Issues

Compile time: what is left for a 2x faster kernel build than gcc-16

tamnd/rucc#4024 · opened

on Nov 1, 2026

1

inline: the unit is measured before small memcpy copies fold, so hot hinted calls in lib/maple_tree.c hit the unit growth limit

tamnd/rucc#4016 · opened

on Nov 1, 2026

aarch64: an integer asm operand with the w constraint is refused

tamnd/rucc#3866 · opened

on Oct 25, 2026

Read gcc's arm_neon.h: fp16, the aarch64 pragma and the builtin_aarch64_ builtins

tamnd/rucc#3861 · opened

on Oct 25, 2026

Put in the KASAN checks for -fsanitize=kernel-address

tamnd/rucc#3854 · opened

on Oct 25, 2026

6

What Linux 7.3 riscv64 defconfig needs from a riscv64 back end

tamnd/rucc#3853 · opened

on Oct 25, 2026

Read the function target attribute for NEON on aarch64

tamnd/rucc#3790 · opened

on Oct 21, 2026

Pass short vectors in v registers on aarch64 as AAPCS64 says

tamnd/rucc#3789 · opened

on Oct 21, 2026

Keep an asm that writes a global register variable, as gcc does with ASM_CALL_CONSTRAINT

tamnd/rucc#3764 · opened

on Oct 21, 2026

Compile Linux 7.3 defconfig in half the CPU time of GCC 16, now 1.78x to 1.87x

tamnd/rucc#3763 · opened

on Oct 21, 2026

__builtin_constant_p on a const local is 0 in an enumerator and 1 elsewhere

tamnd/rucc#3734 · opened

on Oct 19, 2026

Write a memcpy inline when its length is not constant but known to be small, as gcc does for detoast_attr

tamnd/rucc#3705 · opened

on Oct 16, 2026

1

32 byte vectors are passed in memory where gcc uses ymm under -mavx

tamnd/rucc#3682 · opened

on Oct 15, 2026

glibc stubs: loongarch64 gets libutil and libanl that the real system does not have

tamnd/rucc#3677 · opened

on Oct 15, 2026

short-circuit says it kept the loop forest after a collapse in a loop latch, and it did not

tamnd/rucc#3481 · opened

on Oct 9, 2026

aarch64: __builtin_setjmp stops with an internal error on a64.lea_64

tamnd/rucc#3368 · opened

on Oct 8, 2026

Decide on the stack protector canary after dead locals are removed, as gcc does

tamnd/rucc#3361 · opened

on Oct 7, 2026

Write a long memcpy or memset with a known length inline, and turn a fill loop into memset, as gcc does

tamnd/rucc#3360 · opened

on Oct 7, 2026

1

Fold a fortify check that the value range of the length proves cannot fail

tamnd/rucc#3355 · opened

on Oct 7, 2026

1

L9: RISC-V 64 Linux

tamnd/rucc#3284 · opened

on Oct 7, 2026

· L9: RISC-V 64 Linux

8

L8: Distribution rebuild and packages

tamnd/rucc#3283 · opened

on Oct 7, 2026

· L8: Distribution rebuild and packages

L7: AArch64, musl and i686 rows

tamnd/rucc#3282 · opened

on Oct 7, 2026

· L7: AArch64, musl and i686 rows

1

L6: Headers and the C library

tamnd/rucc#3281 · opened

on Oct 7, 2026

· L6: Headers and the C library

23

L5: Debuggers, profilers and unwinders

tamnd/rucc#3280 · opened

on Oct 7, 2026

· L5: Debuggers, profilers and unwinders

17

L2: Build systems see GCC

tamnd/rucc#3277 · opened

on Oct 7, 2026

· L2: Build systems see GCC

2

Navigation Menu

Repository navigation

Issues

Compile time: what is left for a 2x faster kernel build than gcc-16

inline: the unit is measured before small memcpy copies fold, so hot hinted calls in lib/maple_tree.c hit the unit growth limit

aarch64: an integer asm operand with the w constraint is refused

Read gcc's arm_neon.h: fp16, the aarch64 pragma and the builtin_aarch64_ builtins

Put in the KASAN checks for -fsanitize=kernel-address

What Linux 7.3 riscv64 defconfig needs from a riscv64 back end

Read the function target attribute for NEON on aarch64

Pass short vectors in v registers on aarch64 as AAPCS64 says

Keep an asm that writes a global register variable, as gcc does with ASM_CALL_CONSTRAINT

Compile Linux 7.3 defconfig in half the CPU time of GCC 16, now 1.78x to 1.87x

__builtin_constant_p on a const local is 0 in an enumerator and 1 elsewhere

Write a memcpy inline when its length is not constant but known to be small, as gcc does for detoast_attr

32 byte vectors are passed in memory where gcc uses ymm under -mavx

glibc stubs: loongarch64 gets libutil and libanl that the real system does not have

short-circuit says it kept the loop forest after a collapse in a loop latch, and it did not

aarch64: __builtin_setjmp stops with an internal error on a64.lea_64

Decide on the stack protector canary after dead locals are removed, as gcc does

Write a long memcpy or memset with a known length inline, and turn a fill loop into memset, as gcc does

Fold a fortify check that the value range of the length proves cannot fail

L9: RISC-V 64 Linux

L8: Distribution rebuild and packages

L7: AArch64, musl and i686 rows

L6: Headers and the C library

L5: Debuggers, profilers and unwinders

L2: Build systems see GCC