Skip to content

Make LibDeflate finally compile on Linux ARM #2361

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
jviotti merged 1 commit into from
Apr 24, 2026
Merged

Make LibDeflate finally compile on Linux ARM #2361

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
78 changes: 63 additions & 15 deletions cmake/FindLibDeflate.cmake
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -27,27 +27,75 @@ if(NOT LibDeflate_FOUND)
sourcemeta_add_default_options(PRIVATE libdeflate)

# Check if the assembler supports ARM dot-product (udot) instructions.
# GCC 14+ assumes binutils is new enough, but some CI environments
# pair GCC 14 with older binutils that lack udot support
if(CMAKE_SYSTEM_PROCESSOR MATCHES "aarch64|arm64|ARM64")
include(CheckCSourceCompiles)
set(LIBDEFLATE_SAVED_CMAKE_REQUIRED_FLAGS "${CMAKE_REQUIRED_FLAGS}")
set(CMAKE_REQUIRED_FLAGS "${CMAKE_REQUIRED_FLAGS} -march=armv8.2-a+dotprod")
check_c_source_compiles("
#include <arm_neon.h>
int main(void) {
uint32x4_t a = vdupq_n_u32(0);
uint8x16_t b = vdupq_n_u8(0);
uint8x16_t c = vdupq_n_u8(0);
a = vdotq_u32(a, b, c);
return 0;
}
" LIBDEFLATE_HAS_DOTPROD_ASSEMBLER)
set(CMAKE_REQUIRED_FLAGS "${LIBDEFLATE_SAVED_CMAKE_REQUIRED_FLAGS}")
if(CMAKE_C_COMPILER_ID STREQUAL "GNU" AND
CMAKE_C_COMPILER_VERSION VERSION_GREATER_EQUAL 14)
check_c_source_compiles("
#include <arm_neon.h>
__attribute__((target(\"+dotprod\")))
int test(void) {
uint32x4_t a = vdupq_n_u32(0);
uint8x16_t b = vdupq_n_u8(0);
uint8x16_t c = vdupq_n_u8(0);
a = vdotq_u32(a, b, c);
Copy link
Copy Markdown

@augmentcode augmentcode Bot Apr 24, 2026 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

check_c_source_compiles() here uses all-zero vector inputs, which some optimization settings could constant-fold so the generated assembly never contains a dotprod instruction. If that happens, this probe could incorrectly succeed even when the assembler can’t assemble dotprod, leading to later build failures when libdeflate’s real dotprod paths are compiled.

Severity: medium

Fix This in Augment

🤖 Was this useful? React with 👍 or 👎, or 🚀 if it prevented an incident/outage.

return (int)vgetq_lane_u32(a, 0);
}
int main(void) { return test(); }
" LIBDEFLATE_HAS_DOTPROD_ASSEMBLER)
else()
set(LIBDEFLATE_SAVED_CMAKE_REQUIRED_FLAGS "${CMAKE_REQUIRED_FLAGS}")
set(CMAKE_REQUIRED_FLAGS "${CMAKE_REQUIRED_FLAGS} -march=armv8.2-a+dotprod")
check_c_source_compiles("
#include <arm_neon.h>
int main(void) {
uint32x4_t a = vdupq_n_u32(0);
uint8x16_t b = vdupq_n_u8(0);
uint8x16_t c = vdupq_n_u8(0);
a = vdotq_u32(a, b, c);
return 0;
}
" LIBDEFLATE_HAS_DOTPROD_ASSEMBLER)
set(CMAKE_REQUIRED_FLAGS "${LIBDEFLATE_SAVED_CMAKE_REQUIRED_FLAGS}")
endif()
if(NOT LIBDEFLATE_HAS_DOTPROD_ASSEMBLER)
target_compile_definitions(libdeflate PRIVATE
LIBDEFLATE_ASSEMBLER_DOES_NOT_SUPPORT_DOTPROD)
endif()

if(CMAKE_C_COMPILER_ID STREQUAL "GNU" AND
CMAKE_C_COMPILER_VERSION VERSION_GREATER_EQUAL 14)
check_c_source_compiles("
#include <arm_neon.h>
__attribute__((target(\"+crypto,+crc,+sha3\")))
int test(void) {
uint8x16_t a = vdupq_n_u8(0);
uint8x16_t b = vdupq_n_u8(0);
uint8x16_t c = vdupq_n_u8(0);
a = veor3q_u8(a, b, c);
Copy link
Copy Markdown

@augmentcode augmentcode Bot Apr 24, 2026 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Similar to the dotprod probe, the veor3q_u8() test uses compile-time constant inputs and may be optimized into a constant result without emitting eor3/sha3 instructions. That can make LIBDEFLATE_HAS_SHA3_ASSEMBLER a false positive and re-introduce the SHA3 assembler failures this check is meant to guard against.

Severity: medium

Fix This in Augment

🤖 Was this useful? React with 👍 or 👎, or 🚀 if it prevented an incident/outage.

return (int)vgetq_lane_u8(a, 0);
}
int main(void) { return test(); }
" LIBDEFLATE_HAS_SHA3_ASSEMBLER)
else()
set(LIBDEFLATE_SAVED_CMAKE_REQUIRED_FLAGS "${CMAKE_REQUIRED_FLAGS}")
set(CMAKE_REQUIRED_FLAGS "${CMAKE_REQUIRED_FLAGS} -march=armv8.2-a+crypto+crc+sha3")
check_c_source_compiles("
#include <arm_neon.h>
int main(void) {
uint8x16_t a = vdupq_n_u8(0);
uint8x16_t b = vdupq_n_u8(0);
uint8x16_t c = vdupq_n_u8(0);
a = veor3q_u8(a, b, c);
return 0;
}
" LIBDEFLATE_HAS_SHA3_ASSEMBLER)
set(CMAKE_REQUIRED_FLAGS "${LIBDEFLATE_SAVED_CMAKE_REQUIRED_FLAGS}")
endif()
if(NOT LIBDEFLATE_HAS_SHA3_ASSEMBLER)
target_compile_definitions(libdeflate PRIVATE
LIBDEFLATE_ASSEMBLER_DOES_NOT_SUPPORT_SHA3)
endif()
endif()

target_include_directories(libdeflate PUBLIC
Expand Down
Loading