target triple overrides mcpu in zig cc

[SuperFluffy]

Zig Version

0.10.1

Steps to Reproduce and Observed Behavior

I fail to cross compile an asm file that you can find here from an x86_64 machine to aarch64-linux-gnu and an nvidia jetson carmel+v8_2a+crypto:

❯ zig cc -O0 -ffunction-sections -fdata-sections -fPIC -g -fno-omit-frame-pointer -Wall -Wextra -mcpu=carmel+v8_2a+crypto --target=aarch64-linux-gnu -c -o sha256_aarch64.so -c src/sha256_aarch64.S
src/sha256_aarch64.S:75:2: error: instruction requires: sha2
sha256su0 v8.4s, v9.4s

I have also tried -mcpu=generic+v8a+crypto and -mcpu=apple_m1+crypto, but none of them work.

Using verbose output I noticed that neither CPU nor features seem to be passed through:

❯ zig cc -v -O0 -ffunction-sections -fdata-sections -fPIC -g -fno-omit-frame-pointer -Wall -Wextra -mcpu=carmel+v8_2a+crypto --target=aarch64-linux-gnu -c -o sha256_aarch64.so -c src/sha256_aarch64.S
clang version 15.0.7
Target: aarch64-unknown-linux-gnu
Thread model: posix
InstalledDir: /usr/bin
 "/usr/bin/zig" -cc1 -triple aarch64-unknown-linux-gnu -E -disable-free -clear-ast-before-backend -disable-llvm-verifier -discard-value-names -main-file-name sha256_aarch64.S -mrelocation-model pic -pic-level 2 -pic-is-pie -mframe-pointer=non-leaf -fmath-errno -ffp-contract=on -fno-rounding-math -mconstructor-aliases -funwind-tables=2 -target-cpu generic -target-feature +neon -target-feature +v8a -target-abi aapcs -fallow-half-arguments-and-returns -debug-info-kind=constructor -dwarf-version=4 -debugger-tuning=gdb -v -ffunction-sections -fdata-sections -fcoverage-compilation-dir=/home/janis/dev/worldcoin/asm-hashes/sha2 -resource-dir /usr/lib/clang/15.0.7 -D __GLIBC_MINOR__=19 -internal-isystem /usr/lib/clang/15.0.7/include -internal-isystem /usr/local/include -internal-externc-isystem /include -internal-externc-isystem /usr/include -Wall -Wextra -fdebug-compilation-dir=/home/janis/dev/worldcoin/asm-hashes/sha2 -ferror-limit 19 -stack-protector 2 -fno-signed-char -fgnuc-version=4.2.1 -fcolor-diagnostics -faddrsig -D__GCC_HAVE_DWARF2_CFI_ASM=1 -o /tmp/sha256_aarch64-b93408.s -x assembler-with-cpp src/sha256_aarch64.S
clang -cc1 version 15.0.7 based upon LLVM 15.0.7 default target x86_64-pc-linux-gnu
ignoring nonexistent directory "/include"
#include "..." search starts here:
#include <...> search starts here:
 /usr/lib/clang/15.0.7/include
 /usr/local/include
 /usr/include
End of search list.
 "/usr/bin/zig" -cc1as -triple aarch64-unknown-linux-gnu -filetype obj -main-file-name sha256_aarch64.S -target-cpu generic -target-feature +neon -target-feature +v8a -fdebug-compilation-dir=/home/janis/dev/worldcoin/asm-hashes/sha2 -dwarf-debug-producer "clang version 15.0.7" -debug-info-kind=constructor -dwarf-version=4 -mrelocation-model pic --mrelax-relocations -object-file-name=/home/janis/.cache/zig/tmp/5ea73be61cd6ff63-sha256_aarch64.o -o /home/janis/.cache/zig/tmp/5ea73be61cd6ff63-sha256_aarch64.o /tmp/sha256_aarch64-b93408.s
src/sha256_aarch64.S:75:2: error: instruction requires: sha2
 sha256su0 v8.4s, v9.4s
 ^

Note -target-cpu generic -target-feature +neon -target-feature +v8a and the absence of my -mcpu settings. Changing the last command adding my desired features compiles without errors:

❯ "/usr/bin/zig" -cc1as -triple aarch64-unknown-linux-gnu -filetype obj -main-file-name sha256_aarch64.S -target-cpu carmel -target-feature +neon -target-feature +v8a -target-feature crypto -fdebug-compilation-dir
=/home/janis/dev/worldcoin/asm-hashes/sha2 -dwarf-debug-producer "clang version 15.0.7" -debug-info-kind=constructor -dwarf-version=4 -mrelocation-model pic --mrelax-relocations -object-file-name=/home/janis/.cach
e/zig/tmp/4c1793aa0f782f3d-sha256_aarch64.o -o /home/janis/.cache/zig/tmp/4c1793aa0f782f3d-sha256_aarch64.o /tmp/sha256_aarch64-fb28b5.s

(side note: it looks like -target-feature +v8_2a is not a valid target feature? If I set it I get '+v8_2a' is not a recognized feature for this target (ignoring feature))

Expected Behavior

I expect zig to respect -mcpu and not override it using the defaults(?) of the target triple. Or noisily complain that it will do that

It looks like this was previously addressed in these issues: #9196 #4912

[mikdusan]

duplicate of #10411 and #10411 (comment) has a good explanation