@ -242,7 +366,7 @@ struct cpuInfo* get_cpu_info_linux(struct cpuInfo* cpu) {
cpu->num_cpus = sockets;
cpu->hv = emalloc(sizeof(struct hypervisor));
cpu->hv->present = false;
cpu->soc = get_soc();
cpu->soc = get_soc(cpu);
cpu->peak_performance = get_peak_performance(cpu);
return cpu;
@@ -294,7 +418,7 @@ void fill_cpu_info_avalanche_blizzard(struct cpuInfo* cpu, uint32_t pcores, uint
bli->feat = get_features_info();
bli->topo = malloc(sizeof(struct topology));
bli->topo->cach = bli->cach;
bli->topo->total_cores = pcores;
bli->topo->total_cores = ecores;
bli->freq = malloc(sizeof(struct frequency));
bli->freq->base = UNKNOWN_DATA;
bli->freq->max = 2800;
@@ -310,7 +434,7 @@ void fill_cpu_info_avalanche_blizzard(struct cpuInfo* cpu, uint32_t pcores, uint
ava->feat = get_features_info();
ava->topo = malloc(sizeof(struct topology));
ava->topo->cach = ava->cach;
ava->topo->total_cores = ecores;
ava->topo->total_cores = pcores;
ava->freq = malloc(sizeof(struct frequency));
ava->freq->base = UNKNOWN_DATA;
ava->freq->max = 3500;
@@ -319,56 +443,151 @@ void fill_cpu_info_avalanche_blizzard(struct cpuInfo* cpu, uint32_t pcores, uint
ava->next_cpu = NULL;
}
struct cpuInfo* get_cpu_info_mach(struct cpuInfo* cpu) {
uint32_t cpu_family = get_sys_info_by_name("hw.cpufamily");
void fill_cpu_info_everest_sawtooth(struct cpuInfo* cpu, uint32_t pcores, uint32_t ecores) {
// 1. Fill SAWTOOTH
struct cpuInfo* saw = cpu;
saw->midr = MIDR_APPLE_M3_SAWTOOTH;
saw->arch = get_uarch_from_midr(saw->midr, saw);
saw->cach = get_cache_info(saw);
saw->feat = get_features_info();
saw->topo = malloc(sizeof(struct topology));
saw->topo->cach = saw->cach;
saw->topo->total_cores = ecores;
saw->freq = malloc(sizeof(struct frequency));
saw->freq->base = UNKNOWN_DATA;
saw->freq->max = 2750;
saw->hv = malloc(sizeof(struct hypervisor));
saw->hv->present = false;
saw->next_cpu = malloc(sizeof(struct cpuInfo));
// 2. Fill EVEREST
struct cpuInfo* eve = saw->next_cpu;
eve->midr = MIDR_APPLE_M3_EVEREST;
eve->arch = get_uarch_from_midr(eve->midr, eve);
eve->cach = get_cache_info(eve);
eve->feat = get_features_info();
eve->topo = malloc(sizeof(struct topology));
eve->topo->cach = eve->cach;
eve->topo->total_cores = pcores;
eve->freq = malloc(sizeof(struct frequency));
eve->freq->base = UNKNOWN_DATA;
eve->freq->max = 4050;
eve->hv = malloc(sizeof(struct hypervisor));
eve->hv->present = false;
eve->next_cpu = NULL;
}
struct cpuInfo* get_cpu_info_mach(struct cpuInfo* cpu) {
// https://developer.apple.com/documentation/kernel/1387446-sysctlbyname/determining_system_capabilities
uint32_t nperflevels = get_sys_info_by_name("hw.nperflevels");
if((cpu->num_cpus = nperflevels) != 2) {
printBug("Expected to find SoC with 2 perf levels, found: %d", cpu->num_cpus);
return NULL;
}
uint32_t pcores = get_sys_info_by_name("hw.perflevel0.physicalcpu");
uint32_t ecores = get_sys_info_by_name("hw.perflevel1.physicalcpu");
if(ecores <= 0) {
printBug("Expected to find a numer of ecores > 0, found: %d", ecores);
return NULL;
}
if(pcores <= 0) {
printBug("Expected to find a numer of pcores > 0, found: %d", pcores);
return NULL;
}
uint32_t cpu_family = get_sys_info_by_name("hw.cpufamily");
// Manually fill the cpuInfo assuming that
// the CPU is an Apple M1/M2
// the CPU is an Apple SoC
if(cpu_family == CPUFAMILY_ARM_FIRESTORM_ICESTORM) {
cpu->num_cpus = 2;
// Now detect the M1 version
uint32_t cpu_subfamily = get_sys_info_by_name("hw.cpusubfamily");
if(cpu_subfamily == CPUSUBFAMILY_ARM_HG) {
// Apple M1
fill_cpu_info_firestorm_icestorm(cpu, 4, 4);
}
else if(cpu_subfamily == CPUSUBFAMILY_ARM_HS || cpu_subfamily == CPUSUBFAMILY_ARM_HC_HD) {
// Apple M1 Pro/Max/Ultra. Detect number of cores
uint32_t physicalcpu = get_sys_info_by_name("hw.physicalcpu");
if(physicalcpu == 20) {
// M1 Ultra
fill_cpu_info_firestorm_icestorm(cpu, 16, 4);
}
else if(physicalcpu == 8 || physicalcpu == 10) {
// M1 Pro/Max
fill_cpu_info_firestorm_icestorm(cpu, physicalcpu-2, 2);
}
else {
printBug("Found invalid physical cpu number: %d", physicalcpu);
return NULL;
}
}
else {
printBug("Found invalid cpu_subfamily: 0x%.8X", cpu_subfamily);
return NULL;
}
cpu->soc = get_soc();
fill_cpu_info_firestorm_icestorm(cpu, pcores, ecores);
cpu->soc = get_soc(cpu);
cpu->peak_performance = get_peak_performance(cpu);
}
else if(cpu_family == CPUFAMILY_ARM_AVALANCHE_BLIZZARD) {
// Just the "normal" M2 exists for now
cpu->num_cpus = 2;
fill_cpu_info_avalanche_blizzard(cpu, 4, 4);
cpu->soc = get_soc();
fill_cpu_info_avalanche_blizzard(cpu, pcores, ecores);
cpu->soc = get_soc(cpu);
cpu->peak_performance = get_peak_performance(cpu);
}
else if(cpu_family == CPUFAMILY_ARM_EVEREST_SAWTOOTH ||
cpu_family == CPUFAMILY_ARM_EVEREST_SAWTOOTH_2 ||
cpu_family == CPUFAMILY_ARM_EVEREST_SAWTOOTH_PRO ||
cpu_family == CPUFAMILY_ARM_EVEREST_SAWTOOTH_MAX) {
fill_cpu_info_everest_sawtooth(cpu, pcores, ecores);
cpu->soc = get_soc(cpu);
cpu->peak_performance = get_peak_performance(cpu);
}
else {
printBug("Found invalid cpu_family: 0x%.8X", cpu_family);
printBugCheckRelease("Found invalid cpu_family: 0x%.8X", cpu_family);
return NULL;
}
return cpu;
}
#elif defined _WIN32
struct cpuInfo* get_cpu_info_windows(struct cpuInfo* cpu) {
init_cpu_info(cpu);
SYSTEM_INFO sys_info;
GetSystemInfo(&sys_info);
int ncores = sys_info.dwNumberOfProcessors;
uint32_t* midr_array = emalloc(sizeof(uint32_t) * ncores);
int32_t* freq_array = emalloc(sizeof(uint32_t) * ncores);
uint32_t* ids_array = emalloc(sizeof(uint32_t) * ncores);
for(int i=0; i < ncores; i++) {
// Cast from 64 to 32 bit to be able to re-use the pre-existing
// functions such as fill_ids_from_midr and cores_are_equal
int64_t midr_64;
if(!get_win32_core_info_int(i, "CP 4000", &midr_64, true)) {
return NULL;
}
midr_array[i] = midr_64;
if(!get_win32_core_info_int(i, "~MHz", &freq_array[i], false)) {
return NULL;
}
}
uint32_t sockets = fill_ids_from_midr(midr_array, freq_array, ids_array, ncores);
struct cpuInfo* ptr = cpu;
int midr_idx = 0;
int tmp_midr_idx = 0;
for(uint32_t i=0; i < sockets; i++) {
if(i > 0) {
ptr->next_cpu = emalloc(sizeof(struct cpuInfo));
ptr = ptr->next_cpu;
init_cpu_info(ptr);
tmp_midr_idx = midr_idx;
while(cores_are_equal(midr_idx, tmp_midr_idx, midr_array, freq_array)) tmp_midr_idx++;
midr_idx = tmp_midr_idx;
}
ptr->midr = midr_array[midr_idx];
ptr->arch = get_uarch_from_midr(ptr->midr, ptr);
ptr->feat = get_features_info();
ptr->freq = emalloc(sizeof(struct frequency));
ptr->freq->measured = false;
ptr->freq->base = freq_array[midr_idx];
ptr->freq->max = UNKNOWN_DATA;
ptr->cach = get_cache_info(ptr);
ptr->topo = get_topology_info(ptr, ptr->cach, midr_array, freq_array, i, ncores);
}
cpu->num_cpus = sockets;
cpu->hv = emalloc(sizeof(struct hypervisor));
cpu->hv->present = false;
cpu->soc = get_soc(cpu);
cpu->peak_performance = get_peak_performance(cpu);
return cpu;
}
#endif
struct cpuInfo* get_cpu_info(void) {
@@ -379,6 +598,8 @@ struct cpuInfo* get_cpu_info(void) {
return get_cpu_info_linux(cpu);
#elif defined __APPLE__ || __MACH__
return get_cpu_info_mach(cpu);
#elif defined _WIN32
return get_cpu_info_windows(cpu);
#endif
}
@@ -392,7 +613,7 @@ char* get_str_topology(struct cpuInfo* cpu, struct topology* topo, bool dual_soc
char* get_str_features(struct cpuInfo* cpu) {
struct features* feat = cpu->feat;
uint32_t max_len = strlen("NEON,SHA1,SHA2,AES,CRC32,") + 1;
uint32_t max_len = strlen("NEON,SHA1,SHA2,AES,CRC32,SVE,SVE2,") + 1;
uint32_t len = 0;
char* string = ecalloc(max_len, sizeof(char));
@@ -400,6 +621,14 @@ char* get_str_features(struct cpuInfo* cpu) {
strcat(string, "NEON,");
len += 5;
}
if(feat->SVE) {
strcat(string, "SVE,");
len += 4;
}
if(feat->SVE2) {
strcat(string, "SVE2,");
len += 5;
}
if(feat->SHA1) {
strcat(string, "SHA1,");
len += 5;
@@ -448,6 +677,19 @@ void print_debug(struct cpuInfo* cpu) {
printf("%ld MHz\n", freq);
}
}
if (cpu->feat->SVE || cpu->feat->SVE2) {
printf("- cntb: %d\n", (int) cpu->feat->cntb);
}
#if defined(__APPLE__) || defined(__MACH__)
printf("hw.cpufamily: 0x%.8X\n", get_sys_info_by_name("hw.cpufamily"));
printf("hw.cpusubfamily: 0x%.8X\n", get_sys_info_by_name("hw.cpusubfamily"));
printf("hw.nperflevels: %d\n", get_sys_info_by_name("hw.nperflevels"));
printf("hw.physicalcpu: %d\n", get_sys_info_by_name("hw.physicalcpu"));
printf("hw.perflevel0.physicalcpu: %d\n", get_sys_info_by_name("hw.perflevel0.physicalcpu"));
printf("hw.perflevel1.physicalcpu: %d\n", get_sys_info_by_name("hw.perflevel1.physicalcpu"));
#endif
}
void free_topo_struct(struct topology* topo) {

File diff suppressed because it is too large Load Diff

View File

@@ -1,34 +1,10 @@
#ifndef __SOC__
#define __SOC__
#ifndef __SOC_ARM__
#define __SOC_ARM__
#include "../common/cpu.h"
#include "../common/soc.h"
#include <stdint.h>
typedef int32_t SOC;
enum {
SOC_VENDOR_UNKNOWN,
SOC_VENDOR_SNAPDRAGON,
SOC_VENDOR_MEDIATEK,
SOC_VENDOR_EXYNOS,
SOC_VENDOR_KIRIN,
SOC_VENDOR_BROADCOM,
SOC_VENDOR_APPLE,
SOC_VENDOR_ALLWINNER,
SOC_VENDOR_ROCKCHIP
};
struct system_on_chip {
SOC soc_model;
VENDOR soc_vendor;
int32_t process;
char* soc_name;
char* raw_name;
};
struct system_on_chip* get_soc(void);
char* get_soc_name(struct system_on_chip* soc);
VENDOR get_soc_vendor(struct system_on_chip* soc);
char* get_str_process(struct system_on_chip* soc);
struct system_on_chip* get_soc(struct cpuInfo* cpu);
#endif

View File

@@ -10,7 +10,6 @@ enum {
SOC_BCM_2836,
SOC_BCM_2837,
SOC_BCM_2837B0,
SOC_BCM_2711,
SOC_BCM_21553,
SOC_BCM_21553T,
SOC_BCM_21663,
@@ -20,6 +19,8 @@ enum {
SOC_BCM_28145,
SOC_BCM_2157,
SOC_BCM_21654,
SOC_BCM_2711,
SOC_BCM_2712,
// Hisilicon //
SOC_HISILICON_3620,
SOC_HISILICON_3630,
@@ -28,6 +29,10 @@ enum {
SOC_HISILICON_3670,
SOC_HISILICON_3680,
SOC_HISILICON_3690,
SOC_HISILICON_9000S,
// Kunpeng //
SOC_KUNPENG_920,
SOC_KUNPENG_930,
// Exynos //
SOC_EXYNOS_3475,
SOC_EXYNOS_4210,
@@ -63,39 +68,23 @@ enum {
SOC_EXYNOS_980,
SOC_EXYNOS_880,
// Mediatek //
SOC_MTK_MT6893,
SOC_MTK_MT6891,
SOC_MTK_MT6889,
SOC_MTK_MT6885Z,
SOC_MTK_MT6853,
SOC_MTK_MT6873,
SOC_MTK_MT6875,
SOC_MTK_MT6761D,
SOC_MTK_MT6761,
SOC_MTK_MT6762D,
SOC_MTK_MT6755,
SOC_MTK_MT6755M,
SOC_MTK_MT6755T,
SOC_MTK_MT6757,
SOC_MTK_MT6762,
SOC_MTK_MT6763V,
SOC_MTK_MT6763T,
SOC_MTK_MT6757CD,
SOC_MTK_MT6758,
SOC_MTK_MT6765,
SOC_MTK_MT6771,
SOC_MTK_MT6768,
SOC_MTK_MT6771T,
SOC_MTK_MT6771V,
SOC_MTK_MT6779,
SOC_MTK_MT6795,
SOC_MTK_MT6795T,
SOC_MTK_MT6797,
SOC_MTK_MT6797M,
SOC_MTK_MT6797D,
SOC_MTK_MT6797T,
SOC_MTK_MT6797X,
SOC_MTK_MT6799,
SOC_MTK_MT5327,
SOC_MTK_MT5329,
SOC_MTK_MT5366,
SOC_MTK_MT5389,
SOC_MTK_MT5395,
SOC_MTK_MT5396,
SOC_MTK_MT5398,
SOC_MTK_MT5505,
SOC_MTK_MT5561,
SOC_MTK_MT5580,
SOC_MTK_MT5582,
SOC_MTK_MT5592,
SOC_MTK_MT5595,
SOC_MTK_MT5596,
SOC_MTK_MT5597,
SOC_MTK_MT5889,
SOC_MTK_MT5895,
SOC_MTK_MT6515,
SOC_MTK_MT6516,
SOC_MTK_MT6517,
@@ -125,7 +114,51 @@ enum {
SOC_MTK_MT6750T,
SOC_MTK_MT6752,
SOC_MTK_MT6753,
SOC_MTK_MT6755M,
SOC_MTK_MT6755T,
SOC_MTK_MT6757,
SOC_MTK_MT6757CD,
SOC_MTK_MT6758,
SOC_MTK_MT6761,
SOC_MTK_MT6761D,
SOC_MTK_MT6762,
SOC_MTK_MT6762D,
SOC_MTK_MT6762G,
SOC_MTK_MT6763T,
SOC_MTK_MT6763V,
SOC_MTK_MT6765,
SOC_MTK_MT6765G,
SOC_MTK_MT6765H,
SOC_MTK_MT6768,
SOC_MTK_MT6769H,
SOC_MTK_MT6769T,
SOC_MTK_MT6769V,
SOC_MTK_MT6769Z,
SOC_MTK_MT6771,
SOC_MTK_MT6779V_CU,
SOC_MTK_MT6779V_CV,
SOC_MTK_MT6785V_CC,
SOC_MTK_MT6785V_CD,
SOC_MTK_MT6789,
SOC_MTK_MT6795,
SOC_MTK_MT6797,
SOC_MTK_MT6797T,
SOC_MTK_MT6797X,
SOC_MTK_MT6799,
SOC_MTK_MT6833,
SOC_MTK_MT6850,
SOC_MTK_MT6853,
SOC_MTK_MT6853V,
SOC_MTK_MT6873,
SOC_MTK_MT6875,
SOC_MTK_MT6879,
SOC_MTK_MT6883Z,
SOC_MTK_MT6885Z,
SOC_MTK_MT6889,
SOC_MTK_MT6889Z,
SOC_MTK_MT6891,
SOC_MTK_MT6893,
SOC_MTK_MT6893Z,
SOC_MTK_MT8121,
SOC_MTK_MT8125,
SOC_MTK_MT8127,
@@ -140,7 +173,25 @@ enum {
SOC_MTK_MT8581,
SOC_MTK_MT8735,
SOC_MTK_MT8765B,
SOC_MTK_MT8781V,
SOC_MTK_MT8783,
SOC_MTK_MT9602,
SOC_MTK_MT9612,
SOC_MTK_MT9613,
SOC_MTK_MT9615,
SOC_MTK_MT9618,
SOC_MTK_MT9632,
SOC_MTK_MT9638,
SOC_MTK_MT9652,
SOC_MTK_MT9653,
SOC_MTK_MT9675,
SOC_MTK_MT9685,
SOC_MTK_MT9686,
SOC_MTK_MT9689,
SOC_MTK_MT9902,
SOC_MTK_MT9950,
SOC_MTK_MT9972,
SOC_MTK_MT9982,
// Snapdragon //
SOC_SNAPD_QSD8650,
SOC_SNAPD_QSD8250,
@@ -220,11 +271,16 @@ enum {
SOC_SNAPD_SDM660,
SOC_SNAPD_SM6115,
SOC_SNAPD_SM6125,
SOC_SNAPD_SM6375_AC,
SOC_SNAPD_SM6450,
SOC_SNAPD_SDM670,
SOC_SNAPD_SM6150,
SOC_SNAPD_SM6350,
SOC_SNAPD_SDM710,
SOC_SNAPD_SDM712,
SOC_SNAPD_SM4375,
SOC_SNAPD_SM4450,
SOC_SNAPD_SM4635,
SOC_SNAPD_SM7125,
SOC_SNAPD_SM7150_AA,
SOC_SNAPD_SM7150_AB,
@@ -233,6 +289,11 @@ enum {
SOC_SNAPD_SM7250_AA,
SOC_SNAPD_SM7250_AB,
SOC_SNAPD_SM7250_AC,
SOC_SNAPD_SM7435_AB,
SOC_SNAPD_SM7450,
SOC_SNAPD_SM7475,
SOC_SNAPD_SM7550_AB,
SOC_SNAPD_SM7675_AB,
SOC_SNAPD_MSM8974AA,
SOC_SNAPD_MSM8974AB,
SOC_SNAPD_MSM8974AC,
@@ -253,12 +314,23 @@ enum {
SOC_SNAPD_SM8250_AB,
SOC_SNAPD_SM8350,
SOC_SNAPD_SM8450,
SOC_SNAPD_SM8475,
SOC_SNAPD_SM8550_AB,
SOC_SNAPD_SM8635,
SOC_SNAPD_SM8650_AB,
SOC_SNAPD_SC8280XP,
// APPLE
SOC_APPLE_M1,
SOC_APPLE_M1_PRO,
SOC_APPLE_M1_MAX,
SOC_APPLE_M1_ULTRA,
SOC_APPLE_M2,
SOC_APPLE_M2_PRO,
SOC_APPLE_M2_MAX,
SOC_APPLE_M2_ULTRA,
SOC_APPLE_M3,
SOC_APPLE_M3_PRO,
SOC_APPLE_M3_MAX,
// ALLWINNER
SOC_ALLWINNER_A10,
SOC_ALLWINNER_A13,
@@ -276,32 +348,96 @@ enum {
SOC_ALLWINNER_V3S,
SOC_ALLWINNER_HZP,
SOC_ALLWINNER_H2PLUS,
SOC_ALLWINNER_S3,
SOC_ALLWINNER_H3,
SOC_ALLWINNER_H8,
SOC_ALLWINNER_H5,
SOC_ALLWINNER_H6,
SOC_ALLWINNER_H64,
SOC_ALLWINNER_H616,
SOC_ALLWINNER_H618,
SOC_ALLWINNER_R8,
SOC_ALLWINNER_R16,
SOC_ALLWINNER_R40,
SOC_ALLWINNER_R58,
SOC_ALLWINNER_R328,
// ROCKCHIP
SOC_ROCKCHIP_3288,
SOC_ROCKCHIP_3229,
SOC_ROCKCHIP_3308,
SOC_ROCKCHIP_3318,
SOC_ROCKCHIP_3326,
SOC_ROCKCHIP_3328,
SOC_ROCKCHIP_3368,
SOC_ROCKCHIP_3399,
SOC_ROCKCHIP_3528,
SOC_ROCKCHIP_3562,
SOC_ROCKCHIP_3566,
SOC_ROCKCHIP_3568,
SOC_ROCKCHIP_3588,
// GOOGLE
SOC_GOOGLE_TENSOR,
SOC_GOOGLE_TENSOR_G2,
SOC_GOOGLE_TENSOR_G3,
// NVIDIA,
SOC_TEGRA_2,
SOC_TEGRA_3,
SOC_TEGRA_4,
SOC_TEGRA_K1,
SOC_TEGRA_K2,
SOC_TEGRA_X1,
SOC_TEGRA_X2,
SOC_TEGRA_XAVIER,
SOC_TEGRA_ORIN,
// ALTRA
SOC_AMPERE_ALTRA,
// NXP
SOC_NXP_IMX8QM,
SOC_NXP_IMX8QP,
SOC_NXP_IMX8MP,
SOC_NXP_IMX8MN,
SOC_NXP_IMX8MM,
SOC_NXP_IMX8DXP,
SOC_NXP_IMX8QXP,
SOC_NXP_IMX93,
// AMLOGIC
SOC_AMLOGIC_A311D,
SOC_AMLOGIC_A311D2,
SOC_AMLOGIC_S905W,
SOC_AMLOGIC_S905D,
SOC_AMLOGIC_S905X,
SOC_AMLOGIC_S805X,
// MARVELL
SOC_MARVELL_A3700,
SOC_MARVELL_A3710,
SOC_MARVELL_A3720,
SOC_MARVELL_A7200,
SOC_MARVELL_A7400,
SOC_MARVELL_A8020,
SOC_MARVELL_A8040,
SOC_MARVELL_CN9130,
SOC_MARVELL_CN9131,
SOC_MARVELL_CN9132,
// UNKNOWN
SOC_MODEL_UNKNOWN
};
inline static VENDOR get_soc_vendor_from_soc(SOC soc) {
if(soc >= SOC_BCM_2835 && soc <= SOC_BCM_21654) return SOC_VENDOR_BROADCOM;
else if(soc >= SOC_HISILICON_3620 && soc <= SOC_HISILICON_3690) return SOC_VENDOR_KIRIN;
if(soc >= SOC_BCM_2835 && soc <= SOC_BCM_2712) return SOC_VENDOR_BROADCOM;
else if(soc >= SOC_HISILICON_3620 && soc <= SOC_HISILICON_9000S) return SOC_VENDOR_KIRIN;
else if(soc >= SOC_KUNPENG_920 && soc <= SOC_KUNPENG_930) return SOC_VENDOR_KUNPENG;
else if(soc >= SOC_EXYNOS_3475 && soc <= SOC_EXYNOS_880) return SOC_VENDOR_EXYNOS;
else if(soc >= SOC_MTK_MT6893 && soc <= SOC_MTK_MT8783) return SOC_VENDOR_MEDIATEK;
else if(soc >= SOC_SNAPD_QSD8650 && soc <= SOC_SNAPD_SM8450) return SOC_VENDOR_SNAPDRAGON;
else if(soc >= SOC_APPLE_M1 && soc <= SOC_APPLE_M2) return SOC_VENDOR_APPLE;
else if(soc >= SOC_SNAPD_QSD8650 && soc <= SOC_SNAPD_SC8280XP) return SOC_VENDOR_SNAPDRAGON;
else if(soc >= SOC_APPLE_M1 && soc <= SOC_APPLE_M3_MAX) return SOC_VENDOR_APPLE;
else if(soc >= SOC_ALLWINNER_A10 && soc <= SOC_ALLWINNER_R328) return SOC_VENDOR_ALLWINNER;
else if(soc >= SOC_ROCKCHIP_3399 && soc <= SOC_ROCKCHIP_3399) return SOC_VENDOR_ROCKCHIP;
else if(soc >= SOC_ROCKCHIP_3288 && soc <= SOC_ROCKCHIP_3588) return SOC_VENDOR_ROCKCHIP;
else if(soc >= SOC_GOOGLE_TENSOR && soc <= SOC_GOOGLE_TENSOR_G3) return SOC_VENDOR_GOOGLE;
else if(soc >= SOC_TEGRA_2 && soc <= SOC_TEGRA_ORIN) return SOC_VENDOR_NVIDIA;
else if(soc >= SOC_AMPERE_ALTRA && soc <= SOC_AMPERE_ALTRA) return SOC_VENDOR_AMPERE;
else if(soc >= SOC_NXP_IMX8QM && soc <= SOC_NXP_IMX93) return SOC_VENDOR_NXP;
else if(soc >= SOC_AMLOGIC_A311D && soc <= SOC_AMLOGIC_S805X) return SOC_VENDOR_AMLOGIC;
else if(soc >= SOC_MARVELL_A3700 && soc <= SOC_MARVELL_CN9132) return SOC_VENDOR_MARVELL;
return SOC_VENDOR_UNKNOWN;
}

15
src/arm/sve.c Normal file
View File

@@ -0,0 +1,15 @@
#include <stdint.h>
#include "../common/global.h"
// https://learn.arm.com/learning-paths/servers-and-cloud-computing/sve/sve_basics/#:~:text=Using%20a%20text%20editor%20of%20your%20choice%2C%20copy,svcntb%28%29%29%3B%20%7D%20This%20program%20prints%20the%20vector%20length
uint64_t sve_cntb(void) {
#ifdef __ARM_FEATURE_SVE
uint64_t x0 = 0;
__asm volatile("cntb %0"
: "=r"(x0));
return x0;
#else
printWarn("sve_cntb: Hardware supports SVE, but it was not enabled by the compiler");
return 0;
#endif
}

6
src/arm/sve.h Normal file
View File

@@ -0,0 +1,6 @@
#ifndef __SVE_DETECTION__
#define __SVE_DETECTION__
uint64_t sve_cntb(void);
#endif

View File

@@ -10,7 +10,6 @@
// Data not available
#define NA -1
typedef uint32_t MICROARCH;
typedef uint32_t ISA;
struct uarch {
@@ -34,85 +33,9 @@ enum {
ISA_ARMv8_3_A,
ISA_ARMv8_4_A,
ISA_ARMv8_5_A,
ISA_ARMv9_A
};
enum {
UARCH_UNKNOWN,
// ARM
UARCH_ARM7,
UARCH_ARM9,
UARCH_ARM1136,
UARCH_ARM1156,
UARCH_ARM1176,
UARCH_ARM11MPCORE,
UARCH_CORTEX_A5,
UARCH_CORTEX_A7,
UARCH_CORTEX_A8,
UARCH_CORTEX_A9,
UARCH_CORTEX_A12,
UARCH_CORTEX_A15,
UARCH_CORTEX_A17,
UARCH_CORTEX_A32,
UARCH_CORTEX_A35,
UARCH_CORTEX_A53,
UARCH_CORTEX_A55r0, // ARM Cortex-A55 revision 0 (restricted dual-issue capabilities compared to revision 1+).
UARCH_CORTEX_A55,
UARCH_CORTEX_A57,
UARCH_CORTEX_A65,
UARCH_CORTEX_A72,
UARCH_CORTEX_A73,
UARCH_CORTEX_A75,
UARCH_CORTEX_A76,
UARCH_CORTEX_A77,
UARCH_CORTEX_A78,
UARCH_CORTEX_A510,
UARCH_CORTEX_A710,
UARCH_CORTEX_X1,
UARCH_CORTEX_X2,
UARCH_NEOVERSE_N1,
UARCH_NEOVERSE_E1,
UARCH_SCORPION,
UARCH_KRAIT,
UARCH_KYRO,
UARCH_FALKOR,
UARCH_SAPHIRA,
UARCH_DENVER,
UARCH_DENVER2,
UARCH_CARMEL,
// SAMSUNG
UARCH_EXYNOS_M1, // Samsung Exynos M1 (Exynos 8890 big cores)
UARCH_EXYNOS_M2, // Samsung Exynos M2 (Exynos 8895 big cores)
UARCH_EXYNOS_M3, // Samsung Exynos M3 (Exynos 9810 big cores)
UARCH_EXYNOS_M4, // Samsung Exynos M4 (Exynos 9820 big cores)
UARCH_EXYNOS_M5, // Samsung Exynos M5 (Exynos 9830 big cores)
// APPLE
UARCH_SWIFT, // Apple A6 and A6X processors.
UARCH_CYCLONE, // Apple A7 processor.
UARCH_TYPHOON, // Apple A8 and A8X processor
UARCH_TWISTER, // Apple A9 and A9X processor.
UARCH_HURRICANE, // Apple A10 and A10X processor.
UARCH_MONSOON, // Apple A11 processor (big cores).
UARCH_MISTRAL, // Apple A11 processor (little cores).
UARCH_VORTEX, // Apple A12 processor (big cores).
UARCH_TEMPEST, // Apple A12 processor (big cores).
UARCH_LIGHTNING, // Apple A13 processor (big cores).
UARCH_THUNDER, // Apple A13 processor (little cores).
UARCH_ICESTORM, // Apple M1 processor (little cores).
UARCH_FIRESTORM, // Apple M1 processor (big cores).
UARCH_BLIZZARD, // Apple M2 processor (little cores).
UARCH_AVALANCHE, // Apple M2 processor (big cores).
// CAVIUM
UARCH_THUNDERX, // Cavium ThunderX
UARCH_THUNDERX2, // Cavium ThunderX2 (originally Broadcom Vulkan).
// MARVELL
UARCH_PJ4,
UARCH_BRAHMA_B15,
UARCH_BRAHMA_B53,
UARCH_XGENE, // Applied Micro X-Gene.
UARCH_TAISHAN_V110, // HiSilicon TaiShan v110 (Huawei Kunpeng 920 series processors).
// PHYTIUM
UARCH_XIAOMI, // Not to be confused with Xiaomi Inc
ISA_ARMv8_6_A,
ISA_ARMv9_A,
ISA_ARMv9_2_A
};
static const ISA isas_uarch[] = {
@@ -140,17 +63,33 @@ static const ISA isas_uarch[] = {
[UARCH_CORTEX_A76] = ISA_ARMv8_2_A,
[UARCH_CORTEX_A77] = ISA_ARMv8_2_A,
[UARCH_CORTEX_A78] = ISA_ARMv8_2_A,
[UARCH_CORTEX_A78C] = ISA_ARMv8_2_A,
[UARCH_CORTEX_A78AE] = ISA_ARMv8_2_A,
[UARCH_CORTEX_A510] = ISA_ARMv9_A,
[UARCH_CORTEX_A520] = ISA_ARMv9_2_A,
[UARCH_CORTEX_A710] = ISA_ARMv9_A,
[UARCH_CORTEX_A715] = ISA_ARMv9_A,
[UARCH_CORTEX_A720] = ISA_ARMv9_2_A,
[UARCH_CORTEX_A725] = ISA_ARMv9_2_A,
[UARCH_CORTEX_X1] = ISA_ARMv8_2_A,
[UARCH_CORTEX_X1C] = ISA_ARMv8_2_A, // Assuming same as X1
[UARCH_CORTEX_X2] = ISA_ARMv9_A,
[UARCH_CORTEX_X3] = ISA_ARMv9_A,
[UARCH_CORTEX_X4] = ISA_ARMv9_2_A,
[UARCH_CORTEX_X925] = ISA_ARMv9_2_A,
[UARCH_NEOVERSE_N1] = ISA_ARMv8_2_A,
[UARCH_NEOVERSE_N2] = ISA_ARMv9_A,
[UARCH_NEOVERSE_E1] = ISA_ARMv8_2_A,
[UARCH_NEOVERSE_V1] = ISA_ARMv8_4_A,
[UARCH_NEOVERSE_V2] = ISA_ARMv9_A,
[UARCH_NEOVERSE_V3] = ISA_ARMv9_2_A,
[UARCH_BRAHMA_B15] = ISA_ARMv7_A, // Same as Cortex-A15
[UARCH_BRAHMA_B53] = ISA_ARMv8_A, // Same as Cortex-A53
[UARCH_THUNDERX] = ISA_ARMv8_A,
[UARCH_THUNDERX2] = ISA_ARMv8_1_A,
[UARCH_TAISHAN_V110] = ISA_ARMv8_2_A,
[UARCH_TAISHAN_V120] = ISA_ARMv8_2_A, // Not confirmed
[UARCH_TAISHAN_V200] = ISA_ARMv8_2_A, // Not confirmed
[UARCH_DENVER] = ISA_ARMv8_A,
[UARCH_DENVER2] = ISA_ARMv8_A,
[UARCH_CARMEL] = ISA_ARMv8_A,
@@ -167,8 +106,10 @@ static const ISA isas_uarch[] = {
[UARCH_EXYNOS_M5] = ISA_ARMv8_2_A,
[UARCH_ICESTORM] = ISA_ARMv8_5_A, // https://github.com/llvm/llvm-project/blob/main/llvm/include/llvm/Support/AArch64TargetParser.def
[UARCH_FIRESTORM] = ISA_ARMv8_5_A,
[UARCH_BLIZZARD] = ISA_ARMv8_5_A, // Not confirmed
[UARCH_AVALANCHE] = ISA_ARMv8_5_A,
[UARCH_BLIZZARD] = ISA_ARMv8_6_A, // https://github.com/llvm/llvm-project/blob/main/llvm/unittests/TargetParser/TargetParserTest.cpp
[UARCH_AVALANCHE] = ISA_ARMv8_6_A, // https://github.com/llvm/llvm-project/blob/main/llvm/unittests/TargetParser/TargetParserTest.cpp
[UARCH_SAWTOOTH] = ISA_ARMv8_6_A, // https://github.com/llvm/llvm-project/blob/main/llvm/unittests/TargetParser/TargetParserTest.cpp
[UARCH_EVEREST] = ISA_ARMv8_6_A, // https://github.com/llvm/llvm-project/blob/main/llvm/unittests/TargetParser/TargetParserTest.cpp
[UARCH_PJ4] = ISA_ARMv7_A,
[UARCH_XIAOMI] = ISA_ARMv8_A,
};
@@ -186,13 +127,16 @@ static char* isas_string[] = {
[ISA_ARMv8_3_A] = "ARMv8.3",
[ISA_ARMv8_4_A] = "ARMv8.4",
[ISA_ARMv8_5_A] = "ARMv8.5",
[ISA_ARMv9_A] = "ARMv9"
[ISA_ARMv8_6_A] = "ARMv8.6",
[ISA_ARMv9_A] = "ARMv9",
[ISA_ARMv9_2_A] = "ARMv9.2",
};
#define UARCH_START if (false) {}
#define CHECK_UARCH(arch, cpu, im_, p_, v_, r_, str, uarch, vendor) \
else if (im_ == im && p_ == p && (v_ == NA || v_ == v) && (r_ == NA || r_ == r)) fill_uarch(arch, cpu, str, uarch, vendor);
#define UARCH_END else { printBug("Unknown microarchitecture detected: IM=0x%.8X P=0x%.8X V=0x%.8X R=0x%.8X", im, p, v, r); fill_uarch(arch, cpu, "Unknown", UARCH_UNKNOWN, CPU_VENDOR_UNKNOWN); }
#define UARCH_END else { printBugCheckRelease("Unknown microarchitecture detected: IM=0x%X P=0x%X V=0x%X R=0x%X", im, p, v, r); \
fill_uarch(arch, cpu, "Unknown", UARCH_UNKNOWN, CPU_VENDOR_UNKNOWN); }
void fill_uarch(struct uarch* arch, struct cpuInfo* cpu, char* str, MICROARCH u, VENDOR vendor) {
arch->uarch = u;
@@ -208,10 +152,11 @@ void fill_uarch(struct uarch* arch, struct cpuInfo* cpu, char* str, MICROARCH u,
/*
* Codes are based on pytorch/cpuinfo, more precisely:
* - https://github.com/pytorch/cpuinfo/blob/master/src/arm/uarch.c
* - https://github.com/pytorch/cpuinfo/blob/main/src/arm/uarch.c
* Other sources:
* - https://elixir.bootlin.com/linux/latest/source/arch/arm64/include/asm/cputype.h
* - https://elixir.bootlin.com/linux/latest/source/arch/arm/include/asm/cputype.h
* - https://github.com/AsahiLinux/m1n1/blob/main/src/chickens.c
*/
struct uarch* get_uarch_from_midr(uint32_t midr, struct cpuInfo* cpu) {
struct uarch* arch = emalloc(sizeof(struct uarch));
@@ -254,12 +199,26 @@ struct uarch* get_uarch_from_midr(uint32_t midr, struct cpuInfo* cpu) {
CHECK_UARCH(arch, cpu, 'A', 0xD0C, NA, NA, "Neoverse N1", UARCH_NEOVERSE_N1, CPU_VENDOR_ARM)
CHECK_UARCH(arch, cpu, 'A', 0xD0D, NA, NA, "Cortex-A77", UARCH_CORTEX_A77, CPU_VENDOR_ARM)
CHECK_UARCH(arch, cpu, 'A', 0xD0E, NA, NA, "Cortex-A76", UARCH_CORTEX_A76, CPU_VENDOR_ARM)
CHECK_UARCH(arch, cpu, 'A', 0xD40, NA, NA, "Neoverse V1", UARCH_NEOVERSE_V1, CPU_VENDOR_ARM)
CHECK_UARCH(arch, cpu, 'A', 0xD41, NA, NA, "Cortex-A78", UARCH_CORTEX_A78, CPU_VENDOR_ARM)
CHECK_UARCH(arch, cpu, 'A', 0xD42, NA, NA, "Cortex-A78AE", UARCH_CORTEX_A78AE, CPU_VENDOR_ARM)
CHECK_UARCH(arch, cpu, 'A', 0xD44, NA, NA, "Cortex-X1", UARCH_CORTEX_X1, CPU_VENDOR_ARM)
CHECK_UARCH(arch, cpu, 'A', 0xD46, NA, NA, "CortexA510", UARCH_CORTEX_A510, CPU_VENDOR_ARM)
CHECK_UARCH(arch, cpu, 'A', 0xD47, NA, NA, "CortexA710", UARCH_CORTEX_A710, CPU_VENDOR_ARM)
CHECK_UARCH(arch, cpu, 'A', 0xD48, NA, NA, "Cortex-X2", UARCH_CORTEX_X2, CPU_VENDOR_ARM)
CHECK_UARCH(arch, cpu, 'A', 0xD49, NA, NA, "Neoverse N2", UARCH_NEOVERSE_N2, CPU_VENDOR_ARM)
CHECK_UARCH(arch, cpu, 'A', 0xD4A, NA, NA, "Neoverse E1", UARCH_NEOVERSE_E1, CPU_VENDOR_ARM)
CHECK_UARCH(arch, cpu, 'A', 0xD4B, NA, NA, "Cortex-A78C", UARCH_CORTEX_A78C, CPU_VENDOR_ARM)
CHECK_UARCH(arch, cpu, 'A', 0xD4C, NA, NA, "Cortex-X1C", UARCH_CORTEX_X1C, CPU_VENDOR_ARM)
CHECK_UARCH(arch, cpu, 'A', 0xD4D, NA, NA, "Cortex-A715", UARCH_CORTEX_A715, CPU_VENDOR_ARM)
CHECK_UARCH(arch, cpu, 'A', 0xD4E, NA, NA, "Cortex-X3", UARCH_CORTEX_X3, CPU_VENDOR_ARM)
CHECK_UARCH(arch, cpu, 'A', 0xD4F, NA, NA, "Neoverse V2", UARCH_NEOVERSE_V2, CPU_VENDOR_ARM)
CHECK_UARCH(arch, cpu, 'A', 0xD80, NA, NA, "Cortex-A520", UARCH_CORTEX_A520, CPU_VENDOR_ARM)
CHECK_UARCH(arch, cpu, 'A', 0xD81, NA, NA, "Cortex-A720", UARCH_CORTEX_A720, CPU_VENDOR_ARM)
CHECK_UARCH(arch, cpu, 'A', 0xD82, NA, NA, "Cortex-X4", UARCH_CORTEX_X4, CPU_VENDOR_ARM)
CHECK_UARCH(arch, cpu, 'A', 0xD84, NA, NA, "Neoverse V3", UARCH_NEOVERSE_V3, CPU_VENDOR_ARM)
CHECK_UARCH(arch, cpu, 'A', 0xD85, NA, NA, "Cortex-X925", UARCH_CORTEX_X925, CPU_VENDOR_ARM)
CHECK_UARCH(arch, cpu, 'A', 0xD87, NA, NA, "Cortex-A725", UARCH_CORTEX_A725, CPU_VENDOR_ARM)
CHECK_UARCH(arch, cpu, 'B', 0x00F, NA, NA, "Brahma B15", UARCH_BRAHMA_B15, CPU_VENDOR_BROADCOM)
CHECK_UARCH(arch, cpu, 'B', 0x100, NA, NA, "Brahma B53", UARCH_BRAHMA_B53, CPU_VENDOR_BROADCOM)
@@ -271,8 +230,11 @@ struct uarch* get_uarch_from_midr(uint32_t midr, struct cpuInfo* cpu) {
CHECK_UARCH(arch, cpu, 'C', 0x0A3, NA, NA, "ThunderX 81XX", UARCH_THUNDERX, CPU_VENDOR_CAVIUM)
CHECK_UARCH(arch, cpu, 'C', 0x0AF, NA, NA, "ThunderX2 99XX", UARCH_THUNDERX2, CPU_VENDOR_CAVIUM)
CHECK_UARCH(arch, cpu, 'H', 0xD01, NA, NA, "TaiShan v110", UARCH_TAISHAN_V110, CPU_VENDOR_HUAWUEI) // Kunpeng 920 series
CHECK_UARCH(arch, cpu, 'H', 0xD01, NA, NA, "TaiShan v110", UARCH_TAISHAN_V110, CPU_VENDOR_HUAWEI) // Kunpeng 920 series
CHECK_UARCH(arch, cpu, 'H', 0xD02, 2, 2, "TaiShan v120", UARCH_TAISHAN_V120, CPU_VENDOR_HUAWEI) // Kiring 9000S Big cores (https://github.com/Dr-Noob/cpufetch/issues/259)
CHECK_UARCH(arch, cpu, 'H', 0xD02, NA, NA, "TaiShan v200", UARCH_TAISHAN_V200, CPU_VENDOR_HUAWEI) // Kunpeng 930 series (found in openeuler: https://mailweb.openeuler.org/hyperkitty/list/kernel@openeuler.org/message/XQCV7NX2UKRIUWUFKRF4PO3QENCOUFR3)
CHECK_UARCH(arch, cpu, 'H', 0xD40, NA, NA, "Cortex-A76", UARCH_CORTEX_A76, CPU_VENDOR_ARM) // Kirin 980 Big/Medium cores -> Cortex-A76
CHECK_UARCH(arch, cpu, 'H', 0xD42, NA, NA, "TaiShan v120", UARCH_TAISHAN_V120, CPU_VENDOR_HUAWEI) // Kiring 9000S Small Cores (https://github.com/Dr-Noob/cpufetch/issues/259)
CHECK_UARCH(arch, cpu, 'N', 0x000, NA, NA, "Denver", UARCH_DENVER, CPU_VENDOR_NVIDIA)
CHECK_UARCH(arch, cpu, 'N', 0x003, NA, NA, "Denver2", UARCH_DENVER2, CPU_VENDOR_NVIDIA)
@@ -314,8 +276,12 @@ struct uarch* get_uarch_from_midr(uint32_t midr, struct cpuInfo* cpu) {
CHECK_UARCH(arch, cpu, 'a', 0x022, NA, NA, "Icestorm", UARCH_ICESTORM, CPU_VENDOR_APPLE)
CHECK_UARCH(arch, cpu, 'a', 0x023, NA, NA, "Firestorm", UARCH_FIRESTORM, CPU_VENDOR_APPLE)
CHECK_UARCH(arch, cpu, 'a', 0x024, NA, NA, "Icestorm", UARCH_ICESTORM, CPU_VENDOR_APPLE) // https://github.com/Dr-Noob/cpufetch/issues/263
CHECK_UARCH(arch, cpu, 'a', 0x025, NA, NA, "Firestorm", UARCH_FIRESTORM, CPU_VENDOR_APPLE) // https://github.com/Dr-Noob/cpufetch/issues/263
CHECK_UARCH(arch, cpu, 'a', 0x030, NA, NA, "Blizzard", UARCH_BLIZZARD, CPU_VENDOR_APPLE)
CHECK_UARCH(arch, cpu, 'a', 0x031, NA, NA, "Avalanche", UARCH_AVALANCHE, CPU_VENDOR_APPLE)
CHECK_UARCH(arch, cpu, 'a', 0x048, NA, NA, "Sawtooth", UARCH_SAWTOOTH, CPU_VENDOR_APPLE)
CHECK_UARCH(arch, cpu, 'a', 0x049, NA, NA, "Everest", UARCH_EVEREST, CPU_VENDOR_APPLE)
CHECK_UARCH(arch, cpu, 'V', 0x581, NA, NA, "PJ4", UARCH_PJ4, CPU_VENDOR_MARVELL)
CHECK_UARCH(arch, cpu, 'V', 0x584, NA, NA, "PJ4B-MP", UARCH_PJ4, CPU_VENDOR_MARVELL)
@@ -325,10 +291,112 @@ struct uarch* get_uarch_from_midr(uint32_t midr, struct cpuInfo* cpu) {
return arch;
}
bool is_ARMv8_or_newer(struct cpuInfo* cpu) {
return cpu->arch->isa >= ISA_ARMv8_A;
}
bool has_fma_support(struct cpuInfo* cpu) {
// Arm A64 Instruction Set Architecture
// https://developer.arm.com/documentation/ddi0596/2021-12/SIMD-FP-Instructions
return is_ARMv8_or_newer(cpu);
}
int get_vpus_width(struct cpuInfo* cpu) {
// If the CPU has NEON, width can be 64 or 128 [1].
// In >= ARMv8, NEON are 128 bits width [2]
// If the CPU has SVE/SVE2, width can be between 128-2048 [3],
// so we get the exact value from cntb [4]
//
// [1] https://en.wikipedia.org/wiki/ARM_architecture_family#Advanced_SIMD_(Neon)
// [2] https://developer.arm.com/documentation/102474/0100/Fundamentals-of-Armv8-Neon-technology
// [3] https://www.anandtech.com/show/16640/arm-announces-neoverse-v1-n2-platforms-cpus-cmn700-mesh/5
// [4] https://developer.arm.com/documentation/ddi0596/2020-12/SVE-Instructions/CNTB--CNTD--CNTH--CNTW--Set-scalar-to-multiple-of-predicate-constraint-element-count-
if (cpu->feat->SVE && cpu->feat->cntb > 0) {
return cpu->feat->cntb * 8;
}
else if (cpu->feat->NEON) {
if(is_ARMv8_or_newer(cpu)) {
return 128;
}
else {
return 64;
}
}
else {
return 32;
}
}
int get_number_of_vpus(struct cpuInfo* cpu) {
MICROARCH ua = cpu->arch->uarch;
switch(ua) {
case UARCH_CORTEX_X925: // [https://www.anandtech.com/show/21399/arm-unveils-2024-cpu-core-designs-cortex-x925-a725-and-a520-arm-v9-2-redefined-for-3nm-/2]
return 6;
case UARCH_EVEREST: // Just a guess, needs confirmation.
case UARCH_FIRESTORM: // [https://dougallj.github.io/applecpu/firestorm-simd.html]
case UARCH_AVALANCHE: // [https://en.wikipedia.org/wiki/Comparison_of_ARM_processors]
case UARCH_CORTEX_X1: // [https://www.anandtech.com/show/15813/arm-cortex-a78-cortex-x1-cpu-ip-diverging/3]
case UARCH_CORTEX_X1C: // Assuming same as X1
case UARCH_CORTEX_X2: // [https://www.anandtech.com/show/16693/arm-announces-mobile-armv9-cpu-microarchitectures-cortexx2-cortexa710-cortexa510/2]
case UARCH_CORTEX_X3: // [https://www.hwcooling.net/en/cortex-x3-the-new-fastest-arm-core-architecture-analysis: "The FPU and SIMD unit of the core still has four pipelines"]
case UARCH_CORTEX_X4: // [https://www.anandtech.com/show/18871/arm-unveils-armv92-mobile-architecture-cortex-x4-a720-and-a520-64bit-exclusive/2]: "Cortex-X4: Out-of-Order Core"
case UARCH_NEOVERSE_V1: // [https://en.wikichip.org/wiki/arm_holdings/microarchitectures/neoverse_v1]
case UARCH_NEOVERSE_V2: // [https://chipsandcheese.com/2023/09/11/hot-chips-2023-arms-neoverse-v2/]
case UARCH_NEOVERSE_V3: // Assuming same as V2
return 4;
case UARCH_SAWTOOTH: // Needs confirmation, rn this is the best we know: https://mastodon.social/@dougall/111118317031041336
case UARCH_EXYNOS_M3: // [https://www.anandtech.com/show/12361/samsung-exynos-m3-architecture]
case UARCH_EXYNOS_M4: // [https://en.wikichip.org/wiki/samsung/microarchitectures/m4#Block_Diagram]
case UARCH_EXYNOS_M5: // [https://en.wikichip.org/wiki/samsung/microarchitectures/m5]
return 3;
case UARCH_ICESTORM: // [https://dougallj.github.io/applecpu/icestorm-simd.html]
case UARCH_BLIZZARD: // [https://en.wikipedia.org/wiki/Comparison_of_ARM_processors]
case UARCH_TAISHAN_V110:// [https://www-file.huawei.com/-/media/corp2020/pdf/publications/huawei-research/2022/huawei-research-issue1-en.pdf]: "128-bit x 2 for single precision"
case UARCH_TAISHAN_V120:// Not confirmed, asssuming same as v110
case UARCH_TAISHAN_V200:// Not confirmed, asssuming same as v110
case UARCH_CORTEX_A57: // [https://www.anandtech.com/show/8718/the-samsung-galaxy-note-4-exynos-review/5]
case UARCH_CORTEX_A72: // [https://www.anandtech.com/show/10347/arm-cortex-a73-artemis-unveiled/2]
case UARCH_CORTEX_A73: // [https://www.anandtech.com/show/10347/arm-cortex-a73-artemis-unveiled/2]
case UARCH_CORTEX_A75: // [https://www.anandtech.com/show/11441/dynamiq-and-arms-new-cpus-cortex-a75-a55/3]
case UARCH_CORTEX_A76: // [https://www.anandtech.com/show/12785/arm-cortex-a76-cpu-unveiled-7nm-powerhouse/3]
case UARCH_CORTEX_A77: // [https://fuse.wikichip.org/news/2339/arm-unveils-cortex-a77-emphasizes-single-thread-performance]
case UARCH_CORTEX_A78: // [https://fuse.wikichip.org/news/3536/arm-unveils-the-cortex-a78-when-less-is-more]
case UARCH_CORTEX_A78C: // Assuming same as A78
case UARCH_CORTEX_A78AE:// Assuming same as A78
case UARCH_EXYNOS_M1: // [https://www.anandtech.com/show/12361/samsung-exynos-m3-architecture]
case UARCH_EXYNOS_M2: // [https://www.anandtech.com/show/12361/samsung-exynos-m3-architecture]
case UARCH_NEOVERSE_N1: // [https://en.wikichip.org/wiki/arm_holdings/microarchitectures/neoverse_n1#Individual_Core]
case UARCH_NEOVERSE_N2: // [https://chipsandcheese.com/2023/08/18/arms-neoverse-n2-cortex-a710-for-servers/]
case UARCH_CORTEX_A710: // [https://chipsandcheese.com/2023/08/11/arms-cortex-a710-winning-by-default/]: Fig in Core Overview. Table in Instruction Scheduling and Execution
case UARCH_CORTEX_A715: // [https://www.hwcooling.net/en/arm-introduces-new-cortex-a715-core-architecture-analysis/]: "the numbers of ALU and FPU execution units themselves >
case UARCH_CORTEX_A720: // Assuming same as A715: https://www.anandtech.com/show/18871/arm-unveils-armv92-mobile-architecture-cortex-x4-a720-and-a520-64bit-exclusive/3
case UARCH_CORTEX_A725: // Assuming same as A720
return 2;
case UARCH_NEOVERSE_E1: // [https://www.anandtech.com/show/13959/arm-announces-neoverse-n1-platform/5]
// A510 is integrated as part of a Complex. Normally, each complex would incorporate two Cortex-A510 cores.
// Each complex incorporates a single VPU with 2 ports, so for each A510 there is theoretically 1 port.
case UARCH_CORTEX_A510: // [https://en.wikichip.org/wiki/arm_holdings/microarchitectures/cortex-a510#Vector_Processing_Unit_.28VPU.29]
case UARCH_CORTEX_A520: // Assuming same as A50: https://www.anandtech.com/show/18871/arm-unveils-armv92-mobile-architecture-cortex-x4-a720-and-a520-64bit-exclusive/4
return 1;
default:
// ARMv6
// ARMv7
// Remaining UARCH_CORTEX_AXX
// Old Snapdragon (e.g., Scorpion, Krait, etc)
return 1;
}
}
char* get_str_uarch(struct cpuInfo* cpu) {
return cpu->arch->uarch_str;
}
MICROARCH get_uarch(struct uarch* arch) {
return arch->uarch;
}
void free_uarch_struct(struct uarch* arch) {
free(arch->uarch_str);
free(arch);

View File

@@ -5,8 +5,111 @@
#include "midr.h"
enum {
UARCH_UNKNOWN,
// ARM
UARCH_ARM7,
UARCH_ARM9,
UARCH_ARM1136,
UARCH_ARM1156,
UARCH_ARM1176,
UARCH_ARM11MPCORE,
UARCH_CORTEX_A5,
UARCH_CORTEX_A7,
UARCH_CORTEX_A8,
UARCH_CORTEX_A9,
UARCH_CORTEX_A12,
UARCH_CORTEX_A15,
UARCH_CORTEX_A17,
UARCH_CORTEX_A32,
UARCH_CORTEX_A35,
UARCH_CORTEX_A53,
UARCH_CORTEX_A55r0, // ARM Cortex-A55 revision 0 (restricted dual-issue capabilities compared to revision 1+).
UARCH_CORTEX_A55,
UARCH_CORTEX_A57,
UARCH_CORTEX_A65,
UARCH_CORTEX_A72,
UARCH_CORTEX_A73,
UARCH_CORTEX_A75,
UARCH_CORTEX_A76,
UARCH_CORTEX_A77,
UARCH_CORTEX_A78,
UARCH_CORTEX_A78AE,
UARCH_CORTEX_A78C,
UARCH_CORTEX_A510,
UARCH_CORTEX_A520,
UARCH_CORTEX_A710,
UARCH_CORTEX_A715,
UARCH_CORTEX_A720,
UARCH_CORTEX_A725,
UARCH_CORTEX_X1,
UARCH_CORTEX_X1C,
UARCH_CORTEX_X2,
UARCH_CORTEX_X3,
UARCH_CORTEX_X4,
UARCH_CORTEX_X925,
UARCH_NEOVERSE_N1,
UARCH_NEOVERSE_N2,
UARCH_NEOVERSE_E1,
UARCH_NEOVERSE_V1,
UARCH_NEOVERSE_V2,
UARCH_NEOVERSE_V3,
UARCH_SCORPION,
UARCH_KRAIT,
UARCH_KYRO,
UARCH_FALKOR,
UARCH_SAPHIRA,
UARCH_DENVER,
UARCH_DENVER2,
UARCH_CARMEL,
// SAMSUNG
UARCH_EXYNOS_M1, // Samsung Exynos M1 (Exynos 8890 big cores)
UARCH_EXYNOS_M2, // Samsung Exynos M2 (Exynos 8895 big cores)
UARCH_EXYNOS_M3, // Samsung Exynos M3 (Exynos 9810 big cores)
UARCH_EXYNOS_M4, // Samsung Exynos M4 (Exynos 9820 big cores)
UARCH_EXYNOS_M5, // Samsung Exynos M5 (Exynos 9830 big cores)
// APPLE
UARCH_SWIFT, // Apple A6 and A6X processors.
UARCH_CYCLONE, // Apple A7 processor.
UARCH_TYPHOON, // Apple A8 and A8X processor
UARCH_TWISTER, // Apple A9 and A9X processor.
UARCH_HURRICANE, // Apple A10 and A10X processor.
UARCH_MONSOON, // Apple A11 processor (big cores).
UARCH_MISTRAL, // Apple A11 processor (little cores).
UARCH_VORTEX, // Apple A12 processor (big cores).
UARCH_TEMPEST, // Apple A12 processor (big cores).
UARCH_LIGHTNING, // Apple A13 processor (big cores).
UARCH_THUNDER, // Apple A13 processor (little cores).
UARCH_ICESTORM, // Apple M1 processor (little cores).
UARCH_FIRESTORM, // Apple M1 processor (big cores).
UARCH_BLIZZARD, // Apple M2 processor (little cores).
UARCH_AVALANCHE, // Apple M2 processor (big cores).
UARCH_SAWTOOTH, // Apple M3 processor (little cores).
UARCH_EVEREST, // Apple M3 processor (big cores).
// CAVIUM
UARCH_THUNDERX, // Cavium ThunderX
UARCH_THUNDERX2, // Cavium ThunderX2 (originally Broadcom Vulkan).
// MARVELL
UARCH_PJ4,
UARCH_BRAHMA_B15,
UARCH_BRAHMA_B53,
UARCH_XGENE, // Applied Micro X-Gene.
// HUAWEI
UARCH_TAISHAN_V110, // HiSilicon TaiShan v110
UARCH_TAISHAN_V120, // HiSilicon TaiShan v120
UARCH_TAISHAN_V200, // HiSilicon TaiShan v200
// PHYTIUM
UARCH_XIAOMI, // Not to be confused with Xiaomi Inc
};
typedef uint32_t MICROARCH;
struct uarch* get_uarch_from_midr(uint32_t midr, struct cpuInfo* cpu);
int get_number_of_vpus(struct cpuInfo* cpu);
int get_vpus_width(struct cpuInfo* cpu);
bool has_fma_support(struct cpuInfo* cpu);
char* get_str_uarch(struct cpuInfo* cpu);
void free_uarch_struct(struct uarch* arch);
MICROARCH get_uarch(struct uarch* arch);
#endif

View File

@@ -3,8 +3,6 @@
#include "midr.h"
#define _PATH_DEVICETREE_MODEL "/sys/firmware/devicetree/base/model"
#define _PATH_CPUINFO "/proc/cpuinfo"
//#define _PATH_CPUINFO "cpuinfo_debug"
#define CPUINFO_CPU_IMPLEMENTER_STR "CPU implementer\t: "
#define CPUINFO_CPU_ARCHITECTURE_STR "CPU architecture: "
@@ -108,27 +106,6 @@ uint32_t get_midr_from_cpuinfo(uint32_t core, bool* success) {
return midr;
}
char* get_field_from_cpuinfo(char* CPUINFO_FIELD) {
int filelen;
char* buf;
if((buf = read_file(_PATH_CPUINFO, &filelen)) == NULL) {
printWarn("read_file: %s: %s:\n", _PATH_CPUINFO, strerror(errno));
return NULL;
}
char* tmp1 = strstr(buf, CPUINFO_FIELD);
if(tmp1 == NULL) return NULL;
tmp1 = tmp1 + strlen(CPUINFO_FIELD);
char* tmp2 = strstr(tmp1, "\n");
int strlen = (1 + (tmp2-tmp1));
char* hardware = emalloc(sizeof(char) * strlen);
memset(hardware, 0, sizeof(char) * strlen);
strncpy(hardware, tmp1, tmp2-tmp1);
return hardware;
}
char* get_hardware_from_cpuinfo(void) {
return get_field_from_cpuinfo(CPUINFO_HARDWARE_STR);
}

View File

@@ -5,6 +5,7 @@
#define _PATH_SUNXI_NVMEM "/sys/bus/nvmem/devices/sunxi-sid0/nvmem"
#define _PATH_RK_EFUSE0 "/sys/bus/nvmem/devices/rockchip-efuse0/nvmem"
#define _PATH_RK_OTP0 "/sys/bus/nvmem/devices/rockchip-otp0/nvmem"
#define UNKNOWN -1
int get_ncores_from_cpuinfo(void);

View File

@@ -28,6 +28,7 @@ struct args_struct {
bool help_flag;
bool raw_flag;
bool accurate_pp;
bool measure_max_frequency_flag;
bool full_cpu_name_flag;
bool logo_long;
bool logo_short;
@@ -50,6 +51,7 @@ const char args_chr[] = {
/* [ARG_LOGO_INTEL_NEW] = */ 3,
/* [ARG_LOGO_INTEL_OLD] = */ 4,
/* [ARG_ACCURATE_PP] = */ 5,
/* [ARG_MEASURE_MAX_FREQ] = */ 6,
/* [ARG_DEBUG] = */ 'd',
/* [ARG_VERBOSE] = */ 'v',
/* [ARG_VERSION] = */ 'V',
@@ -66,6 +68,7 @@ const char *args_str[] = {
/* [ARG_LOGO_INTEL_NEW] = */ "logo-intel-new",
/* [ARG_LOGO_INTEL_OLD] = */ "logo-intel-old",
/* [ARG_ACCURATE_PP] = */ "accurate-pp",
/* [ARG_MEASURE_MAX_FREQ] = */ "measure-max-freq",
/* [ARG_DEBUG] = */ "debug",
/* [ARG_VERBOSE] = */ "verbose",
/* [ARG_VERSION] = */ "version",
@@ -101,6 +104,10 @@ bool accurate_pp(void) {
return args.accurate_pp;
}
bool measure_max_frequency_flag(void) {
return args.measure_max_frequency_flag;
}
bool show_full_cpu_name(void) {
return args.full_cpu_name_flag;
}
@@ -218,16 +225,23 @@ bool parse_color(char* optarg_str, struct color*** cs) {
char* build_short_options(void) {
const char *c = args_chr;
int len = sizeof(args_chr) / sizeof(args_chr[0]);
char* str = (char *) emalloc(sizeof(char) * (len*2 + 1));
memset(str, 0, sizeof(char) * (len*2 + 1));
char* str = (char *) ecalloc(len*2 + 1, sizeof(char));
#ifdef ARCH_X86
sprintf(str, "%c:%c:%c%c%c%c%c%c%c%c%c%c%c",
sprintf(str, "%c:%c:%c%c%c%c%c%c%c%c%c%c%c%c",
c[ARG_STYLE], c[ARG_COLOR], c[ARG_HELP],
c[ARG_RAW], c[ARG_FULLCPUNAME],
c[ARG_LOGO_SHORT], c[ARG_LOGO_LONG],
c[ARG_LOGO_INTEL_NEW], c[ARG_LOGO_INTEL_OLD],
c[ARG_ACCURATE_PP], c[ARG_DEBUG], c[ARG_VERBOSE],
c[ARG_ACCURATE_PP], c[ARG_MEASURE_MAX_FREQ],
c[ARG_DEBUG], c[ARG_VERBOSE],
c[ARG_VERSION]);
#elif ARCH_ARM
sprintf(str, "%c:%c:%c%c%c%c%c%c%c",
c[ARG_STYLE], c[ARG_COLOR], c[ARG_HELP],
c[ARG_LOGO_SHORT], c[ARG_LOGO_LONG],
c[ARG_MEASURE_MAX_FREQ],
c[ARG_DEBUG], c[ARG_VERBOSE],
c[ARG_VERSION]);
#else
sprintf(str, "%c:%c:%c%c%c%c%c%c",
@@ -270,8 +284,11 @@ bool parse_args(int argc, char* argv[]) {
{args_str[ARG_LOGO_INTEL_NEW], no_argument, 0, args_chr[ARG_LOGO_INTEL_NEW] },
{args_str[ARG_LOGO_INTEL_OLD], no_argument, 0, args_chr[ARG_LOGO_INTEL_OLD] },
{args_str[ARG_ACCURATE_PP], no_argument, 0, args_chr[ARG_ACCURATE_PP] },
{args_str[ARG_MEASURE_MAX_FREQ], no_argument, 0, args_chr[ARG_MEASURE_MAX_FREQ] },
{args_str[ARG_FULLCPUNAME], no_argument, 0, args_chr[ARG_FULLCPUNAME] },
{args_str[ARG_RAW], no_argument, 0, args_chr[ARG_RAW] },
#elif ARCH_ARM
{args_str[ARG_MEASURE_MAX_FREQ], no_argument, 0, args_chr[ARG_MEASURE_MAX_FREQ] },
#endif
{args_str[ARG_LOGO_SHORT], no_argument, 0, args_chr[ARG_LOGO_SHORT] },
{args_str[ARG_LOGO_LONG], no_argument, 0, args_chr[ARG_LOGO_LONG] },
@@ -313,6 +330,9 @@ bool parse_args(int argc, char* argv[]) {
else if(opt == args_chr[ARG_ACCURATE_PP]) {
args.accurate_pp = true;
}
else if(opt == args_chr[ARG_MEASURE_MAX_FREQ]) {
args.measure_max_frequency_flag = true;
}
else if(opt == args_chr[ARG_FULLCPUNAME]) {
args.full_cpu_name_flag = true;
}

View File

@@ -29,6 +29,7 @@ enum {
ARG_LOGO_INTEL_NEW,
ARG_LOGO_INTEL_OLD,
ARG_ACCURATE_PP,
ARG_MEASURE_MAX_FREQ,
ARG_DEBUG,
ARG_VERBOSE,
ARG_VERSION
@@ -43,6 +44,7 @@ int max_arg_str_length(void);
bool parse_args(int argc, char* argv[]);
bool show_help(void);
bool accurate_pp(void);
bool measure_max_frequency_flag(void);
bool show_full_cpu_name(void);
bool show_logo_long(void);
bool show_logo_short(void);

View File

@@ -33,7 +33,7 @@ struct ascii_logo {
uint32_t width;
uint32_t height;
bool replace_blocks;
char color_ascii[3][100];
char color_ascii[4][100];
char color_text[2][100];
};
@@ -105,6 +105,19 @@ $C1 MMM :MMM NMM dMMK dMMX MMN \
$C1 MMM :MMM NMM dMMMoo OMM0....:Nx. MMN \
$C1 MMM :WWW XWW lONMM 'xXMMMMNOc MMN "
#define ASCII_HYGON \
"$C1 \
$C1 \
$C1 \
$C1 ## ## ## ## ###### ###### ## # \
$C1 ##....## ## ## ## ## ## #### # \
$C1 ######## ## ## ##. ## ## # #### \
$C1 ## ## ## *######. ###### # ## \
$C1 \
$C1 \
$C1 \
$C1 "
#define ASCII_SNAPD \
" $C1@@$C2######## \
$C1@@@@@$C2########### \
@@ -145,6 +158,25 @@ $C2 Exynos \
$C2 \
$C2 "
#define ASCII_KUNPENG \
"$C2 . \
$C2 .. \
$C2 .## \
$C1 .$CR $C2.###. \
$C1 ..$CR $C2#####. \
$C1 .#.$CR $C2.#######. \
$C1 .####.$CR $C2.#######. \
$C1 ..######*$CR $C2.#######. . \
$C1 .#########*$CR $C2.#######* . \
$C1 ######*$CR $C2.#######. .#. \
$C1*#######*$CR $C2.#######. *##. \
$C1 ##*$CR $C2.#######. ####### \
$C2 ###.$CR $C1#####$C2 *### \
$C1 *########## \
$C1 *######## \
$C1 #####. \
$C1 *###. "
#define ASCII_KIRIN \
"$C1 ####### \
$C1 ##### #################### \
@@ -218,6 +250,23 @@ $C1 kMMMMMMMMMMMMMMMMMMMMMMd \
$C1 'KMMMMMMMWXXWMMMMMMMk. \
$C1 \"cooc\"* \"*coo'\" "
#define ASCII_GOOGLE \
"$C1 aaaaaaaa \
$C1 .MMMMMMMMMMMMMMMM. \
$C1 .MMMMMMMMMMMMMMMMMMMMM* \
$C1 .MMMMMMMM** *MM* \
$C2 MM$C1MMMMM. \
$C2 *MMM$C1MM. \
$C2*MMMMMM $C4lMMMMMMMMMMMMMMM* \
$C2MMMMMMM $C4lMMMMMMMMMMMMMMMM \
$C2*MMMMMM $C4lMMMMMMMMMMMMMMMd \
$C2 MMMM$C3MMM. $C4*MMMMM. \
$C2 MM$C3MMMMMM. $C4.MMMMMM. \
$C3 *MMMMMMMMM*.......MMM$C4MMMMMM* \
$C3 *MMMMMMMMMMMMMMMMMMMM$C4MM* \
$C3 *MMMMMMMMMMMMMM* \
$C3 ****** "
#define ASCII_ALLWINNER \
"$C1 \
$C1 ################# \
@@ -287,29 +336,6 @@ $C1 ################@@@@@@##@@@@@@################ \
$C1 ####################@@@@@@#################### \
$C1 ############################################## "
#define ASCII_SIFIVE_L \
"$C1 ################################################### \
$C1 ###########@@@@@@@@@@@@@@@@@@@@@@@@@@@@############ \
$C1 ##########@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@########### \
$C1 #########@@@@@#######################@@@@########## \
$C1 ########@@@@@#########################@@@@######### \
$C1 #######@@@@@###########################@@@@######## \
$C1 ######@@@@@########@@@@@@@@@@@@@@@@@@@@@@@@@####### \
$C1 #####@@@@@########@@@@@@@@@@@@@@@@@@@@@@@@@@@###### \
$C1 ####@@@@@################################@@@@@##### \
$C1 ###@@@@@##################################@@@@@#### \
$C1 ##@@@@@####################################@@@@@### \
$C1 ##@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@#########@@@@### \
$C1 ##@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@############@@@@### \
$C1 ####@@@@@#############@@@@@@@############@@@@@##### \
$C1 #######@@@@@@############@############@@@@@@####### \
$C1 ##########@@@@@@###################@@@@@@########## \
$C1 ##############@@@@@@###########@@@@@@############## \
$C1 #################@@@@@@#####@@@@@@################# \
$C1 ####################@@@@@@@@@@@#################### \
$C1 ########################@@@######################## \
$C1 ################################################### "
#define ASCII_STARFIVE \
"$C1 # \
$C1 ########## \
@@ -329,30 +355,95 @@ $C1 ######## ######## \
$C1 ######### \
$C1 # "
#define ASCII_STARFIVE_L \
"$C1 ####### \
$C1 ################. \
$C1 ############ ########### \
$C1 ############ ##########. \
$C1 ############ # ###### \
$C1 ########### ##### ## \
$C1 #######. ########## \
$C1 ###### ### *########### \
$C1 ###### #######. ########## \
$C1 ######### ############ ###### \
$C1 ###########. ###########* # \
$C1 ############ ############ \
$C1 # ############. .########### \
$C1 ###### ########### ######### \
$C1 ########## .######, ##### \
$C1 ############ ##. #####. \
#define ASCII_SIPEED \
"$C1 #################################### \
$C1######@@################################ \
$C1####@@##@@####@@@@@@@@@@@@@@@@########## \
$C1######@@####@@@@@@@@@@@@@@@@@@########## \
$C1#########@@@@########################### \
$C1#######@@@@@@########################### \
$C1#########@@@@########################### \
$C1###########@@@@@@@@@@@@@@@############## \
$C1##############@@@@@@@@@@@@@@@########### \
$C1###########################@@@@@######## \
$C1###########################@@@@@@####### \
$C1###########################@@@@@######## \
$C1##########@@@@@@@@@@@@@@@@@@@########### \
$C1##########@@@@@@@@@@@@@@@@############## \
$C1######################################## \
$C1 #################################### "
#define ASCII_NVIDIA \
"$C1 'cccccccccccccccccccccccccc \
$C1 ;oooooooooooooooooooooooool \
$C1 .:::. .oooooooooooooooooool \
$C1 .:cll; ,c:::. cooooooooooooool \
$C1 ,clo' ;. oolc: ooooooooooool \
$C1.cloo ;cclo . .olc. coooooooool \
$C1oooo :lo, ;ll; looc :oooooooool \
$C1 oooc ool. ;oooc;clol :looooooooool \
$C1 :ooc ,ol; ;oooooo. .cloo; loool \
$C1 ool; .olc. ,:lool .lool \
$C1 ool:. ,::::ccloo. :clooool \
$C1 oolc::. ':cclooooooool \
$C1 ;oooooooooooooooooooooooool \
$C1 \
$C1 \
$C2######. ## ## ## ###### ## ### \
$C2## ## ## ## ## ## ## ## #: :# \
$C2## ## ## ## ## ## ## ## ####### \
$C2## ## ### ## ###### ## ## ## "
#define ASCII_AMPERE \
"$C1 \
$C1 \
$C1 ## \
$C1 #### \
$C1 ### ## \
$C1 ### ### \
$C1 ### ### \
$C1 ### ### \
$C1 ## ### \
$C1 ####### ### ### \
$C1 ###### ## ###### ### \
$C1 ## ##### ##########. \
$C1 ####### # ############ \
$C1 ########### ###########. \
$C1 ###########. ############ \
$C1 ################ \
$C1 ####### "
$C1 #### ### ######## \
$C1 #### ### #### \
$C1 ### ### #### \
$C1 ## ### ### \
$C1 \
$C1 "
#define ASCII_NXP \
"$C1##### # $C2####### ####### $C3########## \
$C1####### ## $C2####### ####### $C3############### \
$C1########## #### $C2###### ###### $C3### ###### \
$C1############ ##### $C2############ $C3##### ##### \
$C1##### ####### ##### $C2########## $C3################### \
$C1##### ######### $C2############## $C3############### \
$C1##### ###### $C2###### ###### $C3#### \
$C1##### ## $C2###### ###### $C3## "
#define ASCII_AMLOGIC \
"$C1 .#####. ### ### \
$C1 ######## ### \
$C1 ####..### ########## ### ### ##### ### ### \
$C1 .## #. ### ## ## ## ### ## ## ## ## ### ## \
$C1 #### #.# ### ## ## ## ### ## ## ## ## ### ## \
$C1#########.### ## ## ## ## ### ###### ## ### \
$C1 ### \
$C1 ### "
#define ASCII_MARVELL \
"$C1 ........... ........... \
$C1 .### . .## . \
$C1 .##### . #### . \
$C1 ####### . ####### . \
$C1 .#########__________. #########__________. \
$C1 .###########|__________|#########|__________| \
$C1 ############ ______############ __________ \
$C1 .######### |__________|###### |__________| \
$C1 ########### ___########### __________ \
$C1.########## |__________| |__________| "
// --------------------- LONG LOGOS ------------------------- //
#define ASCII_AMD_L \
@@ -438,29 +529,103 @@ $C1 ############ ################## ######### #### ######### \
$C1 \
$C1 ############ ################ ######### ## ######### "
#define ASCII_SIFIVE_L \
"$C1 ################################################### \
$C1 ###########@@@@@@@@@@@@@@@@@@@@@@@@@@@@############ \
$C1 ##########@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@########### \
$C1 #########@@@@@#######################@@@@########## \
$C1 ########@@@@@#########################@@@@######### \
$C1 #######@@@@@###########################@@@@######## \
$C1 ######@@@@@########@@@@@@@@@@@@@@@@@@@@@@@@@####### \
$C1 #####@@@@@########@@@@@@@@@@@@@@@@@@@@@@@@@@@###### \
$C1 ####@@@@@################################@@@@@##### \
$C1 ###@@@@@##################################@@@@@#### \
$C1 ##@@@@@####################################@@@@@### \
$C1 ##@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@#########@@@@### \
$C1 ##@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@############@@@@### \
$C1 ####@@@@@#############@@@@@@@############@@@@@##### \
$C1 #######@@@@@@############@############@@@@@@####### \
$C1 ##########@@@@@@###################@@@@@@########## \
$C1 ##############@@@@@@###########@@@@@@############## \
$C1 #################@@@@@@#####@@@@@@################# \
$C1 ####################@@@@@@@@@@@#################### \
$C1 ########################@@@######################## \
$C1 ################################################### "
#define ASCII_STARFIVE_L \
"$C1 ####### \
$C1 ################. \
$C1 ############ ########### \
$C1 ############ ##########. \
$C1 ############ # ###### \
$C1 ########### ##### ## \
$C1 #######. ########## \
$C1 ###### ### *########### \
$C1 ###### #######. ########## \
$C1 ######### ############ ###### \
$C1 ###########. ###########* # \
$C1 ############ ############ \
$C1 # ############. .########### \
$C1 ###### ########### ######### \
$C1 ########## .######, ##### \
$C1 ############ ##. #####. \
$C1 ######### ######## \
$C1 ## ##### ##########. \
$C1 ####### # ############ \
$C1 ########### ###########. \
$C1 ###########. ############ \
$C1 ################ \
$C1 ####### "
#define ASCII_NVIDIA_L \
"$C1 MMMMMMMMMMMMMMMMMMMMMMMMMMMMMM \
$C1 MMMMMMMMMMMMMMMMMMMMMMMMMMMMMM \
$C1 .:: 'MMMMMMMMMMMMMMMMMMMMMMMMM \
$C1 ccllooo;:;. ;MMMMMMMMMMMMMMMMMM \
$C1 cloc :ooollcc: :MMMMMMMMMMMMMMM \
$C1 cloc :ccl; lolc, ;MMMMMMMMMMMM \
$C1.cloo: :clo ;c: .ool; MMMMMMMMMMM \
$C1 ooo: ooo :ool, .cloo. ;lMMMMMMMMMMM \
$C1 ooo: ooc :ooooccooo. :MMMM lMMMMMMM \
$C1 ooc. ool: :oooooo' ,cloo. MMMM \
$C1 ool:. olc: .:cloo. :MMMM \
$C1 olc, ;:::cccloo. :MMMMMMMM \
$C1 olcc::; ,:ccloMMMMMMMMM \
$C1 :......oMMMMMMMMMMMMMMMMMMMMMM \
$C1 :lllMMMMMMMMMMMMMMMMMMMMMMMMMM "
typedef struct ascii_logo asciiL;
// ------------------------------------------------------------------------------------------------------+
// +-----------------------------------------------------------------------------------------------------------------+
// | LOGO | W | H | REPLACE | COLORS LOGO (>0 && <10) | COLORS TEXT (=2) |
// ------------------------------------------------------------------------------------------------------+
// +-----------------------------------------------------------------------------------------------------------------+
asciiL logo_amd = { ASCII_AMD, 39, 15, false, {C_FG_WHITE, C_FG_GREEN}, {C_FG_WHITE, C_FG_GREEN} };
asciiL logo_intel = { ASCII_INTEL, 48, 14, false, {C_FG_CYAN}, {C_FG_CYAN, C_FG_WHITE} };
asciiL logo_intel_new = { ASCII_INTEL_NEW, 51, 9, false, {C_FG_CYAN}, {C_FG_CYAN, C_FG_WHITE} };
asciiL logo_hygon = { ASCII_HYGON, 51, 11, false, {C_FG_RED}, {C_FG_RED, C_FG_WHITE} };
asciiL logo_snapd = { ASCII_SNAPD, 39, 16, false, {C_FG_RED, C_FG_WHITE}, {C_FG_RED, C_FG_WHITE} };
asciiL logo_mtk = { ASCII_MTK, 59, 5, false, {C_FG_BLUE, C_FG_YELLOW}, {C_FG_BLUE, C_FG_YELLOW} };
asciiL logo_exynos = { ASCII_EXYNOS, 22, 13, true, {C_BG_BLUE, C_FG_WHITE}, {C_FG_BLUE, C_FG_WHITE} };
asciiL logo_kirin = { ASCII_KIRIN, 53, 12, false, {C_FG_RED}, {C_FG_WHITE, C_FG_RED} };
asciiL logo_kunpeng = { ASCII_KUNPENG, 48, 17, false, {C_FG_RED, C_FG_WHITE}, {C_FG_WHITE, C_FG_RED} };
asciiL logo_broadcom = { ASCII_BROADCOM, 44, 19, false, {C_FG_WHITE, C_FG_RED}, {C_FG_WHITE, C_FG_RED} };
asciiL logo_arm = { ASCII_ARM, 42, 5, false, {C_FG_CYAN}, {C_FG_WHITE, C_FG_CYAN} };
asciiL logo_ibm = { ASCII_IBM, 42, 9, false, {C_FG_CYAN, C_FG_WHITE}, {C_FG_CYAN, C_FG_WHITE} };
asciiL logo_apple = { ASCII_APPLE, 32, 17, false, {C_FG_WHITE}, {C_FG_CYAN, C_FG_B_WHITE} };
asciiL logo_google = { ASCII_GOOGLE, 35, 15, false, {C_FG_RED, C_FG_YELLOW, C_FG_GREEN, C_FG_BLUE}, {C_FG_BLUE, C_FG_B_WHITE} };
asciiL logo_allwinner = { ASCII_ALLWINNER, 47, 16, false, {C_FG_CYAN}, {C_FG_B_BLACK, C_FG_B_CYAN } };
asciiL logo_rockchip = { ASCII_ROCKCHIP, 58, 8, false, {C_FG_CYAN, C_FG_YELLOW}, {C_FG_CYAN, C_FG_YELLOW} };
asciiL logo_riscv = { ASCII_RISCV, 63, 18, false, {C_FG_CYAN, C_FG_YELLOW}, {C_FG_CYAN, C_FG_YELLOW} };
asciiL logo_sifive = { ASCII_SIFIVE, 48, 19, true, {C_BG_WHITE, C_BG_BLACK}, {C_FG_WHITE, C_FG_BLUE} };
asciiL logo_starfive = { ASCII_STARFIVE, 33, 17, false, {C_FG_WHITE}, {C_FG_WHITE, C_FG_BLUE} };
asciiL logo_sipeed = { ASCII_SIPEED, 41, 16, true, {C_BG_RED, C_BG_WHITE}, {C_FG_RED, C_FG_WHITE} };
asciiL logo_nvidia = { ASCII_NVIDIA, 45, 19, false, {C_FG_GREEN, C_FG_WHITE}, {C_FG_WHITE, C_FG_GREEN} };
asciiL logo_ampere = { ASCII_AMPERE, 50, 17, false, {C_FG_RED}, {C_FG_WHITE, C_FG_RED} };
asciiL logo_nxp = { ASCII_NXP, 55, 8, false, {C_FG_YELLOW, C_FG_CYAN, C_FG_GREEN}, {C_FG_CYAN, C_FG_WHITE} };
asciiL logo_amlogic = { ASCII_AMLOGIC, 58, 8, false, {C_FG_BLUE}, {C_FG_BLUE, C_FG_B_WHITE} };
asciiL logo_marvell = { ASCII_MARVELL, 56, 10, false, {C_FG_B_BLACK}, {C_FG_B_BLACK, C_FG_B_WHITE} };
// Long variants | ----------------------------------------------------------------------------------------------------|
// Long variants | ----------------------------------------------------------------------------------------------------------------|
asciiL logo_amd_l = { ASCII_AMD_L, 62, 19, true, {C_BG_WHITE, C_BG_GREEN}, {C_FG_WHITE, C_FG_GREEN} };
asciiL logo_intel_l = { ASCII_INTEL_L, 62, 19, true, {C_BG_CYAN, C_BG_WHITE}, {C_FG_CYAN, C_FG_WHITE} };
asciiL logo_intel_l_new = { ASCII_INTEL_L_NEW, 57, 14, true, {C_BG_CYAN, C_BG_WHITE, C_BG_BLUE}, {C_FG_CYAN, C_FG_WHITE} };
@@ -468,6 +633,7 @@ asciiL logo_arm_l = { ASCII_ARM_L, 60, 8, true, {C_BG_CYAN},
asciiL logo_ibm_l = { ASCII_IBM_L, 62, 13, true, {C_BG_CYAN, C_FG_WHITE}, {C_FG_CYAN, C_FG_WHITE} };
asciiL logo_starfive_l = { ASCII_STARFIVE_L, 50, 22, false, {C_FG_WHITE}, {C_FG_WHITE, C_FG_BLUE} };
asciiL logo_sifive_l = { ASCII_SIFIVE_L, 53, 21, true, {C_BG_WHITE, C_BG_BLACK}, {C_FG_WHITE, C_FG_CYAN} };
asciiL logo_nvidia_l = { ASCII_NVIDIA_L, 50, 15, false, {C_FG_GREEN, C_FG_WHITE}, {C_FG_WHITE, C_FG_GREEN} };
asciiL logo_unknown = { NULL, 0, 0, false, {COLOR_NONE}, {COLOR_NONE, COLOR_NONE} };
#endif

View File

@@ -34,6 +34,12 @@ int64_t get_freq(struct frequency* freq) {
return freq->max;
}
#ifdef ARCH_X86
int64_t get_freq_pp(struct frequency* freq) {
return freq->max_pp;
}
#endif
#if defined(ARCH_X86) || defined(ARCH_PPC)
char* get_str_cpu_name(struct cpuInfo* cpu, bool fcpuname) {
#ifdef ARCH_X86
@@ -145,17 +151,25 @@ char* get_str_l3(struct cache* cach) {
char* get_str_freq(struct frequency* freq) {
//Max 3 digits and 3 for '(M/G)Hz' plus 1 for '\0'
uint32_t size = (5+1+3+1);
uint32_t size = (1+5+1+3+1);
assert(strlen(STRING_UNKNOWN)+1 <= size);
char* string = emalloc(sizeof(char)*size);
memset(string, 0, sizeof(char)*size);
char* string = ecalloc(size, sizeof(char));
if(freq->max == UNKNOWN_DATA || freq->max < 0)
if(freq->max == UNKNOWN_DATA || freq->max < 0) {
snprintf(string,strlen(STRING_UNKNOWN)+1,STRING_UNKNOWN);
else if(freq->max >= 1000)
}
else if(freq->max >= 1000) {
if (freq->measured)
snprintf(string,size,"~%.3f "STRING_GIGAHERZ,(float)(freq->max)/1000);
else
snprintf(string,size,"%.3f "STRING_GIGAHERZ,(float)(freq->max)/1000);
}
else {
if (freq->measured)
snprintf(string,size,"~%d "STRING_MEGAHERZ,freq->max);
else
snprintf(string,size,"%d "STRING_MEGAHERZ,freq->max);
}
return string;
}
@@ -194,7 +208,7 @@ void init_topology_struct(struct topology* topo, struct cache* cach) {
topo->sockets = 0;
#ifdef ARCH_X86
topo->smt_available = 0;
topo->apic = emalloc(sizeof(struct apic));
topo->apic = ecalloc(1, sizeof(struct apic));
#endif
#endif
}

View File

@@ -8,6 +8,7 @@ enum {
// ARCH_X86
CPU_VENDOR_INTEL,
CPU_VENDOR_AMD,
CPU_VENDOR_HYGON,
// ARCH_ARM
CPU_VENDOR_ARM,
CPU_VENDOR_APPLE,
@@ -16,7 +17,7 @@ enum {
CPU_VENDOR_NVIDIA,
CPU_VENDOR_APM,
CPU_VENDOR_QUALCOMM,
CPU_VENDOR_HUAWUEI,
CPU_VENDOR_HUAWEI,
CPU_VENDOR_SAMSUNG,
CPU_VENDOR_MARVELL,
CPU_VENDOR_PHYTIUM,
@@ -32,10 +33,14 @@ enum {
enum {
HV_VENDOR_KVM,
HV_VENDOR_QEMU,
HV_VENDOR_VBOX,
HV_VENDOR_HYPERV,
HV_VENDOR_VMWARE,
HV_VENDOR_XEN,
HV_VENDOR_PARALLELS,
HV_VENDOR_PHYP,
HV_VENDOR_BHYVE,
HV_VENDOR_APPLEVZ,
HV_VENDOR_INVALID
};
@@ -53,6 +58,13 @@ typedef int32_t VENDOR;
struct frequency {
int32_t base;
int32_t max;
// Indicates if max frequency was measured
bool measured;
#ifdef ARCH_X86
// Max frequency when running vectorized code.
// Used only for peak performance computation.
int32_t max_pp;
#endif
};
struct hypervisor {
@@ -117,12 +129,15 @@ struct features {
bool SHA1;
bool SHA2;
bool CRC32;
bool SVE;
bool SVE2;
uint64_t cntb;
#endif
};
struct extensions {
char* str;
uint32_t mask;
uint64_t mask;
};
struct cpuInfo {
@@ -178,6 +193,8 @@ struct cpuInfo {
#ifdef ARCH_X86
// The index of the first core in the module
uint32_t first_core_id;
// The index of this module
uint32_t module_id;
#endif
#endif
};
@@ -190,6 +207,9 @@ uint32_t get_nsockets(struct topology* topo);
VENDOR get_cpu_vendor(struct cpuInfo* cpu);
int64_t get_freq(struct frequency* freq);
#ifdef ARCH_X86
int64_t get_freq_pp(struct frequency* freq);
#endif
char* get_str_aes(struct cpuInfo* cpu);
char* get_str_sha(struct cpuInfo* cpu);

195
src/common/freq.c Normal file
View File

@@ -0,0 +1,195 @@
#ifdef __linux__
#define _GNU_SOURCE
#include <time.h>
#include <stdio.h>
#include <stdint.h>
#include <string.h>
#include <errno.h>
#include <unistd.h>
#include <asm/unistd.h>
#include <sys/ioctl.h>
#include <linux/perf_event.h>
#include "global.h"
#include "cpu.h"
static long
perf_event_open(struct perf_event_attr *hw_event, pid_t pid,
int cpu, int group_fd, unsigned long flags) {
int ret;
ret = syscall(__NR_perf_event_open, hw_event, pid, cpu,
group_fd, flags);
return ret;
}
#define INSERT_ASM_ONCE __asm volatile("nop");
#define INSERT_ASM_10_TIMES \
INSERT_ASM_ONCE \
INSERT_ASM_ONCE \
INSERT_ASM_ONCE \
INSERT_ASM_ONCE \
INSERT_ASM_ONCE \
INSERT_ASM_ONCE \
INSERT_ASM_ONCE \
INSERT_ASM_ONCE \
INSERT_ASM_ONCE \
INSERT_ASM_ONCE \
#define INSERT_ASM_100_TIMES \
INSERT_ASM_10_TIMES \
INSERT_ASM_10_TIMES \
INSERT_ASM_10_TIMES \
INSERT_ASM_10_TIMES \
INSERT_ASM_10_TIMES \
INSERT_ASM_10_TIMES \
INSERT_ASM_10_TIMES \
INSERT_ASM_10_TIMES \
INSERT_ASM_10_TIMES \
INSERT_ASM_10_TIMES
#define INSERT_ASM_1000_TIMES \
INSERT_ASM_100_TIMES \
INSERT_ASM_100_TIMES \
INSERT_ASM_100_TIMES \
INSERT_ASM_100_TIMES \
INSERT_ASM_100_TIMES \
INSERT_ASM_100_TIMES \
INSERT_ASM_100_TIMES \
INSERT_ASM_100_TIMES \
INSERT_ASM_100_TIMES \
INSERT_ASM_100_TIMES \
void nop_function(uint64_t iters) {
for (uint64_t i = 0; i < iters; i++) {
INSERT_ASM_1000_TIMES
INSERT_ASM_1000_TIMES
INSERT_ASM_1000_TIMES
INSERT_ASM_1000_TIMES
}
}
// Run the nop_function with the number of iterations specified and
// measure both the time and number of cycles
int measure_freq_iters(uint64_t iters, uint32_t core, double* freq) {
clockid_t clock = CLOCK_PROCESS_CPUTIME_ID;
struct timespec start, end;
struct perf_event_attr pe;
uint64_t cycles;
int fd;
int pid = 0;
memset(&pe, 0, sizeof(struct perf_event_attr));
pe.type = PERF_TYPE_HARDWARE;
pe.size = sizeof(struct perf_event_attr);
pe.config = PERF_COUNT_HW_CPU_CYCLES;
pe.disabled = 1;
pe.exclude_kernel = 1;
pe.exclude_hv = 1;
fd = perf_event_open(&pe, pid, core, -1, 0);
if (fd == -1) {
perror("perf_event_open");
if (errno == EPERM || errno == EACCES) {
printErr("You may not have permission to collect stats.\n"\
"Consider tweaking /proc/sys/kernel/perf_event_paranoid or running as root");
}
return -1;
}
if (clock_gettime(clock, &start) == -1) {
perror("clock_gettime");
return -1;
}
if(ioctl(fd, PERF_EVENT_IOC_RESET, 0) == -1) {
perror("ioctl");
return -1;
}
if(ioctl(fd, PERF_EVENT_IOC_ENABLE, 0) == -1) {
perror("ioctl");
return -1;
}
nop_function(iters);
ssize_t ret = read(fd, &cycles, sizeof(uint64_t));
if (ret == -1) {
perror("read");
return -1;
}
if (ret != sizeof(uint64_t)) {
printErr("Read returned %d, expected %d", ret, sizeof(uint64_t));
return -1;
}
if(ioctl(fd, PERF_EVENT_IOC_DISABLE, 0) == -1) {
perror("ioctl");
return -1;
}
if (clock_gettime(clock, &end) == -1) {
perror("clock_gettime");
return -1;
}
uint64_t nsecs = (end.tv_sec*1e9 + end.tv_nsec) - (start.tv_sec*1e9 + start.tv_nsec);
uint64_t usecs = nsecs/1000;
*freq = cycles/((double)usecs);
return 0;
}
// Return a good number of iterations to run the nop_function in
// order to get a precise measurement of the frequency without taking
// too much time.
uint64_t get_num_iters_from_freq(double frequency) {
// Truncate to reduce variability
uint64_t freq_trunc = ((uint64_t) frequency / 100) * 100;
uint64_t osp_per_iter = 4 * 1000;
return freq_trunc * 1e7 * 1/osp_per_iter;
}
// Differences between x86 measure_frequency and this measure_max_frequency:
// - measure_frequency employs all cores simultaneously whereas
// measure_max_frequency only employs 1.
// - measure_frequency runs the computation and checks /proc/cpuinfo whereas
// measure_max_frequency does not rely on /proc/cpuinfo and simply
// counts cpu cycles to measure frequency.
// - measure_frequency uses actual computation while measuring the frequency
// whereas measure_max_frequency uses nop instructions. This makes the former
// x86 dependant whereas the latter is architecture independant.
int64_t measure_max_frequency(uint32_t core) {
if (!bind_to_cpu(core)) {
printErr("Failed binding the process to CPU %d", core);
return UNKNOWN_DATA;
}
// First, get very rough estimation of clock cycle to
// compute a reasonable value for the iterations
double estimation_freq, frequency;
uint64_t iters = 100000;
if (measure_freq_iters(iters, core, &estimation_freq) == -1)
return UNKNOWN_DATA;
if (estimation_freq <= 0.0) {
printErr("First frequency measurement yielded an invalid value: %f", estimation_freq);
return UNKNOWN_DATA;
}
iters = get_num_iters_from_freq(estimation_freq);
printWarn("Running frequency measurement with %ld iterations on core %d...", iters, core);
// Now perform actual measurement
const char* frequency_banner = "cpufetch is measuring the max frequency...";
printf("%s", frequency_banner);
fflush(stdout);
if (measure_freq_iters(iters, core, &frequency) == -1)
return UNKNOWN_DATA;
// Clean screen once measurement is finished
printf("\r%*c\r", (int) strlen(frequency_banner), ' ');
// Discard last digit in the frequency, which should help providing
// more reliable and predictable values.
return (((int) frequency + 5)/10) * 10;
}
#endif // #ifdef __linux__

6
src/common/freq.h Normal file
View File

@@ -0,0 +1,6 @@
#ifndef __COMMON_FREQ__
#define __COMMON_FREQ__
int64_t measure_max_frequency(uint32_t core);
#endif

View File

@@ -1,3 +1,14 @@
#ifdef _WIN32
#define NOMINMAX
#include <windows.h>
#elif defined __linux__
#define _GNU_SOURCE
#include <sched.h>
#elif defined __FreeBSD__
#include <sys/param.h>
#include <sys/cpuset.h>
#endif
#include <stdarg.h>
#include <stdio.h>
#include <stdlib.h>
@@ -21,7 +32,7 @@
#endif
#ifdef ARCH_X86
static const char* ARCH_STR = "x86_64 build";
static const char* ARCH_STR = "x86 / x86_64 build";
#include "../x86/cpuid.h"
#elif ARCH_PPC
static const char* ARCH_STR = "PowerPC build";
@@ -51,7 +62,7 @@
#endif
#ifndef GIT_FULL_VERSION
static const char* VERSION = "1.03";
static const char* VERSION = "1.06";
#endif
enum {
@@ -61,6 +72,14 @@ enum {
int LOG_LEVEL;
void printBugMessage(FILE *restrict stream) {
#if defined(ARCH_X86) || defined(ARCH_PPC)
fprintf(stream, "Please, create a new issue with this error message, the output of 'cpufetch' and 'cpufetch --debug' on https://github.com/Dr-Noob/cpufetch/issues\n");
#elif ARCH_ARM
fprintf(stream, "Please, create a new issue with this error message, your smartphone/computer model, the output of 'cpufetch --verbose' and 'cpufetch --debug' on https://github.com/Dr-Noob/cpufetch/issues\n");
#endif
}
void printWarn(const char *fmt, ...) {
if(LOG_LEVEL == LOG_LEVEL_VERBOSE) {
int buffer_size = 4096;
@@ -95,10 +114,40 @@ void printBug(const char *fmt, ...) {
fprintf(stderr,RED "[ERROR]: "RESET "%s\n",buffer);
fprintf(stderr,"[VERSION]: ");
print_version(stderr);
#if defined(ARCH_X86) || defined(ARCH_PPC)
fprintf(stderr, "Please, create a new issue with this error message, the output of 'cpufetch' and 'cpufetch --debug' on https://github.com/Dr-Noob/cpufetch/issues\n");
#elif ARCH_ARM
fprintf(stderr, "Please, create a new issue with this error message, your smartphone/computer model, the output of 'cpufetch --verbose' and 'cpufetch --debug' on https://github.com/Dr-Noob/cpufetch/issues\n");
printBugMessage(stderr);
}
bool isReleaseVersion(char *git_full_version) {
return strstr(git_full_version, "-") == NULL;
}
/// The unknown uarch errors are by far the most common error a user will encounter.
/// Rather than using the generic printBug function, which asks the user to report
/// the problem on the issues webpage, this function will check if the program is
/// the release version. In such case, support for this feature is most likely already
/// in the last version, so just tell the user to compile that one and not report this
/// in github.
void printBugCheckRelease(const char *fmt, ...) {
int buffer_size = 4096;
char buffer[buffer_size];
va_list args;
va_start(args, fmt);
vsnprintf(buffer,buffer_size, fmt, args);
va_end(args);
fprintf(stderr, RED "[ERROR]: "RESET "%s\n", buffer);
fprintf(stderr, "[VERSION]: ");
print_version(stderr);
#ifdef GIT_FULL_VERSION
if (isReleaseVersion(GIT_FULL_VERSION)) {
fprintf(stderr, RED "[ERROR]: "RESET "You are using an outdated version of cpufetch. Please compile cpufetch from source (see https://github.com/Dr-Noob/cpufetch?tab=readme-ov-file#22-building-from-source)");
}
else {
printBugMessage(stderr);
}
#else
printBugMessage(stderr);
#endif
}
@@ -161,6 +210,34 @@ void* erealloc(void *ptr, size_t size) {
return newptr;
}
#ifndef __APPLE__
bool bind_to_cpu(int cpu_id) {
#ifdef _WIN32
HANDLE process = GetCurrentProcess();
DWORD_PTR processAffinityMask = 1 << cpu_id;
return SetProcessAffinityMask(process, processAffinityMask);
#elif defined __linux__
cpu_set_t currentCPU;
CPU_ZERO(&currentCPU);
CPU_SET(cpu_id, &currentCPU);
if (sched_setaffinity (0, sizeof(currentCPU), &currentCPU) == -1) {
printWarn("sched_setaffinity: %s", strerror(errno));
return false;
}
return true;
#elif defined __FreeBSD__
cpuset_t currentCPU;
CPU_ZERO(&currentCPU);
CPU_SET(cpu_id, &currentCPU);
if(cpuset_setaffinity(CPU_LEVEL_WHICH, CPU_WHICH_TID, -1, sizeof(cpuset_t), &currentCPU) == -1) {
printWarn("cpuset_setaffinity: %s", strerror(errno));
return false;
}
return true;
#endif
}
#endif
void print_version(FILE *restrict stream) {
#ifdef GIT_FULL_VERSION
fprintf(stream, "cpufetch %s (%s %s)\n", GIT_FULL_VERSION, OS_STR, ARCH_STR);

View File

@@ -12,12 +12,16 @@ void set_log_level(bool verbose);
void printWarn(const char *fmt, ...);
void printErr(const char *fmt, ...);
void printBug(const char *fmt, ...);
void printBugCheckRelease(const char *fmt, ...);
int min(int a, int b);
int max(int a, int b);
char *strremove(char *str, const char *sub);
void* emalloc(size_t size);
void* ecalloc(size_t nmemb, size_t size);
void* erealloc(void *ptr, size_t size);
#ifndef __APPLE__
bool bind_to_cpu(int cpu_id);
#endif
void print_version(FILE *restrict stream);
#endif

View File

@@ -30,21 +30,29 @@ void print_help(char *argv[]) {
#ifdef ARCH_X86
#ifdef __linux__
printf(" --%s %*s Compute the peak performance accurately (measure the CPU frequency instead of using the maximum)\n", t[ARG_ACCURATE_PP], (int) (max_len-strlen(t[ARG_ACCURATE_PP])), "");
#endif
printf(" --%s %*s Measure the max CPU frequency instead of reading it\n", t[ARG_MEASURE_MAX_FREQ], (int) (max_len-strlen(t[ARG_MEASURE_MAX_FREQ])), "");
#endif // __linux__
printf(" --%s %*s Show the old Intel logo\n", t[ARG_LOGO_INTEL_OLD], (int) (max_len-strlen(t[ARG_LOGO_INTEL_OLD])), "");
printf(" --%s %*s Show the new Intel logo\n", t[ARG_LOGO_INTEL_NEW], (int) (max_len-strlen(t[ARG_LOGO_INTEL_NEW])), "");
printf(" -%c, --%s %*s Show the full CPU name (do not abbreviate it)\n", c[ARG_FULLCPUNAME], t[ARG_FULLCPUNAME], (int) (max_len-strlen(t[ARG_FULLCPUNAME])), "");
printf(" -%c, --%s %*s Print raw cpuid data (debug purposes)\n", c[ARG_RAW], t[ARG_RAW], (int) (max_len-strlen(t[ARG_RAW])), "");
#endif // ARCH_X86
#ifdef ARCH_ARM
#ifdef __linux__
printf(" --%s %*s Measure the max CPU frequency instead of reading it\n", t[ARG_MEASURE_MAX_FREQ], (int) (max_len-strlen(t[ARG_MEASURE_MAX_FREQ])), "");
#endif
#endif
printf(" -%c, --%s %*s Print this help and exit\n", c[ARG_HELP], t[ARG_HELP], (int) (max_len-strlen(t[ARG_HELP])), "");
printf(" -%c, --%s %*s Print cpufetch version and exit\n", c[ARG_VERSION], t[ARG_VERSION], (int) (max_len-strlen(t[ARG_VERSION])), "");
printf("\nCOLORS: \n");
printf(" * \"intel\": Use Intel default color scheme \n");
printf(" * \"amd\": Use AMD default color scheme \n");
printf(" * \"ibm\", Use IBM default color scheme \n");
printf(" * \"arm\": Use ARM default color scheme \n");
printf(" * \"sifive\": Use SiFive default color scheme \n");
printf(" * \"intel\": Use Intel color scheme \n");
printf(" * \"intel-new\": Use Intel (new logo) color scheme \n");
printf(" * \"amd\": Use AMD color scheme \n");
printf(" * \"ibm\", Use IBM color scheme \n");
printf(" * \"arm\": Use ARM color scheme \n");
printf(" * \"rockchip\": Use Rockchip color scheme \n");
printf(" * \"sifive\": Use SiFive color scheme \n");
printf(" * custom: If the argument of --color does not match any of the previous strings, a custom scheme can be specified.\n");
printf(" 5 colors must be given in RGB with the format: R,G,B:R,G,B:...\n");
printf(" The first 3 colors are the CPU art color and the next 2 colors are the text colors\n");
@@ -74,8 +82,15 @@ void print_help(char *argv[]) {
printf("\nNOTE: \n");
printf(" Peak performance information is NOT accurate. cpufetch computes peak performance using the max\n");
printf(" frequency of the CPU. However, to compute the peak performance, you need to know the frequency of the\n");
printf(" CPU running AVX code. This value is not be fetched by cpufetch since it depends on each specific CPU.\n");
printf(" To correctly measure peak performance, see: https://github.com/Dr-Noob/peakperf\n");
printf(" CPU running AVX code. By default, this value is not fetched by cpufetch, but you can use the\n");
printf(" --accurate-pp option, which will measure the AVX frequency and show a more precise estimation\n");
printf(" (this option is only available in x86 architectures).\n");
printf(" To precisely measure peak performance, see: https://github.com/Dr-Noob/peakperf\n");
printf("\n");
printf(" Both --accurate-pp and --measure-max-freq measure the actual frequency of the CPU. However,\n");
printf(" they differ slightly. The former measures the max frequency while running vectorized SSE/AVX\n");
printf(" instructions and it is thus x86 only, whereas the latter simply measures the max clock cycle\n");
printf(" and is architecture independent.\n");
}
int main(int argc, char* argv[]) {
@@ -116,8 +131,10 @@ int main(int argc, char* argv[]) {
#endif
}
if(print_cpufetch(cpu, get_style(), get_colors(), show_full_cpu_name()))
if(print_cpufetch(cpu, get_style(), get_colors(), show_full_cpu_name())) {
return EXIT_SUCCESS;
else
}
else {
return EXIT_FAILURE;
}
}

133
src/common/pci.c Normal file
View File

@@ -0,0 +1,133 @@
#define _GNU_SOURCE
#include <sys/stat.h>
#include <dirent.h>
#include "udev.h"
#include "global.h"
#include "pci.h"
#ifndef PATH_MAX
#define PATH_MAX 1024
#endif
#define PCI_PATH "/sys/bus/pci/devices/"
#define MAX_LENGTH_PCI_DIR_NAME 1024
// Return a list of PCI devices containing only
// the sysfs path
struct pci_devices * get_pci_paths(void) {
DIR *dirp;
if ((dirp = opendir(PCI_PATH)) == NULL) {
perror("opendir");
return NULL;
}
struct dirent *dp;
int numDirs = 0;
errno = 0;
while ((dp = readdir(dirp)) != NULL) {
if (strcmp(dp->d_name, ".") != 0 && strcmp(dp->d_name, "..") != 0)
numDirs++;
}
if (errno != 0) {
perror("readdir");
return NULL;
}
rewinddir(dirp);
struct pci_devices * pci = emalloc(sizeof(struct pci_devices));
pci->num_devices = numDirs;
pci->devices = emalloc(sizeof(struct pci_device) * pci->num_devices);
char * full_path = emalloc(PATH_MAX * sizeof(char));
struct stat stbuf;
int i = 0;
while ((dp = readdir(dirp)) != NULL) {
if (strcmp(dp->d_name, ".") == 0 || strcmp(dp->d_name, "..") == 0)
continue;
if (strlen(dp->d_name) > MAX_LENGTH_PCI_DIR_NAME) {
printErr("Directory name is too long: %s", dp->d_name);
return NULL;
}
memset(full_path, 0, PATH_MAX * sizeof(char));
snprintf(full_path, min(strlen(PCI_PATH) + strlen(dp->d_name) + 1, PATH_MAX), "%s%s", PCI_PATH, dp->d_name);
if (stat(full_path, &stbuf) == -1) {
perror("stat");
return NULL;
}
if ((stbuf.st_mode & S_IFMT) == S_IFDIR) {
int strLen = min(MAX_LENGTH_PCI_DIR_NAME, strlen(dp->d_name)) + 1;
pci->devices[i] = emalloc(sizeof(struct pci_device));
pci->devices[i]->path = ecalloc(strLen, sizeof(char));
strncpy(pci->devices[i]->path, dp->d_name, strLen);
i++;
}
}
if (errno != 0) {
perror("readdir");
return NULL;
}
return pci;
}
// For each PCI device in the list pci, fetch its vendor and
// device id using sysfs (e.g., /sys/bus/pci/devices/XXX/{vendor/device})
void populate_pci_devices(struct pci_devices * pci) {
int filelen;
char* buf;
for (int i=0; i < pci->num_devices; i++) {
struct pci_device* dev = pci->devices[i];
int path_size = strlen(PCI_PATH) + strlen(dev->path) + 2;
// Read vendor_id
char *vendor_id_path = emalloc(sizeof(char) * (path_size + strlen("vendor") + 1));
sprintf(vendor_id_path, "%s/%s/%s", PCI_PATH, dev->path, "vendor");
if ((buf = read_file(vendor_id_path, &filelen)) == NULL) {
printWarn("read_file: %s: %s\n", vendor_id_path, strerror(errno));
dev->vendor_id = 0;
}
else {
dev->vendor_id = strtol(buf, NULL, 16);
}
// Read device_id
char *device_id_path = emalloc(sizeof(char) * (path_size + strlen("device") + 1));
sprintf(device_id_path, "%s/%s/%s", PCI_PATH, dev->path, "device");
if ((buf = read_file(device_id_path, &filelen)) == NULL) {
printWarn("read_file: %s: %s\n", device_id_path, strerror(errno));
dev->device_id = 0;
}
else {
dev->device_id = strtol(buf, NULL, 16);
}
free(vendor_id_path);
free(device_id_path);
}
}
// Return a list of PCI devices that could be used to infer the SoC.
// The criteria to determine which devices are suitable for this task
// is decided in filter_pci_devices.
struct pci_devices * get_pci_devices(void) {
struct pci_devices * pci = get_pci_paths();
if (pci == NULL)
return NULL;
populate_pci_devices(pci);
return pci;
}

23
src/common/pci.h Normal file
View File

@@ -0,0 +1,23 @@
#ifndef __PCI__
#define __PCI__
#define PCI_VENDOR_NVIDIA 0x10de
#define PCI_VENDOR_AMPERE 0x1def
#define PCI_DEVICE_TEGRA_X1 0x0faf
#define PCI_DEVICE_ALTRA 0xe100
struct pci_device {
char * path;
uint16_t vendor_id;
uint16_t device_id;
};
struct pci_devices {
struct pci_device ** devices;
int num_devices;
};
struct pci_devices * get_pci_devices(void);
#endif

View File

@@ -20,6 +20,7 @@
#include "../arm/uarch.h"
#include "../arm/midr.h"
#include "../arm/soc.h"
#include "../common/soc.h"
#elif ARCH_RISCV
#include "../riscv/riscv.h"
#include "../riscv/uarch.h"
@@ -60,6 +61,7 @@ enum {
ATTRIBUTE_NCORES,
ATTRIBUTE_NCORES_DUAL,
#ifdef ARCH_X86
ATTRIBUTE_SSE,
ATTRIBUTE_AVX,
ATTRIBUTE_FMA,
#elif ARCH_PPC
@@ -95,6 +97,7 @@ static const char* ATTRIBUTE_FIELDS [] = {
"Cores:",
"Cores (Total):",
#ifdef ARCH_X86
"SSE:",
"AVX:",
"FMA:",
#elif ARCH_PPC
@@ -130,6 +133,7 @@ static const char* ATTRIBUTE_FIELDS_SHORT [] = {
"Cores:",
"Cores (Total):",
#ifdef ARCH_X86
"SSE:",
"AVX:",
"FMA:",
#elif ARCH_PPC
@@ -335,6 +339,13 @@ struct ascii_logo* choose_ascii_art_aux(struct ascii_logo* logo_long, struct asc
}
}
// https://no-color.org/
bool is_color_enabled(void) {
const char *var_name = "NO_COLOR";
char *no_color = getenv(var_name);
return no_color == NULL || no_color[0] == '\0';
}
void choose_ascii_art(struct ascii* art, struct color** cs, struct terminal* term, int lf) {
// 1. Choose logo
#ifdef ARCH_X86
@@ -349,6 +360,9 @@ void choose_ascii_art(struct ascii* art, struct color** cs, struct terminal* ter
else if(art->vendor == CPU_VENDOR_AMD) {
art->art = choose_ascii_art_aux(&logo_amd_l, &logo_amd, term, lf);
}
else if(art->vendor == CPU_VENDOR_HYGON) {
art->art = &logo_hygon;
}
else {
art->art = &logo_unknown;
}
@@ -363,14 +377,28 @@ void choose_ascii_art(struct ascii* art, struct color** cs, struct terminal* ter
art->art = &logo_exynos;
else if(art->vendor == SOC_VENDOR_KIRIN)
art->art = &logo_kirin;
else if(art->vendor == SOC_VENDOR_KUNPENG)
art->art = &logo_kunpeng;
else if(art->vendor == SOC_VENDOR_BROADCOM)
art->art = &logo_broadcom;
else if(art->vendor == SOC_VENDOR_APPLE)
art->art = &logo_apple;
else if(art->vendor == SOC_VENDOR_GOOGLE)
art->art = &logo_google;
else if(art->vendor == SOC_VENDOR_ALLWINNER)
art->art = &logo_allwinner;
else if(art->vendor == SOC_VENDOR_ROCKCHIP)
art->art = &logo_rockchip;
else if(art->vendor == SOC_VENDOR_AMPERE)
art->art = &logo_ampere;
else if(art->vendor == SOC_VENDOR_NXP)
art->art = &logo_nxp;
else if(art->vendor == SOC_VENDOR_AMLOGIC)
art->art = &logo_amlogic;
else if(art->vendor == SOC_VENDOR_MARVELL)
art->art = &logo_marvell;
else if(art->vendor == SOC_VENDOR_NVIDIA)
art->art = choose_ascii_art_aux(&logo_nvidia_l, &logo_nvidia, term, lf);
else {
art->art = choose_ascii_art_aux(&logo_arm_l, &logo_arm, term, lf);
}
@@ -381,12 +409,17 @@ void choose_ascii_art(struct ascii* art, struct color** cs, struct terminal* ter
art->art = choose_ascii_art_aux(&logo_starfive_l, &logo_starfive, term, lf);
else if(art->vendor == SOC_VENDOR_ALLWINNER)
art->art = &logo_allwinner;
else if(art->vendor == SOC_VENDOR_SIPEED)
art->art = &logo_sipeed;
else
art->art = &logo_riscv;
#endif
// 2. Choose colors
struct ascii_logo* logo = art->art;
bool color = is_color_enabled();
if (!color)
art->style = STYLE_LEGACY;
switch(art->style) {
case STYLE_LEGACY:
@@ -572,6 +605,7 @@ bool print_cpufetch_x86(struct cpuInfo* cpu, STYLE s, struct color** cs, struct
for(int i = 0; i < cpu->num_cpus; ptr = ptr->next_cpu, i++) {
char* max_frequency = get_str_freq(ptr->freq);
char* avx = get_str_avx(ptr);
char* sse = get_str_sse(ptr);
char* fma = get_str_fma(ptr);
char* cpu_num = emalloc(sizeof(char) * 9);
@@ -606,8 +640,18 @@ bool print_cpufetch_x86(struct cpuInfo* cpu, STYLE s, struct color** cs, struct
setAttribute(art, ATTRIBUTE_NCORES, n_cores);
}
}
// Show the most modern vector instructions.
if (strcmp(avx, "No") == 0) {
if (strcmp(sse, "No") != 0) {
setAttribute(art, ATTRIBUTE_SSE, sse);
}
}
else {
setAttribute(art, ATTRIBUTE_AVX, avx);
setAttribute(art, ATTRIBUTE_FMA, fma);
}
if(l1i != NULL) setAttribute(art, ATTRIBUTE_L1i, l1i);
if(l1d != NULL) setAttribute(art, ATTRIBUTE_L1d, l1d);
if(l2 != NULL) setAttribute(art, ATTRIBUTE_L2, l2);
@@ -680,6 +724,9 @@ bool print_cpufetch_ppc(struct cpuInfo* cpu, STYLE s, struct color** cs, struct
setAttribute(art, ATTRIBUTE_NAME, cpu_name);
}
setAttribute(art, ATTRIBUTE_UARCH, uarch);
if(cpu->hv->present) {
setAttribute(art, ATTRIBUTE_HYPERVISOR, cpu->hv->hv_name);
}
setAttribute(art, ATTRIBUTE_TECHNOLOGY, manufacturing_process);
setAttribute(art, ATTRIBUTE_FREQUENCY, max_frequency);
uint32_t socket_num = get_nsockets(cpu->topo);
@@ -840,21 +887,18 @@ bool print_cpufetch_arm(struct cpuInfo* cpu, STYLE s, struct color** cs, struct
char* soc_name = get_soc_name(cpu->soc);
char* features = get_str_features(cpu);
setAttribute(art, ATTRIBUTE_SOC, soc_name);
// Currently no reliable way to identify the specific SoC on Windows
// https://github.com/Dr-Noob/cpufetch/pull/273
// Hide manufacturing process
#if !defined(_WIN32)
setAttribute(art, ATTRIBUTE_TECHNOLOGY, manufacturing_process);
#endif
if(cpu->num_cpus == 1) {
char* uarch = get_str_uarch(cpu);
char* max_frequency = get_str_freq(cpu->freq);
char* n_cores = get_str_topology(cpu, cpu->topo, false);
/*
* char* l1i = get_str_l1i(cpu->cach);
* char* l1d = get_str_l1d(cpu->cach);
* char* l2 = get_str_l2(cpu->cach);
* char* l3 = get_str_l3(cpu->cach);
* Do not setAttribute for caches.
* Cache functionality may be implemented
* in the future
*/
setAttribute(art, ATTRIBUTE_UARCH, uarch);
setAttribute(art, ATTRIBUTE_FREQUENCY, max_frequency);
@@ -869,17 +913,8 @@ bool print_cpufetch_arm(struct cpuInfo* cpu, STYLE s, struct color** cs, struct
char* uarch = get_str_uarch(ptr);
char* max_frequency = get_str_freq(ptr->freq);
char* n_cores = get_str_topology(ptr, ptr->topo, false);
/*
* char* l1i = get_str_l1i(cpu->cach);
* char* l1d = get_str_l1d(cpu->cach);
* char* l2 = get_str_l2(cpu->cach);
* char* l3 = get_str_l3(cpu->cach);
* Do not setAttribute for caches.
* Cache functionality may be implemented
* in the future
*/
char* cpu_num = emalloc(sizeof(char) * 9);
sprintf(cpu_num, "CPU %d:", i+1);
setAttribute(art, ATTRIBUTE_CPU_NUM, cpu_num);
setAttribute(art, ATTRIBUTE_UARCH, uarch);
@@ -931,16 +966,14 @@ bool print_cpufetch_arm(struct cpuInfo* cpu, STYLE s, struct color** cs, struct
#endif
#ifdef ARCH_RISCV
// https://stackoverflow.com/questions/109023/count-the-number-of-set-bits-in-a-32-bit-integer
int number_of_bits(uint32_t i) {
i = i - ((i >> 1) & 0x55555555); // add pairs of bits
i = (i & 0x33333333) + ((i >> 2) & 0x33333333); // quads
i = (i + (i >> 4)) & 0x0F0F0F0F; // groups of 8
return (i * 0x01010101) >> 24; // horizontal sum of bytes
// https://stackoverflow.com/a/2709523
uint64_t number_of_bits(uint64_t i) {
i = i - ((i >> 1) & 0x5555555555555555);
i = (i & 0x3333333333333333) + ((i >> 2) & 0x3333333333333333);
return (((i + (i >> 4)) & 0xF0F0F0F0F0F0F0F) * 0x101010101010101) >> 56;
}
void print_ascii_riscv(struct ascii* art, uint32_t la, int32_t termw, const char** attribute_fields, uint32_t extensions_mask) {
void print_ascii_riscv(struct ascii* art, uint32_t la, int32_t termw, const char** attribute_fields, uint64_t extensions_mask) {
struct ascii_logo* logo = art->art;
int attr_to_print = 0;
int attr_type;
@@ -954,7 +987,7 @@ void print_ascii_riscv(struct ascii* art, uint32_t la, int32_t termw, const char
int32_t space_up = ((int)logo->height - (int)(art->n_attributes_set + num_extensions))/2;
int32_t space_down = (int)logo->height - (int)(art->n_attributes_set + num_extensions) - (int)space_up;
uint32_t logo_pos = 0;
int32_t iters = max(logo->height, art->n_attributes_set);
int32_t iters = max(logo->height, art->n_attributes_set + num_extensions);
struct line_buffer* lbuf = emalloc(sizeof(struct line_buffer));
lbuf->buf = emalloc(sizeof(char) * LINE_BUFFER_SIZE);
@@ -1035,11 +1068,6 @@ bool print_cpufetch_riscv(struct cpuInfo* cpu, STYLE s, struct color** cs, struc
char* extensions = get_str_extensions(cpu);
char* max_frequency = get_str_freq(cpu->freq);
char* n_cores = get_str_topology(cpu, cpu->topo);
/*char* l1i = get_str_l1i(cpu->cach);
char* l1d = get_str_l1d(cpu->cach);
char* l2 = get_str_l2(cpu->cach);
char* l3 = get_str_l3(cpu->cach);*/
char* pp = get_str_peak_performance(cpu->peak_performance);
// Step 2. Set attributes
@@ -1051,12 +1079,6 @@ bool print_cpufetch_riscv(struct cpuInfo* cpu, STYLE s, struct color** cs, struc
if(extensions != NULL) {
setAttribute(art, ATTRIBUTE_EXTENSIONS, extensions);
}
/*setAttribute(art,ATTRIBUTE_L1i,l1i);
setAttribute(art,ATTRIBUTE_L1d,l1d);
setAttribute(art,ATTRIBUTE_L2,l2);
if(l3 != NULL) {
setAttribute(art,ATTRIBUTE_L3,l3);
}*/
setAttribute(art, ATTRIBUTE_PEAK, pp);
// Step 3. Print output

116
src/common/soc.c Normal file
View File

@@ -0,0 +1,116 @@
#include "soc.h"
#ifdef ARCH_ARM
#include "../arm/socs.h"
#elif ARCH_RISCV
#include "../riscv/socs.h"
#endif
#include "udev.h"
#include "../common/global.h"
#include <string.h>
static char* soc_trademark_string[] = {
// ARM
[SOC_VENDOR_SNAPDRAGON] = "Snapdragon ",
[SOC_VENDOR_MEDIATEK] = "MediaTek ",
[SOC_VENDOR_EXYNOS] = "Exynos ",
[SOC_VENDOR_KIRIN] = "Kirin ",
[SOC_VENDOR_KUNPENG] = "Kunpeng ",
[SOC_VENDOR_BROADCOM] = "Broadcom ",
[SOC_VENDOR_APPLE] = "Apple ",
[SOC_VENDOR_ROCKCHIP] = "Rockchip ",
[SOC_VENDOR_GOOGLE] = "Google ",
[SOC_VENDOR_NVIDIA] = "NVIDIA ",
[SOC_VENDOR_AMPERE] = "Ampere ",
[SOC_VENDOR_NXP] = "NXP ",
[SOC_VENDOR_AMLOGIC] = "Amlogic ",
[SOC_VENDOR_MARVELL] = "Marvell",
// RISC-V
[SOC_VENDOR_SIFIVE] = "SiFive ",
[SOC_VENDOR_STARFIVE] = "StarFive ",
[SOC_VENDOR_SIPEED] = "Sipeed ",
// ARM & RISC-V
[SOC_VENDOR_ALLWINNER] = "Allwinner "
};
VENDOR get_soc_vendor(struct system_on_chip* soc) {
return soc->vendor;
}
char* get_str_process(struct system_on_chip* soc) {
char* str;
if(soc->process == UNKNOWN) {
str = emalloc(sizeof(char) * (strlen(STRING_UNKNOWN)+1));
snprintf(str, strlen(STRING_UNKNOWN)+1, STRING_UNKNOWN);
}
else {
int max_process_len = 5 + 1;
str = ecalloc(max_process_len, sizeof(char));
snprintf(str, max_process_len, "%dnm", soc->process);
}
return str;
}
char* get_soc_name(struct system_on_chip* soc) {
if(soc->model == SOC_MODEL_UNKNOWN)
return soc->raw_name;
return soc->name;
}
void fill_soc(struct system_on_chip* soc, char* soc_name, SOC soc_model, int32_t process) {
soc->model = soc_model;
soc->vendor = get_soc_vendor_from_soc(soc_model);
soc->process = process;
if(soc->vendor == SOC_VENDOR_UNKNOWN) {
printBug("fill_soc: soc->vendor == SOC_VENDOR_UNKOWN");
// If we fall here there is a bug in socs.h
// Reset everything to avoid segfault
soc->vendor = SOC_VENDOR_UNKNOWN;
soc->model = SOC_MODEL_UNKNOWN;
soc->process = UNKNOWN;
soc->raw_name = emalloc(sizeof(char) * (strlen(STRING_UNKNOWN)+1));
snprintf(soc->raw_name, strlen(STRING_UNKNOWN)+1, STRING_UNKNOWN);
}
else {
int len = strlen(soc_name) + strlen(soc_trademark_string[soc->vendor]) + 1;
soc->name = emalloc(sizeof(char) * len);
sprintf(soc->name, "%s%s", soc_trademark_string[soc->vendor], soc_name);
}
}
void fill_soc_raw(struct system_on_chip* soc, char* soc_name, VENDOR vendor) {
soc->model = SOC_MODEL_UNKNOWN;
soc->vendor = vendor;
soc->process = UNKNOWN;
int len = strlen(soc_name) + strlen(soc_trademark_string[soc->vendor]) + 1;
soc->name = emalloc(sizeof(char) * len);
sprintf(soc->name, "%s%s", soc_trademark_string[soc->vendor], soc_name);
}
#ifdef _WIN32
VENDOR try_match_soc_vendor_name(char* vendor_name)
{
for(size_t i=1; i < sizeof(soc_trademark_string)/sizeof(soc_trademark_string[0]); i++) {
if(strstr(vendor_name, soc_trademark_string[i]) != NULL) {
return i;
}
}
return SOC_VENDOR_UNKNOWN;
}
#endif
bool match_soc(struct system_on_chip* soc, char* raw_name, char* expected_name, char* soc_name, SOC soc_model, int32_t process) {
int len1 = strlen(raw_name);
int len2 = strlen(expected_name);
int len = min(len1, len2);
if(strncmp(raw_name, expected_name, len) != 0) {
return false;
}
else {
fill_soc(soc, soc_name, soc_model, process);
return true;
}
}

64
src/common/soc.h Normal file
View File

@@ -0,0 +1,64 @@
#ifndef __SOC__
#define __SOC__
// NOTE:
// soc.c/soc.h are used by
// ARM and RISC-V backends
#include "../common/cpu.h"
#include <stdint.h>
#define UNKNOWN -1
typedef int32_t SOC;
enum {
SOC_VENDOR_UNKNOWN,
// ARM
SOC_VENDOR_SNAPDRAGON,
SOC_VENDOR_MEDIATEK,
SOC_VENDOR_EXYNOS,
SOC_VENDOR_KIRIN,
SOC_VENDOR_KUNPENG,
SOC_VENDOR_BROADCOM,
SOC_VENDOR_APPLE,
SOC_VENDOR_ROCKCHIP,
SOC_VENDOR_GOOGLE,
SOC_VENDOR_NVIDIA,
SOC_VENDOR_AMPERE,
SOC_VENDOR_NXP,
SOC_VENDOR_AMLOGIC,
SOC_VENDOR_MARVELL,
// RISC-V
SOC_VENDOR_SIFIVE,
SOC_VENDOR_STARFIVE,
SOC_VENDOR_SIPEED,
// ARM & RISC-V
SOC_VENDOR_ALLWINNER
};
struct system_on_chip {
SOC model;
VENDOR vendor;
int32_t process;
char* name;
char* raw_name;
};
struct system_on_chip* get_soc(struct cpuInfo* cpu);
char* get_soc_name(struct system_on_chip* soc);
VENDOR get_soc_vendor(struct system_on_chip* soc);
bool match_soc(struct system_on_chip* soc, char* raw_name, char* expected_name, char* soc_name, SOC soc_model, int32_t process);
char* get_str_process(struct system_on_chip* soc);
void fill_soc(struct system_on_chip* soc, char* soc_name, SOC soc_model, int32_t process);
void fill_soc_raw(struct system_on_chip* soc, char* soc_name, VENDOR vendor);
#ifdef _WIN32
VENDOR try_match_soc_vendor_name(char* vendor_name);
#endif
#define SOC_START if (false) {}
#define SOC_EQ(raw_name, expected_name, soc_name, soc_model, soc, process) \
else if (match_soc(soc, raw_name, expected_name, soc_name, soc_model, process)) return true;
#define SOC_END else { return false; }
#endif

View File

@@ -4,8 +4,8 @@
#include <string.h>
#include <errno.h>
#include "../common/global.h"
#include "../common/cpu.h"
#include "global.h"
#include "cpu.h"
uint32_t get_sys_info_by_name(char* name) {
size_t size = 0;

View File

@@ -8,6 +8,9 @@
// APPLE_CPU_PART_M2_BLIZZARD=0x30,M2_AVALANCHE=0x31
#define MIDR_APPLE_M2_BLIZZARD 0x610F0300
#define MIDR_APPLE_M2_AVALANCHE 0x610F0310
// https://github.com/AsahiLinux/m1n1/blob/main/src/chickens.c
#define MIDR_APPLE_M3_SAWTOOTH 0x610F0480
#define MIDR_APPLE_M3_EVEREST 0x610F0490
// M1 / A14
#ifndef CPUFAMILY_ARM_FIRESTORM_ICESTORM
@@ -17,6 +20,15 @@
#ifndef CPUFAMILY_ARM_AVALANCHE_BLIZZARD
#define CPUFAMILY_ARM_AVALANCHE_BLIZZARD 0xDA33D83D
#endif
// M3 / A16 / A17
// M3: https://ratfactor.com/zig/stdlib-browseable2/c/darwin.zig.html
// M3_2: https://github.com/Dr-Noob/cpufetch/issues/230
// PRO: https://github.com/Dr-Noob/cpufetch/issues/225
// MAX: https://github.com/Dr-Noob/cpufetch/issues/210
#define CPUFAMILY_ARM_EVEREST_SAWTOOTH 0x8765EDEA
#define CPUFAMILY_ARM_EVEREST_SAWTOOTH_2 0xFA33415E
#define CPUFAMILY_ARM_EVEREST_SAWTOOTH_PRO 0x5F4DEA93
#define CPUFAMILY_ARM_EVEREST_SAWTOOTH_MAX 0x72015832
// For detecting different M1 types
// NOTE: Could also be achieved detecting different
@@ -31,6 +43,14 @@
#define CPUSUBFAMILY_ARM_HC_HD 5
#endif
// For alternative way to get CPU frequency on macOS and *BSD
#ifdef __APPLE__
#define CPUFREQUENCY_SYSCTL "hw.cpufrequency_max"
#else
// For FreeBSD, not sure about other *BSD
#define CPUFREQUENCY_SYSCTL "dev.cpu.0.freq"
#endif
uint32_t get_sys_info_by_name(char* name);
#endif

View File

@@ -1,7 +1,10 @@
#include "../common/global.h"
#include "udev.h"
#include "global.h"
#include "cpu.h"
#define _PATH_DEVTREE "/proc/device-tree/compatible"
// https://www.kernel.org/doc/html/latest/core-api/cpu_hotplug.html
int get_ncores_from_cpuinfo(void) {
// Examples:
@@ -129,6 +132,26 @@ long get_cache_size_from_file(char* path) {
return ret * 1024;
}
char* get_field_from_cpuinfo(char* CPUINFO_FIELD) {
int filelen;
char* buf;
if((buf = read_file(_PATH_CPUINFO, &filelen)) == NULL) {
printWarn("read_file: %s: %s:\n", _PATH_CPUINFO, strerror(errno));
return NULL;
}
char* tmp1 = strstr(buf, CPUINFO_FIELD);
if(tmp1 == NULL) return NULL;
tmp1 = tmp1 + strlen(CPUINFO_FIELD);
char* tmp2 = strstr(tmp1, "\n");
int strlen = (1 + (tmp2-tmp1));
char* hardware = ecalloc(strlen, sizeof(char));
strncpy(hardware, tmp1, tmp2-tmp1);
return hardware;
}
long get_max_freq_from_file(uint32_t core) {
char path[_PATH_FREQUENCY_MAX_LEN];
sprintf(path, "%s%s/cpu%d%s%s", _PATH_SYS_SYSTEM, _PATH_SYS_CPU, core, _PATH_FREQUENCY, _PATH_FREQUENCY_MAX);
@@ -178,7 +201,13 @@ bool maps_equal(uint32_t* map1, uint32_t* map2, int n) {
return true;
}
int get_num_caches_from_files(char** paths, int num_paths) {
// Generic function to count the number of distinct
// elements in the list of files passed in char** paths.
// An element can be potentially anything.
// We use this function to count:
// - The number of caches
// - The number of sockets
int get_num_elements_from_files(char** paths, int num_paths) {
int filelen;
char* buf;
char* tmpbuf;
@@ -193,10 +222,10 @@ int get_num_caches_from_files(char** paths, int num_paths) {
num_bitmasks += (buf[i] == ',');
}
// 2. Read cpu_shared_map from every core
uint32_t** shared_maps = emalloc(sizeof(uint32_t *) * num_paths);
// 2. Read map from every core
uint32_t** maps = emalloc(sizeof(uint32_t *) * num_paths);
for(int i=0; i < num_paths; i++) {
shared_maps[i] = emalloc(sizeof(uint32_t) * num_bitmasks);
maps[i] = emalloc(sizeof(uint32_t) * num_bitmasks);
if((buf = read_file(paths[i], &filelen)) == NULL) {
printWarn("Could not open '%s'", paths[i]);
@@ -206,6 +235,7 @@ int get_num_caches_from_files(char** paths, int num_paths) {
for(int j=0; j < num_bitmasks; j++) {
char* end;
tmpbuf = emalloc(sizeof(char) * (strlen(buf) + 1));
memset(tmpbuf, 0, sizeof(char) * (strlen(buf) + 1));
char* commaend = strstr(buf, ",");
if(commaend == NULL) {
strcpy(tmpbuf, buf);
@@ -221,35 +251,43 @@ int get_num_caches_from_files(char** paths, int num_paths) {
return -1;
}
shared_maps[i][j] = (uint32_t) ret;
maps[i][j] = (uint32_t) ret;
buf = commaend + 1;
free(tmpbuf);
}
}
// 2. Count number of different masks; this is the number of caches
int num_caches = 0;
// 2. Count number of different masks; this is the number of elements
int num_elements = 0;
bool found = false;
uint32_t** unique_shared_maps = emalloc(sizeof(uint32_t *) * num_paths);
uint32_t** unique_maps = emalloc(sizeof(uint32_t *) * num_paths);
for(int i=0; i < num_paths; i++) {
unique_shared_maps[i] = emalloc(sizeof(uint32_t) * num_bitmasks);
unique_maps[i] = emalloc(sizeof(uint32_t) * num_bitmasks);
for(int j=0; j < num_bitmasks; j++) {
unique_shared_maps[i][j] = 0;
unique_maps[i][j] = 0;
}
}
for(int i=0; i < num_paths; i++) {
for(int j=0; j < num_paths && !found; j++) {
if(maps_equal(shared_maps[i], unique_shared_maps[j], num_bitmasks)) found = true;
if(maps_equal(maps[i], unique_maps[j], num_bitmasks)) found = true;
}
if(!found) {
add_shared_map(shared_maps, i, unique_shared_maps, num_caches, num_bitmasks);
num_caches++;
add_shared_map(maps, i, unique_maps, num_elements, num_bitmasks);
num_elements++;
}
found = false;
}
return num_caches;
return num_elements;
}
int get_num_caches_from_files(char** paths, int num_paths) {
return get_num_elements_from_files(paths, num_paths);
}
int get_num_sockets_from_files(char** paths, int num_paths) {
return get_num_elements_from_files(paths, num_paths);
}
int get_num_caches_by_level(struct cpuInfo* cpu, uint32_t level) {
@@ -278,3 +316,94 @@ int get_num_caches_by_level(struct cpuInfo* cpu, uint32_t level) {
return ret;
}
int get_num_sockets_package_cpus(struct topology* topo) {
// Get number of sockets using
// /sys/devices/system/cpu/cpu*/topology/package_cpus
char** paths = emalloc(sizeof(char *) * topo->total_cores);
for(int i=0; i < topo->total_cores; i++) {
paths[i] = emalloc(sizeof(char) * _PATH_PACKAGE_MAX_LEN);
sprintf(paths[i], "%s%s/cpu%d%s", _PATH_SYS_SYSTEM, _PATH_SYS_CPU, i, _PATH_TOPO_PACKAGE_CPUS);
}
int ret = get_num_sockets_from_files(paths, topo->total_cores);
for(int i=0; i < topo->total_cores; i++)
free(paths[i]);
free(paths);
return ret;
}
// Inspired in is_devtree_compatible from lscpu
bool is_devtree_compatible(char* str) {
int filelen;
char* buf;
if((buf = read_file("/proc/device-tree/compatible", &filelen)) == NULL) {
return false;
}
char* tmp;
if((tmp = strstr(buf, str)) == NULL) {
return false;
}
return true;
}
char* get_devtree_compatible(int *filelen) {
char* buf;
if ((buf = read_file(_PATH_DEVTREE, filelen)) == NULL) {
printWarn("read_file: %s: %s", _PATH_DEVTREE, strerror(errno));
}
return buf;
}
// TODO:
// Returns a list of strings containing the vendors of the compatible
// file from the device tree. In this context, vendor refers to the first
// string of every entry. For instance, given a compatible file with:
// "str1,foo1.str2,foo2" (where . denotes the NULL byte, i.e., the separator),
// then this function will return a list with str1,str2.
struct devtree** get_devtree_compatible_struct(int *num_vendors_ptr) {
int len;
char* dt = get_devtree_compatible(&len);
if (dt == NULL) {
return NULL;
}
int num_vendors = 0;
char* ptr = dt;
for (int ptrpos = 0; ptrpos < len; ptrpos = (ptr-dt)) {
ptr = memchr(ptr, '\0', len)+1;
num_vendors++;
}
struct devtree** vendors = emalloc(sizeof(struct devtree *) * num_vendors);
ptr = dt;
for (int ptrpos = 0, i = 0; ptrpos < len; ptrpos = (ptr-dt), i++) {
char* comma_ptr = strstr(ptr, ",")-1;
char* end_ptr = memchr(comma_ptr, '\0', ptrpos - len);
// TODO check NULL
int vendor_str_len = (comma_ptr-ptr)+1;
int model_str_len = (end_ptr-comma_ptr)+1;
vendors[i] = emalloc(sizeof(struct devtree));
vendors[i]->vendor = ecalloc(vendor_str_len, sizeof(char));
vendors[i]->model = ecalloc(model_str_len, sizeof(char));
strncpy(vendors[i]->vendor, ptr, vendor_str_len);
strncpy(vendors[i]->model, comma_ptr, model_str_len);
ptr = memchr(ptr, '\0', len)+1;
}
*num_vendors_ptr = num_vendors;
return vendors;
}

View File

@@ -12,6 +12,7 @@
#include "cpu.h"
#define _PATH_CPUINFO "/proc/cpuinfo"
#define _PATH_SYS_SYSTEM "/sys/devices/system"
#define _PATH_SYS_CPU "/cpu"
#define _PATH_FREQUENCY "/cpufreq"
@@ -24,9 +25,16 @@
#define _PATH_CACHE_SIZE "/size"
#define _PATH_CACHE_SHARED_MAP "/shared_cpu_map"
#define _PATH_CPUS_PRESENT _PATH_SYS_SYSTEM _PATH_SYS_CPU "/present"
#define _PATH_TOPO_PACKAGE_CPUS "/topology/package_cpus"
#define _PATH_FREQUENCY_MAX_LEN 100
#define _PATH_CACHE_MAX_LEN 200
#define _PATH_PACKAGE_MAX_LEN 200
struct devtree {
char* vendor;
char* model;
};
char* read_file(char* path, int* len);
long get_max_freq_from_file(uint32_t core);
@@ -36,6 +44,11 @@ long get_l1d_cache_size(uint32_t core);
long get_l2_cache_size(uint32_t core);
long get_l3_cache_size(uint32_t core);
int get_num_caches_by_level(struct cpuInfo* cpu, uint32_t level);
int get_num_sockets_package_cpus(struct topology* topo);
int get_ncores_from_cpuinfo(void);
char* get_field_from_cpuinfo(char* CPUINFO_FIELD);
bool is_devtree_compatible(char* str);
char* get_devtree_compatible(int *filelen);
struct devtree** get_devtree_compatible_struct(int *num_vendors);
#endif

View File

@@ -11,6 +11,20 @@
#include "../common/udev.h"
#include "../common/global.h"
static char *hv_vendors_name[] = {
[HV_VENDOR_KVM] = "KVM",
[HV_VENDOR_QEMU] = "QEMU",
[HV_VENDOR_VBOX] = "VirtualBox",
[HV_VENDOR_HYPERV] = "Microsoft Hyper-V",
[HV_VENDOR_VMWARE] = "VMware",
[HV_VENDOR_XEN] = "Xen",
[HV_VENDOR_PARALLELS] = "Parallels",
[HV_VENDOR_PHYP] = "pHyp",
[HV_VENDOR_BHYVE] = "bhyve",
[HV_VENDOR_APPLEVZ] = "Apple VZ",
[HV_VENDOR_INVALID] = STRING_UNKNOWN
};
struct cache* get_cache_info(struct cpuInfo* cpu) {
struct cache* cach = emalloc(sizeof(struct cache));
init_cache_struct(cach);
@@ -63,12 +77,21 @@ struct topology* get_topology_info(struct cache* cach) {
printWarn("fill_core_ids_from_sys failed, output may be incomplete/invalid");
for(int i=0; i < topo->total_cores; i++) core_ids[i] = 0;
}
if(!fill_package_ids_from_sys(package_ids, topo->total_cores)) {
printWarn("fill_package_ids_from_sys failed, output may be incomplete/invalid");
for(int i=0; i < topo->total_cores; i++) package_ids[i] = 0;
// fill_package_ids_from_sys failed, use udev to try
// to find the number of sockets
topo->sockets = get_num_sockets_package_cpus(topo);
if (topo->sockets == UNKNOWN_DATA) {
printWarn("get_num_sockets_package_cpus failed: assuming 1 socket");
topo->sockets = 1;
}
// 2. Socket detection
}
else {
// fill_package_ids_from_sys succeeded, use the
// traditional socket detection algorithm
int *package_ids_count = emalloc(sizeof(int) * topo->total_cores);
for(int i=0; i < topo->total_cores; i++) {
package_ids_count[i] = 0;
@@ -81,6 +104,8 @@ struct topology* get_topology_info(struct cache* cach) {
topo->sockets++;
}
}
free(package_ids_count);
}
// 3. Physical cores detection
int *core_ids_unified = emalloc(sizeof(int) * topo->total_cores);
@@ -105,7 +130,6 @@ struct topology* get_topology_info(struct cache* cach) {
free(core_ids);
free(package_ids);
free(package_ids_count);
free(core_ids_unified);
return topo;
@@ -126,9 +150,16 @@ struct uarch* get_cpu_uarch(struct cpuInfo* cpu) {
struct frequency* get_frequency_info(void) {
struct frequency* freq = emalloc(sizeof(struct frequency));
freq->measured = false;
freq->max = get_max_freq_from_file(0);
freq->base = get_min_freq_from_file(0);
if(freq->max == UNKNOWN_DATA) {
// If we are unable to find it in the
// standard path, try /proc/cpuinfo
freq->max = get_frequency_from_cpuinfo();
}
return freq;
}
@@ -158,6 +189,28 @@ int64_t get_peak_performance(struct cpuInfo* cpu, struct topology* topo, int64_t
return flops;
}
struct hypervisor* get_hp_info(void) {
struct hypervisor* hv = emalloc(sizeof(struct hypervisor));
hv->present = false;
// Weird heuristic found in lscpu:
// https://github.com/util-linux/util-linux/blob/master/sys-utils/lscpu-virt.c
if(access("/proc" _PATH_DT_IBM_PARTIT_NAME, F_OK) == 0 &&
access("/proc" _PATH_DT_HMC_MANAGED, F_OK) == 0 &&
access("/proc" _PATH_DT_QEMU_WIDTH, F_OK) != 0) {
hv->present = true;
hv->hv_vendor = HV_VENDOR_PHYP;
}
else if(is_devtree_compatible("qemu,pseries")) {
hv->present = true;
hv->hv_vendor = HV_VENDOR_QEMU;
}
hv->hv_name = hv_vendors_name[hv->hv_vendor];
return hv;
}
struct cpuInfo* get_cpu_info(void) {
struct cpuInfo* cpu = emalloc(sizeof(struct cpuInfo));
struct features* feat = emalloc(sizeof(struct features));
@@ -172,8 +225,11 @@ struct cpuInfo* get_cpu_info(void) {
char* path = emalloc(sizeof(char) * (strlen(_PATH_DT) + strlen(_PATH_DT_PART) + 1));
sprintf(path, "%s%s", _PATH_DT, _PATH_DT_PART);
cpu->cpu_name = read_file(path, &len);
if((cpu->cpu_name = read_file(path, &len)) == NULL) {
printWarn("Could not open '%s'", path);
}
cpu->pvr = mfpvr();
cpu->hv = get_hp_info();
cpu->arch = get_cpu_uarch(cpu);
cpu->freq = get_frequency_info();
cpu->topo = get_topology_info(cpu->cach);

View File

@@ -25,6 +25,7 @@ enum {
UARCH_PPC603,
UARCH_PPC440,
UARCH_PPC470,
UARCH_ESPRESSO, // Not exactly an uarch, but the codename of Wii U
UARCH_PPC970,
UARCH_PPC970FX,
UARCH_PPC970MP,
@@ -35,6 +36,7 @@ enum {
UARCH_POWER7,
UARCH_POWER7PLUS,
UARCH_POWER8,
UARCH_POWER8_DD21,
UARCH_POWER9,
UARCH_POWER9_DD20,
UARCH_POWER9_DD21,
@@ -74,6 +76,7 @@ void fill_uarch(struct uarch* arch, MICROARCH u) {
FILL_UARCH(arch->uarch, UARCH_PPC603, "PowerPC 603", UNK) // varies
FILL_UARCH(arch->uarch, UARCH_PPC440, "PowerPC 440", UNK)
FILL_UARCH(arch->uarch, UARCH_PPC470, "PowerPC 470", 45) // strange...
FILL_UARCH(arch->uarch, UARCH_ESPRESSO, "Espresso", 45) // https://en.wikipedia.org/wiki/PowerPC_7xx#Espresso, https://en.wikipedia.org/wiki/Espresso_(processor), https://github.com/Dr-Noob/cpufetch/issues/231
FILL_UARCH(arch->uarch, UARCH_PPC970, "PowerPC 970", 130)
FILL_UARCH(arch->uarch, UARCH_PPC970FX, "PowerPC 970FX", 90)
FILL_UARCH(arch->uarch, UARCH_PPC970MP, "PowerPC 970MP", 90)
@@ -84,6 +87,7 @@ void fill_uarch(struct uarch* arch, MICROARCH u) {
FILL_UARCH(arch->uarch, UARCH_POWER7, "POWER7", 45)
FILL_UARCH(arch->uarch, UARCH_POWER7PLUS, "POWER7+", 32)
FILL_UARCH(arch->uarch, UARCH_POWER8, "POWER8", 22)
FILL_UARCH(arch->uarch, UARCH_POWER8_DD21, "POWER8 (DD2.1)", 22)
FILL_UARCH(arch->uarch, UARCH_POWER9, "POWER9", 14)
FILL_UARCH(arch->uarch, UARCH_POWER9_DD20, "POWER9 (DD2.0)", 14)
FILL_UARCH(arch->uarch, UARCH_POWER9_DD21, "POWER9 (DD2.1)", 14)
@@ -100,7 +104,19 @@ void fill_uarch(struct uarch* arch, MICROARCH u) {
}
/*
* PVR masks/values from arch/powerpc/kernel/cputable.c (Linux kernel)
* PVR masks/values from Linux kernel:
* - arch/powerpc/kernel/cputable.c (kernel <= 6.0)
* - arch/powerpc/kernel/cpu_specs_book3s_64.h (kernel >= 6.1)
*
* In the kernel, there is a POWER8E identifier. In
* https://wiki.raptorcs.com/wiki/POWER8E it says it is
* actually DD2.1, while other POWER8 should be DD2.0.
* The last assumption does not seem to be correct according
* to https://openbenchmarking.org/s/POWER8NVL, which shows a
* POWER8NVL where kernel says it is DD1.0. We implement this
* to show only the uarch, not the revision, since it seems a bit
* redundant?
*
* This list may be incorrect, incomplete or overly simplified,
* specially in the case of 32 bit entries
*/
@@ -125,7 +141,7 @@ struct uarch* get_uarch_from_pvr(uint32_t pvr) {
CHECK_UARCH(arch, pvr, 0xffffffff, 0x0f000006, UARCH_POWER10)
CHECK_UARCH(arch, pvr, 0xffff0000, 0x003f0000, UARCH_POWER7)
CHECK_UARCH(arch, pvr, 0xffff0000, 0x004A0000, UARCH_POWER7PLUS)
CHECK_UARCH(arch, pvr, 0xffff0000, 0x004b0000, UARCH_POWER8)
CHECK_UARCH(arch, pvr, 0xffff0000, 0x004b0000, UARCH_POWER8_DD21)
CHECK_UARCH(arch, pvr, 0xffff0000, 0x004c0000, UARCH_POWER8)
CHECK_UARCH(arch, pvr, 0xffff0000, 0x004d0000, UARCH_POWER8)
CHECK_UARCH(arch, pvr, 0xffffefff, 0x004e0200, UARCH_POWER9_DD20)
@@ -220,6 +236,7 @@ struct uarch* get_uarch_from_pvr(uint32_t pvr) {
CHECK_UARCH(arch, pvr, 0xffff0000, 0x7ff50000, UARCH_PPC470)
CHECK_UARCH(arch, pvr, 0xffff0000, 0x00050000, UARCH_PPC470)
CHECK_UARCH(arch, pvr, 0xffff0000, 0x11a50000, UARCH_PPC470)
CHECK_UARCH(arch, pvr, 0xffffffff, 0x70010201, UARCH_ESPRESSO)
UARCH_END
return arch;
@@ -234,6 +251,7 @@ bool has_altivec(struct uarch* arch) {
case UARCH_POWER7:
case UARCH_POWER7PLUS:
case UARCH_POWER8:
case UARCH_POWER8_DD21:
case UARCH_POWER9:
case UARCH_POWER9_DD20:
case UARCH_POWER9_DD21:
@@ -265,9 +283,6 @@ char* get_str_process(struct cpuInfo* cpu) {
if(process == UNK) {
snprintf(str, strlen(STRING_UNKNOWN)+1, STRING_UNKNOWN);
}
else if(process > 100) {
sprintf(str, "%.2fum", (double)process/100);
}
else if(process > 0){
sprintf(str, "%dnm", process);
}

View File

@@ -1,16 +1,19 @@
#include <errno.h>
#include "../common/udev.h"
#include "../common/global.h"
#include "udev.h"
#define _PATH_TOPO_CORE_ID "topology/core_id"
#define _PATH_TOPO_PACKAGE_ID "topology/physical_package_id"
#define CPUINFO_FREQUENCY_STR "clock\t\t: "
bool fill_array_from_sys(int *core_ids, int total_cores, char* SYS_PATH) {
int filelen;
char* buf;
char* end;
char path[128];
memset(name, 0, sizeof(char) * 128);
for(int i=0; i < total_cores; i++) {
sprintf(path, "%s%s/cpu%d/%s", _PATH_SYS_SYSTEM, _PATH_SYS_CPU, i, SYS_PATH);
@@ -36,5 +39,52 @@ bool fill_core_ids_from_sys(int *core_ids, int total_cores) {
}
bool fill_package_ids_from_sys(int* package_ids, int total_cores) {
return fill_array_from_sys(package_ids, total_cores, _PATH_TOPO_PACKAGE_ID);
bool status = fill_array_from_sys(package_ids, total_cores, _PATH_TOPO_PACKAGE_ID);
if(status) {
// fill_array_from_sys completed successfully, but we
// must to check the integrity of the package_ids array
for(int i=0; i < total_cores; i++) {
if(package_ids[i] == -1) {
printWarn("fill_package_ids_from_sys: package_ids[%d] = -1", i);
return false;
}
else if(package_ids[i] >= total_cores || package_ids[i] < 0) {
printBug("fill_package_ids_from_sys: package_ids[%d] = %d", i, package_ids[i]);
return false;
}
}
return true;
}
return false;
}
long get_frequency_from_cpuinfo(void) {
char* freq_str = get_field_from_cpuinfo(CPUINFO_FREQUENCY_STR);
if(freq_str == NULL) {
return UNKNOWN_DATA;
}
else {
// freq_str should be in the form XXXX.YYYYYYMHz
char* dot = strstr(freq_str, ".");
freq_str[dot-freq_str] = '\0';
char* end;
errno = 0;
long ret = strtol(freq_str, &end, 10);
if(errno != 0) {
printBug("strtol: %s", strerror(errno));
free(freq_str);
return UNKNOWN_DATA;
}
// We consider it an error if frequency is
// greater than 10 GHz or less than 100 MHz
if(ret > 10000 || ret < 100) {
printBug("Invalid data was read from file '%s': %ld\n", CPUINFO_FREQUENCY_STR, ret);
return UNKNOWN_DATA;
}
return ret;
}
}

View File

@@ -4,8 +4,13 @@
#include "../common/udev.h"
#define _PATH_DT "/proc/device-tree/vpd/root-node-vpd@a000/enclosure@1e00/backplane@800/processor@1000"
#define _PATH_DT_PART "/part-number"
#define _PATH_DT_IBM_PARTIT_NAME "/device-tree/ibm,partition-name"
#define _PATH_DT_HMC_MANAGED "/device-tree/hmc-managed?"
#define _PATH_DT_QEMU_WIDTH "/device-tree/chosen/qemu,graphic-width"
bool fill_core_ids_from_sys(int *core_ids, int total_cores);
bool fill_package_ids_from_sys(int* package_ids, int total_cores);
int get_num_sockets_package_cpus(struct topology* topo);
long get_frequency_from_cpuinfo(void);
#endif

View File

@@ -9,9 +9,17 @@
#include "uarch.h"
#include "soc.h"
#define SET_ISA_EXT_MAP(name, bit) \
if(strncmp(multi_letter_extension, name, \
multi_letter_extension_len) == 0) { \
ext->mask |= 1UL << bit; \
maskset = true; \
} \
struct frequency* get_frequency_info(uint32_t core) {
struct frequency* freq = emalloc(sizeof(struct frequency));
freq->measured = false;
freq->base = UNKNOWN_DATA;
freq->max = get_max_freq_from_file(core);
@@ -28,6 +36,62 @@ int64_t get_peak_performance(struct cpuInfo* cpu) {
return flops;
}
// Returns the length of the multi-letter
// extension, or -1 if an error occurs
int parse_multi_letter_extension(struct extensions* ext, char* e) {
if(*e != '_') return -1;
char* multi_letter_extension_end = strstr(e+1, "_");
if(multi_letter_extension_end == NULL) {
// This is the last extension, find the end
// of the string
multi_letter_extension_end = e + strlen(e);
}
int multi_letter_extension_len = multi_letter_extension_end-(e+1);
bool maskset = false;
char* multi_letter_extension = emalloc(multi_letter_extension_len);
strncpy(multi_letter_extension, e+1, multi_letter_extension_len);
// This should be up-to-date with
// https://elixir.bootlin.com/linux/latest/source/arch/riscv/kernel/cpufeature.c
// which should represent the list of extensions available in real chips
SET_ISA_EXT_MAP("sscofpmf", RISCV_ISA_EXT_SSCOFPMF)
SET_ISA_EXT_MAP("sstc", RISCV_ISA_EXT_SSTC)
SET_ISA_EXT_MAP("svinval", RISCV_ISA_EXT_SVINVAL)
SET_ISA_EXT_MAP("svpbmt", RISCV_ISA_EXT_SVPBMT)
SET_ISA_EXT_MAP("zbb", RISCV_ISA_EXT_ZBB)
SET_ISA_EXT_MAP("zicbom", RISCV_ISA_EXT_ZICBOM)
SET_ISA_EXT_MAP("zihintpause", RISCV_ISA_EXT_ZIHINTPAUSE)
SET_ISA_EXT_MAP("svnapot", RISCV_ISA_EXT_SVNAPOT)
SET_ISA_EXT_MAP("zicbop", RISCV_ISA_EXT_ZICBOP)
SET_ISA_EXT_MAP("zicboz", RISCV_ISA_EXT_ZICBOZ)
SET_ISA_EXT_MAP("smaia", RISCV_ISA_EXT_SMAIA)
SET_ISA_EXT_MAP("ssaia", RISCV_ISA_EXT_SSAIA)
SET_ISA_EXT_MAP("zba", RISCV_ISA_EXT_ZBA)
SET_ISA_EXT_MAP("zbs", RISCV_ISA_EXT_ZBS)
SET_ISA_EXT_MAP("zicntr", RISCV_ISA_EXT_ZICNTR)
SET_ISA_EXT_MAP("zicsr", RISCV_ISA_EXT_ZICSR)
SET_ISA_EXT_MAP("zifencei", RISCV_ISA_EXT_ZIFENCEI)
SET_ISA_EXT_MAP("zihpm", RISCV_ISA_EXT_ZIHPM)
if(!maskset) {
printBug("parse_multi_letter_extension: Unknown multi-letter extension: %s", multi_letter_extension);
return -1;
}
return multi_letter_extension_len;
}
bool valid_extension(char ext) {
bool found = false;
uint64_t idx = 0;
while(idx < sizeof(extension_list)/sizeof(extension_list[0]) && !found) {
found = (extension_list[idx].id == (ext - 'a'));
if(!found) idx++;
}
return found;
}
struct extensions* get_extensions_from_str(char* str) {
struct extensions* ext = emalloc(sizeof(struct extensions));
ext->mask = 0;
@@ -37,12 +101,11 @@ struct extensions* get_extensions_from_str(char* str) {
return ext;
}
int len = sizeof(char) * (strlen(str)+1);
ext->str = emalloc(sizeof(char) * len);
memset(ext->str, 0, len);
int len = strlen(str);
ext->str = ecalloc(len+1, sizeof(char));
strncpy(ext->str, str, sizeof(char) * len);
// Code inspired in Linux kernel:
// Code inspired in Linux kernel (riscv_fill_hwcap):
// https://elixir.bootlin.com/linux/v6.2.10/source/arch/riscv/kernel/cpufeature.c
char* isa = str;
if (!strncmp(isa, "rv32", 4))
@@ -55,9 +118,33 @@ struct extensions* get_extensions_from_str(char* str) {
}
for(char* e = isa; *e != '\0'; e++) {
if(*e == '_') {
// Multi-letter extension
int multi_letter_extension_len = parse_multi_letter_extension(ext, e);
if(multi_letter_extension_len == -1) {
return ext;
}
e += multi_letter_extension_len;
}
else {
// Single-letter extensions 's' and 'u' are invalid
// according to Linux kernel (arch/riscv/kernel/cpufeature.c:
// riscv_fill_hwcap). Optionally, we could opt for using
// hwcap instead of cpuinfo to avoid this
if (*e == 's' || *e == 'u') {
continue;
}
// Make sure that the extension is valid before
// adding it to the mask
if(valid_extension(*e)) {
int n = *e - 'a';
ext->mask |= 1UL << n;
}
else {
printBug("get_extensions_from_str: Invalid extension: '%c'", *e);
}
}
}
return ext;
}
@@ -77,7 +164,7 @@ struct cpuInfo* get_cpu_info(void) {
cpu->ext = get_extensions_from_str(ext_str);
if(cpu->ext->str != NULL && cpu->ext->mask == 0) return NULL;
cpu->arch = get_uarch_from_cpuinfo_str(cpuinfo_str, cpu);
cpu->soc = get_soc();
cpu->soc = get_soc(cpu);
cpu->freq = get_frequency_info(0);
cpu->peak_performance = get_peak_performance(cpu);
@@ -101,5 +188,20 @@ char* get_str_extensions(struct cpuInfo* cpu) {
}
void print_debug(struct cpuInfo* cpu) {
printf("Unimplemented!\n");
printf("- soc: ");
if(cpu->soc->raw_name == NULL) {
printf("NULL\n");
}
else {
printf("'%s'\n", cpu->soc->raw_name);
}
printf("- uarch: ");
char* arch_cpuinfo_str = get_arch_cpuinfo_str(cpu);
if(arch_cpuinfo_str == NULL) {
printf("NULL\n");
}
else {
printf("'%s'\n", arch_cpuinfo_str);
}
}

View File

@@ -8,7 +8,37 @@ struct extension {
char* str;
};
#define RISCV_ISA_EXT_NAME_LEN_MAX 32
#define RISCV_ISA_EXT_BASE 26
// https://elixir.bootlin.com/linux/latest/source/arch/riscv/include/asm/hwcap.h
// This enum represent the logical ID for multi-letter RISC-V ISA extensions.
// The logical ID should start from RISCV_ISA_EXT_BASE
enum riscv_isa_ext_id {
RISCV_ISA_EXT_SSCOFPMF = RISCV_ISA_EXT_BASE,
RISCV_ISA_EXT_SSTC,
RISCV_ISA_EXT_SVINVAL,
RISCV_ISA_EXT_SVPBMT,
RISCV_ISA_EXT_ZBB,
RISCV_ISA_EXT_ZICBOM,
RISCV_ISA_EXT_ZIHINTPAUSE,
RISCV_ISA_EXT_SVNAPOT,
RISCV_ISA_EXT_ZICBOP,
RISCV_ISA_EXT_ZICBOZ,
RISCV_ISA_EXT_SMAIA,
RISCV_ISA_EXT_SSAIA,
RISCV_ISA_EXT_ZBA,
RISCV_ISA_EXT_ZBS,
RISCV_ISA_EXT_ZICNTR,
RISCV_ISA_EXT_ZICSR,
RISCV_ISA_EXT_ZIFENCEI,
RISCV_ISA_EXT_ZIHPM,
RISCV_ISA_EXT_ID_MAX
};
// https://five-embeddev.com/riscv-isa-manual/latest/preface.html#preface
// https://en.wikichip.org/wiki/risc-v/standard_extensions
// (Zicbop) https://github.com/riscv/riscv-CMOs/blob/master/cmobase/Zicbop.adoc
// Included all except for G
static const struct extension extension_list[] = {
{ 'i' - 'a', "(I) Integer Instruction Set" },
@@ -26,7 +56,25 @@ static const struct extension extension_list[] = {
{ 'v' - 'a', "(V) Vector Operations" },
{ 'n' - 'a', "(N) User-Level Interrupts" },
{ 'h' - 'a', "(H) Hypervisor" },
{ 's' - 'a', "(S) Supervisor-level Instructions" }
// multi-letter extensions
{ RISCV_ISA_EXT_SSCOFPMF, "(Sscofpmf) Count OverFlow and Privilege Mode Filtering" },
{ RISCV_ISA_EXT_SSTC, "(Sstc) S and VS level Time Compare" },
{ RISCV_ISA_EXT_SVINVAL, "(Svinval) Fast TLB Invalidation" },
{ RISCV_ISA_EXT_SVPBMT, "(Svpbmt) Page-based Memory Types" },
{ RISCV_ISA_EXT_ZBB, "(Zbb) Basic bit-manipulation" },
{ RISCV_ISA_EXT_ZICBOM, "(Zicbom) Cache Block Management Operations" },
{ RISCV_ISA_EXT_ZIHINTPAUSE, "(Zihintpause) Pause Hint" },
{ RISCV_ISA_EXT_SVNAPOT, "(Svnapot) Naturally Aligned Power of Two Pages" },
{ RISCV_ISA_EXT_ZICBOZ, "(Zicboz) Cache Block Zero Operations" },
{ RISCV_ISA_EXT_ZICBOP, "(Zicbop) Cache Block Prefetch Operations" },
{ RISCV_ISA_EXT_SMAIA, "(Smaia) Advanced Interrupt Architecture" },
{ RISCV_ISA_EXT_SSAIA, "(Ssaia) Advanced Interrupt Architecture" },
{ RISCV_ISA_EXT_ZBA, "(Zba) Address Generation" },
{ RISCV_ISA_EXT_ZBS, "(Zbs) Single-bit Instructions" },
{ RISCV_ISA_EXT_ZICNTR, "(Zicntr) Base Counters and Timers" },
{ RISCV_ISA_EXT_ZICSR, "(Zicsr) Control and Status Register" },
{ RISCV_ISA_EXT_ZIFENCEI, "(Zifencei) Instruction-Fetch Fence" },
{ RISCV_ISA_EXT_ZIHPM, "(Zihpm) Hardware Performance Counters" }
};
struct cpuInfo* get_cpu_info(void);

View File

@@ -5,66 +5,6 @@
#include <string.h>
VENDOR get_soc_vendor(struct system_on_chip* soc) {
return soc->soc_vendor;
}
char* get_str_process(struct system_on_chip* soc) {
char* str;
if(soc->process == UNKNOWN) {
str = emalloc(sizeof(char) * (strlen(STRING_UNKNOWN)+1));
snprintf(str, strlen(STRING_UNKNOWN)+1, STRING_UNKNOWN);
}
else {
str = emalloc(sizeof(char) * 5);
memset(str, 0, sizeof(char) * 5);
snprintf(str, 5, "%dnm", soc->process);
}
return str;
}
char* get_soc_name(struct system_on_chip* soc) {
if(soc->soc_model == SOC_MODEL_UNKNOWN)
return soc->raw_name;
return soc->soc_name;
}
static char* soc_trademark_string[] = {
[SOC_VENDOR_SIFIVE] = "SiFive ",
[SOC_VENDOR_STARFIVE] = "StarFive ",
[SOC_VENDOR_ALLWINNER] = "Allwinner "
};
void fill_soc(struct system_on_chip* soc, char* soc_name, SOC soc_model, int32_t process) {
soc->soc_model = soc_model;
soc->soc_vendor = get_soc_vendor_from_soc(soc_model);
soc->process = process;
int len = strlen(soc_name) + strlen(soc_trademark_string[soc->soc_vendor]) + 1;
soc->soc_name = emalloc(sizeof(char) * len);
memset(soc->soc_name, 0, sizeof(char) * len);
sprintf(soc->soc_name, "%s%s", soc_trademark_string[soc->soc_vendor], soc_name);
}
bool match_soc(struct system_on_chip* soc, char* raw_name, char* expected_name, char* soc_name, SOC soc_model, int32_t process) {
int len1 = strlen(raw_name);
int len2 = strlen(expected_name);
int len = min(len1, len2);
if(strncmp(raw_name, expected_name, len) != 0) {
return false;
}
else {
fill_soc(soc, soc_name, soc_model, process);
return true;
}
}
#define SOC_START if (false) {}
#define SOC_EQ(raw_name, expected_name, soc_name, soc_model, soc, process) \
else if (match_soc(soc, raw_name, expected_name, soc_name, soc_model, process)) return true;
#define SOC_END else { return false; }
bool match_sifive(char* soc_name, struct system_on_chip* soc) {
char* tmp = soc_name;
@@ -72,7 +12,7 @@ bool match_sifive(char* soc_name, struct system_on_chip* soc) {
/*if((tmp = strstr(soc_name, "???")) == NULL)
return false;*/
//soc->soc_vendor = ???
//soc->vendor = ???
SOC_START
SOC_EQ(tmp, "fu740", "Freedom U740", SOC_SIFIVE_U740, soc, 40)
@@ -92,6 +32,12 @@ bool match_allwinner(char* soc_name, struct system_on_chip* soc) {
SOC_END
}
bool match_sipeed(char* soc_name, struct system_on_chip* soc) {
SOC_START
SOC_EQ(soc_name, "light", "Lichee Pi 4A", SOC_SIPEED_LICHEEPI4A, soc, 12) // https://github.com/Dr-Noob/cpufetch/issues/200, https://sipeed.com/licheepi4a
SOC_END
}
struct system_on_chip* parse_soc_from_string(struct system_on_chip* soc) {
char* raw_name = soc->raw_name;
@@ -101,7 +47,10 @@ struct system_on_chip* parse_soc_from_string(struct system_on_chip* soc) {
if(match_allwinner(raw_name, soc))
return soc;
match_sifive(raw_name, soc);
if(match_sifive(raw_name, soc))
return soc;
match_sipeed(raw_name, soc);
return soc;
}
@@ -116,15 +65,15 @@ struct system_on_chip* guess_soc_from_devtree(struct system_on_chip* soc) {
return soc;
}
struct system_on_chip* get_soc(void) {
struct system_on_chip* get_soc(struct cpuInfo* cpu) {
struct system_on_chip* soc = emalloc(sizeof(struct system_on_chip));
soc->raw_name = NULL;
soc->soc_vendor = SOC_VENDOR_UNKNOWN;
soc->soc_model = SOC_MODEL_UNKNOWN;
soc->vendor = SOC_VENDOR_UNKNOWN;
soc->model = SOC_MODEL_UNKNOWN;
soc->process = UNKNOWN;
soc = guess_soc_from_devtree(soc);
if(soc->soc_vendor == SOC_VENDOR_UNKNOWN) {
if(soc->vendor == SOC_VENDOR_UNKNOWN) {
if(soc->raw_name != NULL) {
printWarn("SoC detection failed using device tree: Found '%s' string", soc->raw_name);
}
@@ -133,7 +82,7 @@ struct system_on_chip* get_soc(void) {
}
}
if(soc->soc_model == SOC_MODEL_UNKNOWN) {
if(soc->model == SOC_MODEL_UNKNOWN) {
// raw_name might not be NULL, but if we were unable to find
// the exact SoC, just print "Unkwnown"
soc->raw_name = emalloc(sizeof(char) * (strlen(STRING_UNKNOWN)+1));

View File

@@ -1,29 +1,10 @@
#ifndef __SOC__
#define __SOC__
#ifndef __SOC_RISCV__
#define __SOC_RISCV__
#include "../common/soc.h"
#include "../common/cpu.h"
#include <stdint.h>
typedef int32_t SOC;
enum {
SOC_VENDOR_UNKNOWN,
SOC_VENDOR_SIFIVE,
SOC_VENDOR_STARFIVE,
SOC_VENDOR_ALLWINNER
};
struct system_on_chip {
SOC soc_model;
VENDOR soc_vendor;
int32_t process;
char* soc_name;
char* raw_name;
};
struct system_on_chip* get_soc(void);
char* get_soc_name(struct system_on_chip* soc);
VENDOR get_soc_vendor(struct system_on_chip* soc);
char* get_str_process(struct system_on_chip* soc);
struct system_on_chip* get_soc(struct cpuInfo* cpu);
#endif

View File

@@ -11,6 +11,8 @@ enum {
SOC_STARFIVE_VF2,
// ALLWINNER
SOC_ALLWINNER_D1H,
// SIPEED
SOC_SIPEED_LICHEEPI4A,
// UNKNOWN
SOC_MODEL_UNKNOWN
};
@@ -19,6 +21,7 @@ inline static VENDOR get_soc_vendor_from_soc(SOC soc) {
if(soc >= SOC_SIFIVE_U740 && soc <= SOC_SIFIVE_U740) return SOC_VENDOR_SIFIVE;
if(soc >= SOC_STARFIVE_VF2 && soc <= SOC_STARFIVE_VF2) return SOC_VENDOR_STARFIVE;
if(soc >= SOC_ALLWINNER_D1H && soc <= SOC_ALLWINNER_D1H) return SOC_VENDOR_ALLWINNER;
if(soc >= SOC_SIPEED_LICHEEPI4A && soc <= SOC_SIPEED_LICHEEPI4A) return SOC_VENDOR_SIPEED;
return SOC_VENDOR_UNKNOWN;
}

View File

@@ -11,6 +11,7 @@ typedef uint32_t MICROARCH;
struct uarch {
MICROARCH uarch;
char* uarch_str;
char* cpuinfo_str;
};
enum {
@@ -40,6 +41,7 @@ void fill_uarch(struct uarch* arch, struct cpuInfo* cpu, char* str, MICROARCH u,
// T-Head: https://www.t-head.cn/product/c906
struct uarch* get_uarch_from_cpuinfo_str(char* cpuinfo_str, struct cpuInfo* cpu) {
struct uarch* arch = emalloc(sizeof(struct uarch));
arch->cpuinfo_str = cpuinfo_str;
if(cpuinfo_str == NULL) {
printWarn("get_uarch_from_cpuinfo: Unable to detect microarchitecture, cpuinfo_str is NULL");
fill_uarch(arch, cpu, "Unknown", UARCH_UNKNOWN, CPU_VENDOR_UNKNOWN);
@@ -72,6 +74,10 @@ char* get_str_uarch(struct cpuInfo* cpu) {
return cpu->arch->uarch_str;
}
char* get_arch_cpuinfo_str(struct cpuInfo* cpu) {
return cpu->arch->cpuinfo_str;
}
void free_uarch_struct(struct uarch* arch) {
free(arch->uarch_str);
free(arch);

View File

@@ -6,6 +6,7 @@
struct uarch;
char* get_arch_cpuinfo_str(struct cpuInfo* cpu);
char* get_str_uarch(struct cpuInfo* cpu);
void free_uarch_struct(struct uarch* arch);
struct uarch* get_uarch_from_cpuinfo_str(char* cpuinfo_str, struct cpuInfo* cpu);

View File

@@ -4,7 +4,6 @@
#include "../common/global.h"
#include "udev.h"
#define _PATH_CPUINFO "/proc/cpuinfo"
#define _PATH_DEVTREE "/proc/device-tree/compatible"
#define CPUINFO_UARCH_STR "uarch\t\t: "
#define CPUINFO_EXTENSIONS_STR "isa\t\t: "
@@ -41,8 +40,7 @@ char* get_field_from_devtree(int DEVTREE_FIELD) {
tmp1++;
int strlen = filelen-(tmp1-buf);
char* hardware = emalloc(sizeof(char) * strlen);
memset(hardware, 0, sizeof(char) * strlen);
char* hardware = ecalloc(strlen, sizeof(char));
strncpy(hardware, tmp1, strlen-1);
return hardware;
@@ -71,9 +69,8 @@ char* parse_cpuinfo_field(char* field_str) {
}
int ret_strlen = (end-tmp);
char* ret = emalloc(sizeof(char) * (ret_strlen+1));
memset(ret, 0, sizeof(char) * (ret_strlen+1));
strncpy(ret, tmp, ret_strlen);
char* ret = ecalloc(ret_strlen+1, sizeof(char));
strncpy(ret, tmp, sizeof(char) * ret_strlen);
return ret;
}
@@ -86,7 +83,7 @@ char* get_uarch_from_cpuinfo(void) {
return parse_cpuinfo_field(CPUINFO_UARCH_STR);
}
char* get_extensions_from_cpuinfo() {
char* get_extensions_from_cpuinfo(void) {
return parse_cpuinfo_field(CPUINFO_EXTENSIONS_STR);
}

View File

@@ -72,34 +72,6 @@ uint32_t get_apic_id(bool x2apic_id) {
}
}
#ifndef __APPLE__
bool bind_to_cpu(int cpu_id) {
#ifdef _WIN32
HANDLE process = GetCurrentProcess();
DWORD_PTR processAffinityMask = 1 << cpu_id;
return SetProcessAffinityMask(process, processAffinityMask);
#elif defined __linux__
cpu_set_t currentCPU;
CPU_ZERO(&currentCPU);
CPU_SET(cpu_id, &currentCPU);
if (sched_setaffinity (0, sizeof(currentCPU), &currentCPU) == -1) {
printWarn("sched_setaffinity: %s", strerror(errno));
return false;
}
return true;
#elif defined __FreeBSD__
cpuset_t currentCPU;
CPU_ZERO(&currentCPU);
CPU_SET(cpu_id, &currentCPU);
if(cpuset_setaffinity(CPU_LEVEL_WHICH, CPU_WHICH_TID, -1, sizeof(cpuset_t), &currentCPU) == -1) {
printWarn("cpuset_setaffinity: %s", strerror(errno));
return false;
}
return true;
#endif
}
#endif
#ifdef __linux__
int get_total_cores_module(int total_cores, int module) {
int total_modules = 2;
@@ -262,15 +234,11 @@ uint32_t max_apic_id_size(uint32_t** cache_id_apic, struct topology* topo) {
bool build_topo_from_apic(uint32_t* apic_pkg, uint32_t* apic_smt, uint32_t** cache_id_apic, struct topology* topo) {
uint32_t size = max_apic_id_size(cache_id_apic, topo);
uint32_t* sockets = emalloc(sizeof(uint32_t) * size);
uint32_t* smt = emalloc(sizeof(uint32_t) * size);
uint32_t* apic_id = emalloc(sizeof(uint32_t) * size);
uint32_t* sockets = ecalloc(size, sizeof(uint32_t));
uint32_t* smt = ecalloc(size, sizeof(uint32_t));
uint32_t* apic_id = ecalloc(size, sizeof(uint32_t));
uint32_t num_caches = 0;
memset(sockets, 0, sizeof(uint32_t) * size);
memset(smt, 0, sizeof(uint32_t) * size);
memset(apic_id, 0, sizeof(uint32_t) * size);
// System topology
for(int i=0; i < topo->total_cores_module; i++) {
sockets[apic_pkg[i]] = 1;
@@ -397,6 +365,11 @@ bool fill_apic_ids(uint32_t* apic_ids, int first_core, int n, bool x2apic_id) {
}
bool get_topology_from_apic(struct cpuInfo* cpu, struct topology* topo) {
if (topo->cach == NULL) {
printWarn("get_topology_from_apic: cach is NULL");
return false;
}
uint32_t apic_id;
uint32_t* apic_ids = emalloc(sizeof(uint32_t) * topo->total_cores_module);
uint32_t* apic_pkg = emalloc(sizeof(uint32_t) * topo->total_cores_module);

View File

@@ -17,10 +17,6 @@ struct apic {
bool get_topology_from_apic(struct cpuInfo* cpu, struct topology* topo);
uint32_t is_smt_enabled_amd(struct topology* topo);
#ifndef __APPLE__
bool bind_to_cpu(int cpu_id);
#endif
#ifdef __linux__
int get_total_cores_module(int total_cores, int module);
#endif

View File

@@ -5,6 +5,13 @@
#include "../common/udev.h"
#include <unistd.h>
#endif
#if defined (__FreeBSD__) || defined (__APPLE__)
#include "../common/sysctl.h"
#endif
#ifdef __linux__
#include "../common/freq.h"
#endif
#include <stdio.h>
#include <stdlib.h>
@@ -22,28 +29,35 @@
#define CPU_VENDOR_INTEL_STRING "GenuineIntel"
#define CPU_VENDOR_AMD_STRING "AuthenticAMD"
#define CPU_VENDOR_HYGON_STRING "HygonGenuine"
static const char *hv_vendors_string[] = {
[HV_VENDOR_KVM] = "KVMKVMKVM",
[HV_VENDOR_QEMU] = "TCGTCGTCGTCG",
[HV_VENDOR_VBOX] = "VBoxVBoxVBox",
[HV_VENDOR_HYPERV] = "Microsoft Hv",
[HV_VENDOR_VMWARE] = "VMwareVMware",
[HV_VENDOR_XEN] = "XenVMMXenVMM",
[HV_VENDOR_PARALLELS] = "lrpepyh vr",
[HV_VENDOR_PHYP] = NULL,
[HV_VENDOR_BHYVE] = "bhyve bhyve ",
[HV_VENDOR_APPLEVZ] = "Apple VZ"
};
static char *hv_vendors_name[] = {
[HV_VENDOR_KVM] = "KVM",
[HV_VENDOR_QEMU] = "QEMU",
[HV_VENDOR_VBOX] = "VirtualBox",
[HV_VENDOR_HYPERV] = "Microsoft Hyper-V",
[HV_VENDOR_VMWARE] = "VMware",
[HV_VENDOR_XEN] = "Xen",
[HV_VENDOR_PARALLELS] = "Parallels",
[HV_VENDOR_PHYP] = "pHyp",
[HV_VENDOR_BHYVE] = "bhyve",
[HV_VENDOR_APPLEVZ] = "Apple VZ",
[HV_VENDOR_INVALID] = STRING_UNKNOWN
};
#define HYPERVISOR_NAME_MAX_LENGTH 17
#define MASK 0xFF
/*
@@ -77,8 +91,7 @@ char* get_str_cpu_name_internal(void) {
uint32_t edx = 0;
uint32_t c = 0;
char * name = emalloc(sizeof(char) * CPU_NAME_MAX_LENGTH);
memset(name, 0, CPU_NAME_MAX_LENGTH);
char * name = ecalloc(CPU_NAME_MAX_LENGTH, sizeof(char));
for(int i=0; i < 3; i++) {
eax = 0x80000002 + i;
@@ -197,22 +210,18 @@ int64_t get_peak_performance(struct cpuInfo* cpu, bool accurate_pp) {
for(int i=0; i < cpu->num_cpus; ptr = ptr->next_cpu, i++) {
struct topology* topo = ptr->topo;
int64_t max_freq = get_freq(ptr->freq);
int64_t freq = get_freq(ptr->freq);
int64_t freq;
#ifdef __linux__
if(accurate_pp)
freq = measure_frequency(ptr);
else
freq = max_freq;
freq = get_freq_pp(ptr->freq);
#else
// Silence compiler warning
(void)(accurate_pp);
freq = max_freq;
#endif
//First, check we have consistent data
if(freq == UNKNOWN_DATA || topo->logical_cores == UNKNOWN_DATA) {
if(freq == UNKNOWN_DATA || topo == NULL || topo->logical_cores == UNKNOWN_DATA) {
return -1;
}
@@ -223,9 +232,10 @@ int64_t get_peak_performance(struct cpuInfo* cpu, bool accurate_pp) {
if(feat->FMA3 || feat->FMA4)
flops = flops*2;
// Ice Lake has AVX512, but it has 1 VPU for AVX512, while
// it has 2 for AVX2. If this is a Ice Lake CPU, we are computing
// the peak performance supposing AVX2, not AVX512
// NOTE:
// Some CPUs (Ice Lake, Zen 4) have AVX512, but they have only
// 1 VPU for AVX512, while they have 2 for AVX2. In such cases,
// we are computing the peak performance supposing AVX2, not AVX512
if(feat->AVX512 && vpus_are_AVX512(ptr))
flops = flops*16;
else if(feat->AVX || feat->AVX2)
@@ -260,15 +270,20 @@ struct hypervisor* get_hp_info(bool hv_present) {
cpuid(&eax, &ebx, &ecx, &edx);
if(ebx == 0x0 && ecx == 0x0 && edx == 0x0) {
hv->hv_vendor = HV_VENDOR_INVALID;
printWarn("Hypervisor vendor is empty");
}
else {
char name[13];
memset(name, 0, 13);
memset(name, 0, sizeof(char) * 13);
get_name_cpuid(name, ebx, ecx, edx);
bool found = false;
uint8_t len = sizeof(hv_vendors_string) / sizeof(hv_vendors_string[0]);
for(uint8_t v=0; v < len && !found; v++) {
if(strcmp(hv_vendors_string[v], name) == 0) {
if(hv_vendors_string[v] != NULL && strcmp(hv_vendors_string[v], name) == 0) {
hv->hv_vendor = v;
found = true;
}
@@ -276,7 +291,8 @@ struct hypervisor* get_hp_info(bool hv_present) {
if(!found) {
hv->hv_vendor = HV_VENDOR_INVALID;
printWarn("Unknown hypervisor vendor: %s", name);
printBug("Unknown hypervisor vendor: '%s'", name);
}
}
hv->hv_name = hv_vendors_name[hv->hv_vendor];
@@ -430,6 +446,23 @@ int32_t get_core_type(void) {
}
}
#ifdef __linux__
// Gets the max frequency for estimating the peak performance,
// filling in the passed cpuInfo parameter with this information.
void fill_frequency_info_pp(struct cpuInfo* cpu) {
int32_t unused;
int32_t *max_freq_pp_vec = malloc(sizeof(int32_t) * cpu->num_cpus);
struct cpuInfo* ptr = cpu;
for (uint32_t i=0; i < cpu->num_cpus; i++) {
set_cpu_module(i, cpu->num_cpus, &unused);
ptr->freq->max_pp = measure_frequency(ptr, max_freq_pp_vec);
ptr = ptr->next_cpu;
}
}
#endif
struct cpuInfo* get_cpu_info(void) {
struct cpuInfo* cpu = emalloc(sizeof(struct cpuInfo));
cpu->peak_performance = -1;
@@ -438,7 +471,7 @@ struct cpuInfo* get_cpu_info(void) {
cpu->cach = NULL;
cpu->feat = NULL;
uint32_t modules = 1;
cpu->num_cpus = 1;
uint32_t eax = 0;
uint32_t ebx = 0;
uint32_t ecx = 0;
@@ -450,13 +483,15 @@ struct cpuInfo* get_cpu_info(void) {
//Fill vendor
char name[13];
memset(name,0,13);
memset(name, 0, sizeof(char) * 13);
get_name_cpuid(name, ebx, edx, ecx);
if(strcmp(CPU_VENDOR_INTEL_STRING,name) == 0)
cpu->cpu_vendor = CPU_VENDOR_INTEL;
else if (strcmp(CPU_VENDOR_AMD_STRING,name) == 0)
cpu->cpu_vendor = CPU_VENDOR_AMD;
else if (strcmp(CPU_VENDOR_HYGON_STRING,name) == 0)
cpu->cpu_vendor = CPU_VENDOR_HYGON;
else {
cpu->cpu_vendor = CPU_VENDOR_INVALID;
printErr("Unknown CPU vendor: %s", name);
@@ -475,9 +510,8 @@ struct cpuInfo* get_cpu_info(void) {
cpu->cpu_name = get_str_cpu_name_internal();
}
else {
cpu->cpu_name = emalloc(sizeof(char) * (strlen(STRING_UNKNOWN) + 1));
strcpy(cpu->cpu_name, STRING_UNKNOWN);
printWarn("Can't read cpu name from cpuid (needed extended level is 0x%.8X, max is 0x%.8X)", 0x80000004, cpu->maxExtendedLevels);
cpu->cpu_name = NULL;
printWarn("Can't read CPU name from cpuid (needed extended level is 0x%.8X, max is 0x%.8X)", 0x80000004, cpu->maxExtendedLevels);
}
cpu->topology_extensions = false;
@@ -495,12 +529,12 @@ struct cpuInfo* get_cpu_info(void) {
cpu->hybrid_flag = (edx >> 15) & 0x1;
}
if(cpu->hybrid_flag) modules = 2;
if(cpu->hybrid_flag) cpu->num_cpus = 2;
struct cpuInfo* ptr = cpu;
for(uint32_t i=0; i < modules; i++) {
for(uint32_t i=0; i < cpu->num_cpus; i++) {
int32_t first_core;
set_cpu_module(i, modules, &first_core);
set_cpu_module(i, cpu->num_cpus, &first_core);
if(i > 0) {
ptr->next_cpu = emalloc(sizeof(struct cpuInfo));
@@ -525,16 +559,18 @@ struct cpuInfo* get_cpu_info(void) {
ptr->core_type = get_core_type();
}
ptr->first_core_id = first_core;
ptr->module_id = i;
ptr->feat = get_features_info(ptr);
// If any field of the struct is NULL,
// return inmideately, as further functions
// require valid fields (cach, topo, etc)
ptr->arch = get_cpu_uarch(ptr);
ptr->freq = get_frequency_info(ptr);
if (cpu->cpu_name == NULL && ptr == cpu) {
// If we couldnt read CPU name from cpuid, infer it now
cpu->cpu_name = infer_cpu_name_from_uarch(cpu->arch);
}
ptr->cach = get_cache_info(ptr);
if(ptr->cach == NULL) return cpu;
if(cpu->hybrid_flag) {
ptr->topo = get_topology_info(ptr, ptr->cach, i);
@@ -542,16 +578,30 @@ struct cpuInfo* get_cpu_info(void) {
else {
ptr->topo = get_topology_info(ptr, ptr->cach, -1);
}
if(cpu->topo == NULL) return cpu;
// If topo is NULL, return early, as get_peak_performance
// requries non-NULL topology.
if(ptr->topo == NULL) return cpu;
}
cpu->num_cpus = modules;
#ifdef __linux__
// If accurate_pp is requested, we need to get the max frequency
// after fetching the topology for all CPU modules, since the topology
// is required by fill_frequency_info_pp
if (accurate_pp()) fill_frequency_info_pp(cpu);
#endif
cpu->peak_performance = get_peak_performance(cpu, accurate_pp());
return cpu;
}
bool get_cache_topology_amd(struct cpuInfo* cpu, struct topology* topo) {
if (topo->cach == NULL) {
printWarn("get_cache_topology_amd: cach is NULL");
return false;
}
if(cpu->maxExtendedLevels >= 0x8000001D && cpu->topology_extensions) {
uint32_t i, eax, ebx, ecx, edx, num_sharing_cache, cache_type, cache_level;
@@ -625,15 +675,24 @@ bool get_cache_topology_amd(struct cpuInfo* cpu, struct topology* topo) {
return true;
}
#ifdef __linux__
void get_topology_from_udev(struct topology* topo) {
// TODO: To be improved in the future
topo->total_cores = get_ncores_from_cpuinfo();
// TODO: To be improved in the future
if (topo->total_cores == 1) {
// We can assume it's a single core CPU
topo->logical_cores = topo->total_cores;
topo->physical_cores = topo->total_cores;
}
else {
topo->logical_cores = UNKNOWN_DATA;
topo->physical_cores = UNKNOWN_DATA;
}
topo->smt_available = 1;
topo->smt_supported = 1;
topo->sockets = 1;
}
#endif
// Main reference: https://software.intel.com/content/www/us/en/develop/articles/intel-64-architecture-processor-topology-enumeration.html
// Very interesting resource: https://wiki.osdev.org/Detecting_CPU_Topology_(80x86)
@@ -673,32 +732,31 @@ struct topology* get_topology_info(struct cpuInfo* cpu, struct cache* cach, int
topo->total_cores_module = topo->total_cores;
}
bool toporet = false;
switch(cpu->cpu_vendor) {
case CPU_VENDOR_INTEL:
if (cpu->maxLevels >= 0x00000004) {
bool toporet = get_topology_from_apic(cpu, topo);
toporet = get_topology_from_apic(cpu, topo);
}
else {
printWarn("Can't read topology information from cpuid (needed level is 0x%.8X, max is 0x%.8X)", 0x00000004, cpu->maxLevels);
}
if(!toporet) {
#ifdef __linux__
printWarn("Failed to retrieve topology from APIC, using udev...\n");
printWarn("Failed to retrieve topology from APIC, using udev...");
get_topology_from_udev(topo);
#else
printErr("Failed to retrieve topology from APIC, assumming default values...\n");
if (cpu->maxLevels >= 0x00000004)
printErr("Failed to retrieve topology from APIC, assumming default values...");
topo->logical_cores = UNKNOWN_DATA;
topo->physical_cores = UNKNOWN_DATA;
topo->smt_available = 1;
topo->smt_supported = 1;
#endif
}
}
else {
printWarn("Can't read topology information from cpuid (needed level is 0x%.8X, max is 0x%.8X)", 0x00000001, cpu->maxLevels);
topo->physical_cores = 1;
topo->logical_cores = 1;
topo->smt_available = 1;
topo->smt_supported = 1;
}
break;
case CPU_VENDOR_AMD:
case CPU_VENDOR_HYGON:
if (cpu->maxExtendedLevels >= 0x80000008) {
eax = 0x80000008;
cpuid(&eax, &ebx, &ecx, &edx);
@@ -715,10 +773,15 @@ struct topology* get_topology_info(struct cpuInfo* cpu, struct cache* cach, int
}
}
else {
#ifdef __linux__
printWarn("Can't read topology information from cpuid (needed extended level is 0x%.8X, max is 0x%.8X), using udev...", 0x80000008, cpu->maxExtendedLevels);
get_topology_from_udev(topo);
#else
printWarn("Can't read topology information from cpuid (needed extended level is 0x%.8X, max is 0x%.8X)", 0x80000008, cpu->maxExtendedLevels);
topo->physical_cores = 1;
topo->logical_cores = 1;
topo->smt_supported = 1;
#endif
}
if (cpu->maxLevels >= 0x00000001) {
@@ -895,16 +958,27 @@ struct cache* get_cache_info(struct cpuInfo* cpu) {
struct frequency* get_frequency_info(struct cpuInfo* cpu) {
struct frequency* freq = emalloc(sizeof(struct frequency));
freq->measured = false;
if(cpu->maxLevels < 0x00000016) {
#if defined (_WIN32) || defined (__APPLE__)
#if defined (_WIN32)
printWarn("Can't read frequency information from cpuid (needed level is 0x%.8X, max is 0x%.8X)", 0x00000016, cpu->maxLevels);
freq->base = UNKNOWN_DATA;
freq->max = UNKNOWN_DATA;
#elif defined (__FreeBSD__) || defined (__APPLE__)
printWarn("Can't read frequency information from cpuid (needed level is 0x%.8X, max is 0x%.8X). Using sysctl", 0x00000016, cpu->maxLevels);
uint32_t freq_hz = get_sys_info_by_name(CPUFREQUENCY_SYSCTL);
if (freq_hz == 0) {
printWarn("Read max CPU frequency from sysctl and got 0 MHz");
freq->max = UNKNOWN_DATA;
}
freq->base = UNKNOWN_DATA;
freq->max = freq_hz;
#else
printWarn("Can't read frequency information from cpuid (needed level is 0x%.8X, max is 0x%.8X). Using udev", 0x00000016, cpu->maxLevels);
freq->base = UNKNOWN_DATA;
freq->max = get_max_freq_from_file(0);
freq->max = get_max_freq_from_file(cpu->first_core_id);
if(freq->max == 0) {
printWarn("Read max CPU frequency from udev and got 0 MHz");
@@ -931,7 +1005,7 @@ struct frequency* get_frequency_info(struct cpuInfo* cpu) {
printWarn("Read max CPU frequency from CPUID and got 0 MHz");
#ifdef __linux__
printWarn("Using udev to detect frequency");
freq->max = get_max_freq_from_file(0);
freq->max = get_max_freq_from_file(cpu->first_core_id);
if(freq->max == 0) {
printWarn("Read max CPU frequency from udev and got 0 MHz");
@@ -943,6 +1017,16 @@ struct frequency* get_frequency_info(struct cpuInfo* cpu) {
}
}
#ifdef __linux__
if (freq->max == UNKNOWN_DATA || measure_max_frequency_flag()) {
if (freq->max == UNKNOWN_DATA)
printWarn("All previous methods failed, measuring CPU frequency");
freq->max = measure_max_frequency(cpu->first_core_id);
freq->measured = true;
}
#endif
freq->max_pp = UNKNOWN_DATA;
return freq;
}
@@ -964,24 +1048,33 @@ char* get_str_topology(struct cpuInfo* cpu, struct topology* topo, bool dual_soc
string = emalloc(sizeof(char) * (strlen(STRING_UNKNOWN) + 1));
strcpy(string, STRING_UNKNOWN);
}
else if(topo->smt_supported > 1) {
else {
char cores_str[6];
memset(cores_str, 0, sizeof(char) * 6);
if (topo->physical_cores * topo_sockets > 1)
strcpy(cores_str, "cores");
else
strcpy(cores_str, "core");
if(topo->smt_supported > 1) {
// 4 for digits, 21 for ' cores (SMT disabled)' which is the longest possible output
uint32_t max_size = 4+21+1;
string = emalloc(sizeof(char) * max_size);
if(topo->smt_available > 1)
snprintf(string, max_size, "%d cores (%d threads)", topo->physical_cores * topo_sockets, topo->logical_cores * topo_sockets);
snprintf(string, max_size, "%d %s (%d threads)", topo->physical_cores * topo_sockets, cores_str, topo->logical_cores * topo_sockets);
else {
if(cpu->cpu_vendor == CPU_VENDOR_AMD)
snprintf(string, max_size, "%d cores (SMT disabled)", topo->physical_cores * topo_sockets);
snprintf(string, max_size, "%d %s (SMT disabled)", topo->physical_cores * topo_sockets, cores_str);
else
snprintf(string, max_size, "%d cores (HT disabled)", topo->physical_cores * topo_sockets);
snprintf(string, max_size, "%d %s (HT disabled)", topo->physical_cores * topo_sockets, cores_str);
}
}
else {
uint32_t max_size = 4+7+1;
string = emalloc(sizeof(char) * max_size);
snprintf(string, max_size, "%d cores",topo->physical_cores * topo_sockets);
snprintf(string, max_size, "%d %s",topo->physical_cores * topo_sockets, cores_str);
}
}
return string;
@@ -1042,8 +1135,14 @@ char* get_str_sse(struct cpuInfo* cpu) {
last+=SSE4_2_sl;
}
if (last == 0) {
snprintf(string, 2+1, "No");
}
else {
//Purge last comma
string[last-1] = '\0';
}
return string;
}
@@ -1084,12 +1183,29 @@ void print_debug(struct cpuInfo* cpu) {
free_cpuinfo_struct(cpu);
}
// TODO: Query HV and Xeon Phi levels
void print_raw_level(uint32_t reg) {
uint32_t eax = reg;
uint32_t ebx = 0;
uint32_t ecx = 0;
uint32_t edx = 0;
cpuid(&eax, &ebx, &ecx, &edx);
printf(" 0x%.8X 0x%.2X: 0x%.8X 0x%.8X 0x%.8X 0x%.8X\n", reg, 0x00, eax, ebx, ecx, edx);
}
void print_raw_sublevel(uint32_t reg, uint32_t reg2) {
uint32_t eax = reg;
uint32_t ebx = 0;
uint32_t ecx = reg2;
uint32_t edx = 0;
cpuid(&eax, &ebx, &ecx, &edx);
printf(" 0x%.8X 0x%.2X: 0x%.8X 0x%.8X 0x%.8X 0x%.8X\n", reg, reg2, eax, ebx, ecx, edx);
}
void print_raw(struct cpuInfo* cpu) {
uint32_t eax;
uint32_t ebx;
uint32_t ecx;
uint32_t edx;
printf("%s\n\n", cpu->cpu_name);
printf(" CPUID leaf sub EAX EBX ECX EDX \n");
printf("--------------------------------------------------------------\n");
@@ -1104,66 +1220,40 @@ void print_raw(struct cpuInfo* cpu) {
printf("CPU %d:\n", c);
// Standard levels
for(uint32_t reg=0x00000000; reg <= cpu->maxLevels; reg++) {
if(reg == 0x00000004) {
for(uint32_t reg2=0x00000000; reg2 < cpu->cach->max_cache_level; reg2++) {
eax = reg;
ebx = 0;
ecx = reg2;
edx = 0;
cpuid(&eax, &ebx, &ecx, &edx);
printf(" 0x%.8X 0x%.2X: 0x%.8X 0x%.8X 0x%.8X 0x%.8X\n", reg, reg2, eax, ebx, ecx, edx);
print_raw_sublevel(reg, reg2);
}
}
else if(reg == 0x0000000B) {
for(uint32_t reg2=0x00000000; reg2 < cpu->topo->smt_supported; reg2++) {
eax = reg;
ebx = 0;
ecx = reg2;
edx = 0;
cpuid(&eax, &ebx, &ecx, &edx);
printf(" 0x%.8X 0x%.2X: 0x%.8X 0x%.8X 0x%.8X 0x%.8X\n", reg, reg2, eax, ebx, ecx, edx);
print_raw_sublevel(reg, reg2);
}
}
else {
eax = reg;
ebx = 0;
ecx = 0;
edx = 0;
cpuid(&eax, &ebx, &ecx, &edx);
printf(" 0x%.8X 0x%.2X: 0x%.8X 0x%.8X 0x%.8X 0x%.8X\n", reg, 0x00, eax, ebx, ecx, edx);
print_raw_level(reg);
}
}
// Hypervisor levels
for(uint32_t reg=0x40000000; reg <= 0x40000006; reg++) {
print_raw_level(reg);
}
// Extended levels
for(uint32_t reg=0x80000000; reg <= cpu->maxExtendedLevels; reg++) {
if(reg == 0x8000001D) {
for(uint32_t reg2=0x00000000; reg2 < cpu->cach->max_cache_level; reg2++) {
eax = reg;
ebx = 0;
ecx = reg2;
edx = 0;
cpuid(&eax, &ebx, &ecx, &edx);
printf(" 0x%.8X 0x%.2X: 0x%.8X 0x%.8X 0x%.8X 0x%.8X\n", reg, reg2, eax, ebx, ecx, edx);
print_raw_sublevel(reg, reg2);
}
}
else {
eax = reg;
ebx = 0;
ecx = 0;
edx = 0;
cpuid(&eax, &ebx, &ecx, &edx);
printf(" 0x%.8X 0x%.2X: 0x%.8X 0x%.8X 0x%.8X 0x%.8X\n", reg, 0x00, eax, ebx, ecx, edx);
print_raw_level(reg);
}
}
}
}

View File

@@ -21,9 +21,12 @@
#define FREQ_VECTOR_SIZE 1<<16
struct freq_thread {
// Inputs
struct cpuInfo* cpu;
bool end;
bool measure;
double freq;
// Output
int32_t *max_pp;
};
double vector_average_harmonic(double* v, int len) {
@@ -48,6 +51,7 @@ void* measure_freq(void *freq_ptr) {
char* line = NULL;
size_t len = 0;
ssize_t read;
struct cpuInfo* cpu = freq->cpu;
int v = 0;
double* freq_vector = malloc(sizeof(double) * FREQ_VECTOR_SIZE);
@@ -76,18 +80,43 @@ void* measure_freq(void *freq_ptr) {
sleep_ms(500);
}
freq->freq = vector_average_harmonic(freq_vector, v);
printWarn("AVX2 measured freq=%f\n", freq->freq);
if (cpu->hybrid_flag) {
// We have an heterogeneous architecture. After measuring the
// frequency for all cores, we now need to compute the average
// independently for each CPU module.
struct cpuInfo* ptr = cpu;
double* freq_vector_ptr = freq_vector;
for (int i=0; i < cpu->num_cpus; ptr = ptr->next_cpu, i++) {
freq->max_pp[i] = vector_average_harmonic(freq_vector_ptr, ptr->topo->total_cores_module);
printWarn("AVX2 measured freq=%d (module %d)", freq->max_pp[i], i);
freq_vector_ptr = freq_vector_ptr + ptr->topo->total_cores_module;
}
}
else {
freq->max_pp[0] = vector_average_harmonic(freq_vector, v);
printWarn("AVX2 measured freq=%d\n", freq->max_pp[0]);
}
return NULL;
}
int64_t measure_frequency(struct cpuInfo* cpu) {
int32_t measure_frequency(struct cpuInfo* cpu, int32_t *max_freq_pp_vec) {
if (cpu->hybrid_flag && cpu->module_id > 0) {
// We have a hybrid architecture and we have already
// measured the frequency for this module in a previous
// call to this function, so now just return it.
return max_freq_pp_vec[cpu->module_id];
}
int ret;
int num_spaces;
struct freq_thread* freq_struct = malloc(sizeof(struct freq_thread));
freq_struct->end = false;
freq_struct->measure = false;
freq_struct->cpu = cpu;
freq_struct->max_pp = max_freq_pp_vec;
void* (*compute_function)(void*);
@@ -116,8 +145,25 @@ int64_t measure_frequency(struct cpuInfo* cpu) {
}
pthread_t* compute_th = malloc(sizeof(pthread_t) * cpu->topo->total_cores);
cpu_set_t cpus;
pthread_attr_t attr;
if ((ret = pthread_attr_init(&attr)) != 0) {
printErr("pthread_attr_init: %s", strerror(ret));
return -1;
}
for(int i=0; i < cpu->topo->total_cores; i++) {
ret = pthread_create(&compute_th[i], NULL, compute_function, NULL);
// We might have called bind_to_cpu previously, binding the threads
// to a specific core, so now we must make sure we run the new thread
// on the correct core.
CPU_ZERO(&cpus);
CPU_SET(i, &cpus);
if ((ret = pthread_attr_setaffinity_np(&attr, sizeof(cpu_set_t), &cpus)) != 0) {
printErr("pthread_attr_setaffinity_np: %s", strerror(ret));
return -1;
}
ret = pthread_create(&compute_th[i], &attr, compute_function, NULL);
if(ret != 0) {
fprintf(stderr, "Error creating thread\n");
@@ -142,5 +188,5 @@ int64_t measure_frequency(struct cpuInfo* cpu) {
}
printf("\r%*c", num_spaces, ' ');
return freq_struct->freq;
return max_freq_pp_vec[0];
}

View File

@@ -8,6 +8,6 @@
#define MEASURE_TIME_SECONDS 5
#define LOOP_ITERS 100000000
int64_t measure_frequency(struct cpuInfo* cpu);
int32_t measure_frequency(struct cpuInfo* cpu, int32_t *max_freq_pp_vec);
#endif

View File

@@ -47,8 +47,12 @@ typedef uint32_t MICROARCH;
enum {
UARCH_UNKNOWN,
// INTEL //
UARCH_I486,
UARCH_P5,
UARCH_P6,
UARCH_P5_MMX,
UARCH_P6_PRO,
UARCH_P6_PENTIUM_II,
UARCH_P6_PENTIUM_III,
UARCH_DOTHAN,
UARCH_YONAH,
UARCH_MEROM,
@@ -89,12 +93,15 @@ enum {
UARCH_CEDAR_MILL,
UARCH_ITANIUM2,
UARCH_ICE_LAKE,
UARCH_SAPPHIRE_RAPIDS,
UARCH_TIGER_LAKE,
UARCH_ALDER_LAKE,
UARCH_RAPTOR_LAKE,
// AMD //
UARCH_AM486,
UARCH_AM5X86,
UARCH_SSA5,
UARCH_K5,
UARCH_K6,
UARCH_K7,
UARCH_K8,
@@ -112,7 +119,10 @@ enum {
UARCH_ZEN2,
UARCH_ZEN3,
UARCH_ZEN3_PLUS,
UARCH_ZEN4
UARCH_ZEN4,
UARCH_ZEN4C,
UARCH_ZEN5,
UARCH_ZEN5C,
};
struct uarch {
@@ -124,7 +134,8 @@ struct uarch {
#define UARCH_START if (false) {}
#define CHECK_UARCH(arch, ef_, f_, em_, m_, s_, str, uarch, process) \
else if (ef_ == ef && f_ == f && (em_ == NA || em_ == em) && (m_ == NA || m_ == m) && (s_ == NA || s_ == s)) fill_uarch(arch, str, uarch, process);
#define UARCH_END else { printBug("Unknown microarchitecture detected: M=0x%.8X EM=0x%.8X F=0x%.8X EF=0x%.8X S=0x%.8X", m, em, f, ef, s); fill_uarch(arch, STRING_UNKNOWN, UARCH_UNKNOWN, 0); }
#define UARCH_END else { printBugCheckRelease("Unknown microarchitecture detected: M=0x%X EM=0x%X F=0x%X EF=0x%X S=0x%X", m, em, f, ef, s); \
fill_uarch(arch, STRING_UNKNOWN, UARCH_UNKNOWN, UNK); }
void fill_uarch(struct uarch* arch, char* str, MICROARCH u, uint32_t process) {
arch->uarch_str = emalloc(sizeof(char) * (strlen(str)+1));
@@ -142,31 +153,45 @@ struct uarch* get_uarch_from_cpuid_intel(uint32_t ef, uint32_t f, uint32_t em, u
// EM: Extended Model //
// M: Model //
// S: Stepping //
// ----------------------------------------------------------------------------- //
// ------------------------------------------------------------------------------- //
// EF F EM M S //
UARCH_START
CHECK_UARCH(arch, 0, 4, 0, 0, NA, "i80486DX", UARCH_I486, UNK) // https://sandpile.org/x86/cpuid.htm#level_0000_0001h
CHECK_UARCH(arch, 0, 4, 0, 1, NA, "i80486DX-50", UARCH_I486, UNK) // https://sandpile.org/x86/cpuid.htm#level_0000_0001h
CHECK_UARCH(arch, 0, 4, 0, 2, NA, "i80486SX", UARCH_I486, UNK) // https://sandpile.org/x86/cpuid.htm#level_0000_0001h
CHECK_UARCH(arch, 0, 4, 0, 3, NA, "i80486DX2", UARCH_I486, UNK) // https://sandpile.org/x86/cpuid.htm#level_0000_0001h
CHECK_UARCH(arch, 0, 4, 0, 4, NA, "i80486SL", UARCH_I486, UNK) // https://sandpile.org/x86/cpuid.htm#level_0000_0001h
CHECK_UARCH(arch, 0, 4, 0, 5, NA, "i80486SX2", UARCH_I486, UNK) // https://sandpile.org/x86/cpuid.htm#level_0000_0001h
CHECK_UARCH(arch, 0, 4, 0, 7, NA, "i80486DX2WB", UARCH_I486, UNK) // https://sandpile.org/x86/cpuid.htm#level_0000_0001h
CHECK_UARCH(arch, 0, 4, 0, 8, NA, "i80486DX4", UARCH_I486, UNK) // https://sandpile.org/x86/cpuid.htm#level_0000_0001h
CHECK_UARCH(arch, 0, 4, 0, 9, NA, "i80486DX4WB", UARCH_I486, UNK) // https://sandpile.org/x86/cpuid.htm#level_0000_0001h
CHECK_UARCH(arch, 0, 5, 0, 0, NA, "P5", UARCH_P5, 800)
CHECK_UARCH(arch, 0, 5, 0, 1, NA, "P5", UARCH_P5, 800)
CHECK_UARCH(arch, 0, 5, 0, 2, NA, "P5", UARCH_P5, UNK)
CHECK_UARCH(arch, 0, 5, 0, 3, NA, "P5", UARCH_P5, 600)
CHECK_UARCH(arch, 0, 5, 0, 4, NA, "P5 MMX", UARCH_P5, UNK)
CHECK_UARCH(arch, 0, 5, 0, 7, NA, "P5 MMX", UARCH_P5, UNK)
CHECK_UARCH(arch, 0, 5, 0, 8, NA, "P5 MMX", UARCH_P5, 250)
CHECK_UARCH(arch, 0, 5, 0, 2, NA, "P54C", UARCH_P5, UNK) // https://sandpile.org/x86/cpuid.htm#level_0000_0001h
CHECK_UARCH(arch, 0, 5, 0, 3, NA, "P24T (Overdrive)", UARCH_P5, 600) // https://sandpile.org/x86/cpuid.htm#level_0000_0001h
CHECK_UARCH(arch, 0, 5, 0, 4, NA, "P55C (MMX)", UARCH_P5_MMX, 350) // https://www.cpu-world.com/CPUs/Pentium/TYPE-Pentium%20MMX.html
CHECK_UARCH(arch, 0, 5, 0, 7, NA, "P54C", UARCH_P5, 350) // https://sandpile.org/x86/cpuid.htm#level_0000_0001h
CHECK_UARCH(arch, 0, 5, 0, 8, NA, "Tillamook", UARCH_P5_MMX, 250) // http://instlatx64.atw.hu./
CHECK_UARCH(arch, 0, 5, 0, 9, 0, "Lakemont", UARCH_LAKEMONT, 32)
CHECK_UARCH(arch, 0, 5, 0, 9, NA, "P5 MMX", UARCH_P5, UNK)
CHECK_UARCH(arch, 0, 5, 0, 9, NA, "P5 (MMX)", UARCH_P5_MMX, UNK)
CHECK_UARCH(arch, 0, 5, 0, 10, 0, "Lakemont", UARCH_LAKEMONT, 32)
CHECK_UARCH(arch, 0, 6, 0, 0, NA, "P6 Pentium II", UARCH_P6, UNK)
CHECK_UARCH(arch, 0, 6, 0, 1, NA, "P6 Pentium II", UARCH_P6, UNK) // process depends on core
CHECK_UARCH(arch, 0, 6, 0, 2, NA, "P6 Pentium II", UARCH_P6, UNK)
CHECK_UARCH(arch, 0, 6, 0, 3, NA, "P6 Pentium II", UARCH_P6, 350)
CHECK_UARCH(arch, 0, 6, 0, 4, NA, "P6 Pentium II", UARCH_P6, UNK)
CHECK_UARCH(arch, 0, 6, 0, 5, NA, "P6 Pentium II", UARCH_P6, 250)
CHECK_UARCH(arch, 0, 6, 0, 6, NA, "P6 Pentium II", UARCH_P6, UNK)
CHECK_UARCH(arch, 0, 6, 0, 7, NA, "P6 Pentium III", UARCH_P6, 250)
CHECK_UARCH(arch, 0, 6, 0, 8, NA, "P6 Pentium III", UARCH_P6, 180)
CHECK_UARCH(arch, 0, 6, 0, 9, NA, "P6 Pentium M", UARCH_P6, 130)
CHECK_UARCH(arch, 0, 6, 0, 10, NA, "P6 Pentium III", UARCH_P6, 180)
CHECK_UARCH(arch, 0, 6, 0, 11, NA, "P6 Pentium III", UARCH_P6, 130)
CHECK_UARCH(arch, 0, 6, 0, 1, 1, "P6", UARCH_P6_PRO, UNK)
CHECK_UARCH(arch, 0, 6, 0, 1, 2, "P6", UARCH_P6_PRO, UNK)
CHECK_UARCH(arch, 0, 6, 0, 1, 6, "P6", UARCH_P6_PRO, 350)
CHECK_UARCH(arch, 0, 6, 0, 1, 7, "P6", UARCH_P6_PRO, 350)
CHECK_UARCH(arch, 0, 6, 0, 1, 9, "P6", UARCH_P6_PRO, 350)
CHECK_UARCH(arch, 0, 6, 0, 0, NA, "P6 (Pentium II)", UARCH_P6_PENTIUM_II, UNK)
CHECK_UARCH(arch, 0, 6, 0, 1, NA, "P6 (Pentium II)", UARCH_P6_PENTIUM_II, UNK) // process depends on core
CHECK_UARCH(arch, 0, 6, 0, 2, NA, "P6 (Pentium II)", UARCH_P6_PENTIUM_II, UNK)
CHECK_UARCH(arch, 0, 6, 0, 3, NA, "P6 (Klamath)", UARCH_P6_PENTIUM_II, 350) // http://instlatx64.atw.hu.
CHECK_UARCH(arch, 0, 6, 0, 4, NA, "P6 (Pentium II)", UARCH_P6_PENTIUM_II, UNK)
CHECK_UARCH(arch, 0, 6, 0, 5, NA, "P6 (Deschutes)", UARCH_P6_PENTIUM_II, 250) // http://instlatx64.atw.hu.
CHECK_UARCH(arch, 0, 6, 0, 6, NA, "P6 (Dixon)", UARCH_P6_PENTIUM_II, UNK) // http://instlatx64.atw.hu.
CHECK_UARCH(arch, 0, 6, 0, 7, NA, "P6 (Katmai)", UARCH_P6_PENTIUM_III, 250) // Core names from: https://en.wikichip.org/wiki/intel/cpuid. NOTE: Xeon core names are different! https://www.techpowerup.com/cpu-specs/?generation=Intel+Pentium+III+Xeon
CHECK_UARCH(arch, 0, 6, 0, 8, NA, "P6 (Coppermine)", UARCH_P6_PENTIUM_III, 180) // Also: https://en.wikipedia.org/wiki/Pentium_III
CHECK_UARCH(arch, 0, 6, 0, 9, NA, "P6 (Pentium M)", UARCH_P6_PENTIUM_III, 130)
CHECK_UARCH(arch, 0, 6, 0, 10, NA, "P6 (Coppermine T)", UARCH_P6_PENTIUM_III, 180)
CHECK_UARCH(arch, 0, 6, 0, 11, NA, "P6 (Tualatin)", UARCH_P6_PENTIUM_III, 130)
CHECK_UARCH(arch, 0, 6, 0, 13, NA, "Dothan", UARCH_DOTHAN, UNK) // process depends on core
CHECK_UARCH(arch, 0, 6, 0, 14, NA, "Yonah", UARCH_YONAH, 65)
CHECK_UARCH(arch, 0, 6, 0, 15, NA, "Merom", UARCH_MEROM, 65)
@@ -231,6 +256,7 @@ struct uarch* get_uarch_from_cpuid_intel(uint32_t ef, uint32_t f, uint32_t em, u
// CHECK_UARCH(arch, 0, 6, 8, 14, 10, ...) It is not possible to determine uarch only from CPUID dump (can be Kaby Lake R or Coffee Lake U)
CHECK_UARCH(arch, 0, 6, 8, 14, 11, "Whiskey Lake", UARCH_WHISKEY_LAKE, 14) // wikichip
CHECK_UARCH(arch, 0, 6, 8, 14, 12, "Comet Lake", UARCH_COMET_LAKE, 14) // wikichip
CHECK_UARCH(arch, 0, 6, 8, 15, 8, "Sapphire Rapids", UARCH_SAPPHIRE_RAPIDS, 7) // wikichip
CHECK_UARCH(arch, 0, 6, 9, 6, NA, "Tremont", UARCH_TREMONT, 10) // LX*
CHECK_UARCH(arch, 0, 6, 9, 7, NA, "Alder Lake", UARCH_ALDER_LAKE, 10) // instlatx64 (Alder Lake-S)
CHECK_UARCH(arch, 0, 6, 9, 10, NA, "Alder Lake", UARCH_ALDER_LAKE, 10) // instlatx64 (Alder Lake-P)
@@ -245,6 +271,9 @@ struct uarch* get_uarch_from_cpuid_intel(uint32_t ef, uint32_t f, uint32_t em, u
CHECK_UARCH(arch, 0, 6, 10, 6, NA, "Comet Lake", UARCH_COMET_LAKE, 14) // instlatx64.atw.hu (i7-10710U)
CHECK_UARCH(arch, 0, 6, 10, 7, NA, "Rocket Lake", UARCH_ROCKET_LAKE, 14) // instlatx64.atw.hu (i7-11700K)
CHECK_UARCH(arch, 0, 6, 11, 7, NA, "Raptor Lake", UARCH_RAPTOR_LAKE, 10) // instlatx64.atw.hu (i5-13600K)
CHECK_UARCH(arch, 0, 6, 11, 10, NA, "Raptor Lake", UARCH_RAPTOR_LAKE, 10) // instlatx64.atw.hu (i7-1370P)
CHECK_UARCH(arch, 0, 6, 11, 14, NA, "Alder Lake", UARCH_ALDER_LAKE, 10) // instlatx64.atw.hu (Alder Lake-N)
CHECK_UARCH(arch, 0, 6, 11, 15, NA, "Raptor Lake", UARCH_RAPTOR_LAKE, 10) // instlatx64.atw.hu (i5-13500)
CHECK_UARCH(arch, 0, 11, 0, 0, NA, "Knights Ferry", UARCH_KNIGHTS_FERRY, 45) // found only on en.wikichip.org
CHECK_UARCH(arch, 0, 11, 0, 1, NA, "Knights Corner", UARCH_KNIGHTS_CORNER, 22)
CHECK_UARCH(arch, 0, 15, 0, 0, NA, "Willamette", UARCH_WILLAMETTE, 180)
@@ -257,7 +286,6 @@ struct uarch* get_uarch_from_cpuid_intel(uint32_t ef, uint32_t f, uint32_t em, u
CHECK_UARCH(arch, 1, 15, 0, 1, NA, "Itanium2", UARCH_ITANIUM2, 130)
CHECK_UARCH(arch, 1, 15, 0, 2, NA, "Itanium2", UARCH_ITANIUM2, 130)
UARCH_END
return arch;
}
@@ -273,11 +301,16 @@ struct uarch* get_uarch_from_cpuid_amd(uint32_t ef, uint32_t f, uint32_t em, uin
// ----------------------------------------------------------------------------- //
// EF F EM M S //
UARCH_START
CHECK_UARCH(arch, 0, 4, 0, 3, NA, "Am486", UARCH_AM486, UNK)
CHECK_UARCH(arch, 0, 4, 0, 7, NA, "Am486", UARCH_AM486, UNK)
CHECK_UARCH(arch, 0, 4, 0, 8, NA, "Am486", UARCH_AM486, UNK)
CHECK_UARCH(arch, 0, 4, 0, 9, NA, "Am486", UARCH_AM486, UNK)
CHECK_UARCH(arch, 0, 4, NA, NA, NA, "Am5x86", UARCH_AM5X86, UNK)
CHECK_UARCH(arch, 0, 4, 0, 3, NA, "Am486DX2", UARCH_AM486, UNK) // https://sandpile.org/x86/cpuid.htm#level_0000_0001h
CHECK_UARCH(arch, 0, 4, 0, 7, NA, "Am486DX2WB", UARCH_AM486, UNK) // https://sandpile.org/x86/cpuid.htm#level_0000_0001h
CHECK_UARCH(arch, 0, 4, 0, 8, NA, "Am486DX4", UARCH_AM486, UNK) // https://sandpile.org/x86/cpuid.htm#level_0000_0001h
CHECK_UARCH(arch, 0, 4, 0, 9, NA, "Am486DX4WB", UARCH_AM486, UNK) // https://sandpile.org/x86/cpuid.htm#level_0000_0001h
CHECK_UARCH(arch, 0, 4, 0, 14, NA, "Am5x86", UARCH_AM5X86, 350) // https://sandpile.org/x86/cpuid.htm#level_0000_0001h
CHECK_UARCH(arch, 0, 4, 0, 15, NA, "Am5x86WB", UARCH_AM5X86, 350) // https://sandpile.org/x86/cpuid.htm#level_0000_0001h
CHECK_UARCH(arch, 0, 5, 0, 0, NA, "SSA5 (K5)", UARCH_SSA5, 350) // https://sandpile.org/x86/cpuid.htm#level_0000_0001h
CHECK_UARCH(arch, 0, 5, 0, 1, NA, "K5", UARCH_K5, 350) // https://sandpile.org/x86/cpuid.htm#level_0000_0001h
CHECK_UARCH(arch, 0, 5, 0, 2, NA, "K5", UARCH_K5, 350) // https://sandpile.org/x86/cpuid.htm#level_0000_0001h
CHECK_UARCH(arch, 0, 5, 0, 3, NA, "K5", UARCH_K5, 350) // https://sandpile.org/x86/cpuid.htm#level_0000_0001h
CHECK_UARCH(arch, 0, 5, 0, 6, NA, "K6", UARCH_K6, 300)
CHECK_UARCH(arch, 0, 5, 0, 7, NA, "K6", UARCH_K6, 250) // *p from sandpile.org
CHECK_UARCH(arch, 0, 5, 0, 10, NA, "K7", UARCH_K7, 130) // Geode NX
@@ -348,6 +381,7 @@ struct uarch* get_uarch_from_cpuid_amd(uint32_t ef, uint32_t f, uint32_t em, uin
CHECK_UARCH(arch, 6, 15, 6, 5, NA, "Excavator", UARCH_EXCAVATOR, 28) // undocumented, but sample from Alexandros Couloumbis
CHECK_UARCH(arch, 6, 15, 7, 0, NA, "Excavator", UARCH_EXCAVATOR, 28)
CHECK_UARCH(arch, 7, 15, 0, 0, NA, "Jaguar", UARCH_JAGUAR, 28)
CHECK_UARCH(arch, 7, 15, 1, NA, NA, "Jaguar", UARCH_JAGUAR, 14) // instlatx64 (PS4) Normal PS4 is 28nm, Slim and Pro are 16nm
CHECK_UARCH(arch, 7, 15, 2, 6, NA, "Jaguar", UARCH_JAGUAR, 28) // AMD Cato (Xbox One?)
CHECK_UARCH(arch, 7, 15, 3, 0, NA, "Puma 2014", UARCH_PUMA_2014, 28)
CHECK_UARCH(arch, 8, 15, 0, 0, NA, "Zen", UARCH_ZEN, 14) // instlatx64 engr sample
@@ -362,12 +396,49 @@ struct uarch* get_uarch_from_cpuid_amd(uint32_t ef, uint32_t f, uint32_t em, uin
CHECK_UARCH(arch, 8, 15, 6, 0, NA, "Zen 2", UARCH_ZEN2, 7) // undocumented, geekbench.com example
CHECK_UARCH(arch, 8, 15, 6, 8, NA, "Zen 2", UARCH_ZEN2, 7) // found on instlatx64
CHECK_UARCH(arch, 8, 15, 7, 1, NA, "Zen 2", UARCH_ZEN2, 7) // samples from Steven Noonan and instlatx64
CHECK_UARCH(arch, 8, 15, 8, 4, NA, "Zen 2", UARCH_ZEN2, 7) // instlatx64 (Xbox Series X?)
CHECK_UARCH(arch, 8, 15, 9, 0, 2, "Zen 2", UARCH_ZEN2, 7) // Steam Deck (instlatx64)
CHECK_UARCH(arch, 8, 15, 10, 0, NA, "Zen 2", UARCH_ZEN2, 6) // instlatx64
CHECK_UARCH(arch, 10, 15, 0, 1, NA, "Zen 3", UARCH_ZEN3, 7) // instlatx64
CHECK_UARCH(arch, 10, 15, 0, 8, NA, "Zen 3", UARCH_ZEN3, 7) // instlatx64
CHECK_UARCH(arch, 10, 15, 1, 1, NA, "Zen 4", UARCH_ZEN4, 5) // instlatx64
CHECK_UARCH(arch, 10, 15, 1, 8, NA, "Zen 4", UARCH_ZEN4, 5) // instlatx64
CHECK_UARCH(arch, 10, 15, 2, 1, NA, "Zen 3", UARCH_ZEN3, 7) // instlatx64
CHECK_UARCH(arch, 10, 15, 3, NA, NA, "Zen 3", UARCH_ZEN3, 7) // instlatx64
CHECK_UARCH(arch, 10, 15, 4, 4, NA, "Zen 3+", UARCH_ZEN3_PLUS, 6) // instlatx64 (they say it is Zen3...)
CHECK_UARCH(arch, 10, 15, 5, 0, NA, "Zen 3", UARCH_ZEN3, 7) // instlatx64
CHECK_UARCH(arch, 10, 15, 6, 1, 2, "Zen 4", UARCH_ZEN4, 5) // instlatx64
CHECK_UARCH(arch, 10, 15, 7, 4, 1, "Zen 4", UARCH_ZEN4, 4) // instlatx64
CHECK_UARCH(arch, 10, 15, 7, 5, 2, "Zen 4", UARCH_ZEN4, 4) // instlatx64
CHECK_UARCH(arch, 10, 15, 7, 8, 0, "Zen 4", UARCH_ZEN4, 4) // instlatx64
CHECK_UARCH(arch, 10, 15, 8, NA, NA, "Zen 4", UARCH_ZEN4, 5) // instlatx64 (AMD MI300C)
CHECK_UARCH(arch, 10, 15, 9, NA, NA, "Zen 4", UARCH_ZEN4, 5) // instlatx64 (AMD MI300A)
CHECK_UARCH(arch, 10, 15, 10, NA, NA, "Zen 4c", UARCH_ZEN4C, 5) // instlatx64
CHECK_UARCH(arch, 11, 15, 0, NA, NA, "Zen 5", UARCH_ZEN5, 4) // Turin/EPYC (instlatx64)
CHECK_UARCH(arch, 11, 15, 1, NA, NA, "Zen 5c", UARCH_ZEN5C, 3) // Zen5c EPYC (instlatx64, https://en.wikipedia.org/wiki/Zen_5#cite_note-10)
CHECK_UARCH(arch, 11, 15, 2, NA, NA, "Zen 5", UARCH_ZEN5, 4) // Strix Point (instlatx64)
CHECK_UARCH(arch, 11, 15, 4, NA, NA, "Zen 5", UARCH_ZEN5, 4) // Granite Ridge (instlatx64)
CHECK_UARCH(arch, 11, 15, 6, NA, NA, "Zen 5", UARCH_ZEN5, 4) // Krackan Point (instlatx64)
CHECK_UARCH(arch, 11, 15, 7, NA, NA, "Zen 5", UARCH_ZEN5, 4) // Strix Halo (instlatx64)
UARCH_END
return arch;
}
struct uarch* get_uarch_from_cpuid_hygon(uint32_t ef, uint32_t f, uint32_t em, uint32_t m, int s) {
struct uarch* arch = emalloc(sizeof(struct uarch));
// EF: Extended Family //
// F: Family //
// EM: Extended Model //
// M: Model //
// S: Stepping //
// ----------------------------------------------------------------------------- //
// EF F EM M S //
UARCH_START
// https://www.phoronix.com/news/Hygon-Dhyana-AMD-China-CPUs
CHECK_UARCH(arch, 9, 15, 0, 1, NA, "Zen", UARCH_ZEN, UNK) // https://github.com/Dr-Noob/cpufetch/issues/244
// CHECK_UARCH(arch, 9, 15, 0, 2, NA, "???", ?????????, UNK) // http://instlatx64.atw.hu/
UARCH_END
return arch;
@@ -375,11 +446,16 @@ struct uarch* get_uarch_from_cpuid_amd(uint32_t ef, uint32_t f, uint32_t em, uin
struct uarch* get_uarch_from_cpuid(struct cpuInfo* cpu, uint32_t dump, uint32_t ef, uint32_t f, uint32_t em, uint32_t m, int s) {
if(cpu->cpu_vendor == CPU_VENDOR_INTEL) {
struct uarch* arch = emalloc(sizeof(struct uarch));
if(dump == 0x000806E9) {
if (cpu->cpu_name == NULL) {
printErr("Unable to find uarch without CPU name");
fill_uarch(arch, STRING_UNKNOWN, UARCH_UNKNOWN, UNK);
return arch;
}
// It is not possible to determine uarch only from CPUID dump (can be Kaby Lake or Amber Lake)
// See issue https://github.com/Dr-Noob/cpufetch/issues/122
struct uarch* arch = emalloc(sizeof(struct uarch));
if(strstr(cpu->cpu_name, "Y") != NULL) {
fill_uarch(arch, "Amber Lake", UARCH_AMBER_LAKE, 14);
}
@@ -390,10 +466,14 @@ struct uarch* get_uarch_from_cpuid(struct cpuInfo* cpu, uint32_t dump, uint32_t
return arch;
}
else if (dump == 0x000806EA) {
if (cpu->cpu_name == NULL) {
printErr("Unable to find uarch without CPU name");
fill_uarch(arch, STRING_UNKNOWN, UARCH_UNKNOWN, UNK);
return arch;
}
// It is not possible to determine uarch only from CPUID dump (can be Kaby Lake R or Coffee Lake U)
// See issue https://github.com/Dr-Noob/cpufetch/issues/149
struct uarch* arch = emalloc(sizeof(struct uarch));
if(strstr(cpu->cpu_name, "i5-8250U") != NULL ||
strstr(cpu->cpu_name, "i5-8350U") != NULL ||
strstr(cpu->cpu_name, "i7-8550U") != NULL ||
@@ -408,12 +488,86 @@ struct uarch* get_uarch_from_cpuid(struct cpuInfo* cpu, uint32_t dump, uint32_t
}
return get_uarch_from_cpuid_intel(ef, f, em, m, s);
}
else
else if(cpu->cpu_vendor == CPU_VENDOR_AMD) {
return get_uarch_from_cpuid_amd(ef, f, em, m, s);
}
else if(cpu->cpu_vendor == CPU_VENDOR_HYGON) {
return get_uarch_from_cpuid_hygon(ef, f, em, m, s);
}
else {
printBug("Invalid CPU vendor: %d", cpu->cpu_vendor);
return NULL;
}
}
// If we cannot get the CPU name from CPUID, try to infer it from uarch
char* infer_cpu_name_from_uarch(struct uarch* arch) {
char* cpu_name = NULL;
if (arch == NULL) {
printErr("infer_cpu_name_from_uarch: Unable to find CPU name");
cpu_name = ecalloc(strlen(STRING_UNKNOWN) + 1, sizeof(char));
strcpy(cpu_name, STRING_UNKNOWN);
return cpu_name;
}
char *str = NULL;
switch (arch->uarch) {
// Intel
case UARCH_I486:
str = "Intel 486";
break;
case UARCH_P5:
str = "Intel Pentium";
break;
case UARCH_P5_MMX:
str = "Intel Pentium MMX";
break;
case UARCH_P6_PRO:
str = "Intel Pentium Pro";
break;
case UARCH_P6_PENTIUM_II:
str = "Intel Pentium II";
break;
case UARCH_P6_PENTIUM_III:
str = "Intel Pentium III";
break;
// AMD
case UARCH_AM486:
str = "AMD 486";
break;
case UARCH_AM5X86:
str = "AMD 5x86";
break;
case UARCH_SSA5:
str = "AMD 5k86";
break;
default:
printErr("Unable to find name from uarch: %d", arch->uarch);
break;
}
if (str == NULL) {
cpu_name = ecalloc(strlen(STRING_UNKNOWN) + 1, sizeof(char));
strcpy(cpu_name, STRING_UNKNOWN);
}
else {
cpu_name = ecalloc(strlen(str) + 1, sizeof(char));
strcpy(cpu_name, str);
}
return cpu_name;
}
bool vpus_are_AVX512(struct cpuInfo* cpu) {
return cpu->arch->uarch != UARCH_ICE_LAKE && cpu->arch->uarch != UARCH_TIGER_LAKE;
// Zen5 actually has 2 x AVX512 units
// https://www.anandtech.com/show/21469/amd-details-ryzen-ai-300-series-for-mobile-strix-point-with-rdna-35-igpu-xdna-2-npu
return cpu->arch->uarch != UARCH_ICE_LAKE &&
cpu->arch->uarch != UARCH_TIGER_LAKE &&
cpu->arch->uarch != UARCH_ZEN4 &&
cpu->arch->uarch != UARCH_ZEN4C;
}
bool is_knights_landing(struct cpuInfo* cpu) {
@@ -439,6 +593,7 @@ int get_number_of_vpus(struct cpuInfo* cpu) {
case UARCH_KNIGHTS_LANDING:
case UARCH_KNIGHTS_MILL:
case UARCH_SAPPHIRE_RAPIDS:
case UARCH_ICE_LAKE:
case UARCH_TIGER_LAKE:
case UARCH_ALDER_LAKE:
@@ -449,6 +604,9 @@ int get_number_of_vpus(struct cpuInfo* cpu) {
case UARCH_ZEN3:
case UARCH_ZEN3_PLUS:
case UARCH_ZEN4:
case UARCH_ZEN4C:
case UARCH_ZEN5:
case UARCH_ZEN5C:
return 2;
default:
return 1;
@@ -478,9 +636,6 @@ char* get_str_process(struct cpuInfo* cpu) {
if(process == UNK) {
snprintf(str, strlen(STRING_UNKNOWN)+1, STRING_UNKNOWN);
}
else if(process > 100) {
sprintf(str, "%.2fum", (double)process/100);
}
else if(process > 0){
sprintf(str, "%dnm", process);
}

View File

@@ -8,6 +8,7 @@
struct uarch;
struct uarch* get_uarch_from_cpuid(struct cpuInfo* cpu, uint32_t dump, uint32_t ef, uint32_t f, uint32_t em, uint32_t m, int s);
char* infer_cpu_name_from_uarch(struct uarch* arch);
bool vpus_are_AVX512(struct cpuInfo* cpu);
bool is_knights_landing(struct cpuInfo* cpu);
int get_number_of_vpus(struct cpuInfo* cpu);