Skip to contentSkip to conversation

package/libabseil-cpp: fix build on x86 32-bit with AES

PATCH Applied Yegor Yefremov via buildroot 1 participant 2 messages Sep 25 – Sep 28

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 package/libabseil-cpp: fix build on x86 32-bit with AES

From: Yegor Yefremov <[email protected]>

Since the bump to version 20260817.0, the AES-based hash for long strings is used whenever __SSE4_2__ and __AES__ are defined. It relies on _mm_cvtsi128_si64() and _mm_extract_epi64(), which are only available on x86_64, so the build fails on 32-bit x86 for CPUs that support SSE4.2 and AES:

hash.cc:115:32: error: '_mm_cvtsi128_si64' was not declared in this scope

Add a patch from a pending upstream pull request restricting this code path to x86_64.

Fixes:
https://autobuild.buildroot.org/results/48198506ef76c6ebad0a802cb1fdd5270494eae1

Assisted-by Claude:claude-opus-5
---
 ...0002-Fix-32-bit-x86-build-of-hash.cc.patch | 33 +++++++++++++++++++
 1 file changed, 33 insertions(+)
 create mode 100644 package/libabseil-cpp/0002-Fix-32-bit-x86-build-of-hash.cc.patch
package/libabseil-cpp/0002-Fix-32-bit-x86-build-of-hash.cc.patch new file +33 -0
new file mode 100644
index 0000000000..1a289fe115
@@ -0,0 +1,33 @@
+From 66cd3bf8a9c19f03c5b0bf58031ab8176fb5608a Mon Sep 17 00:00:00 2001
+From: Jaipaul Cheernam <[email protected]>
+Date: Thu, 24 Sep 2026 14:59:26 +0000
+Subject: [PATCH] Fix 32-bit x86 build of hash.cc
+
+The x86 AES SIMD path in absl/hash/internal/hash.cc uses the 64-bit-only
+_mm_cvtsi128_si64 and _mm_extract_epi64 intrinsics but is gated only on
+__SSE4_2__ && __AES__, so 32-bit x86 builds fail to compile.
+
+Require __x86_64__/_M_X64 for that path, matching the existing guard in
+absl/crc/internal/crc32_x86_arm_combined_simd.h.
+
+Fixes: https://github.com/abseil/abseil-cpp/issues/2176
+Upstream: https://github.com/abseil/abseil-cpp/pull/2177
+Signed-off-by: Yegor Yefremov <[email protected]>
+---
+ absl/hash/internal/hash.cc | 3 ++-
+ 1 file changed, 2 insertions(+), 1 deletion(-)
+
+diff --git a/absl/hash/internal/hash.cc b/absl/hash/internal/hash.cc
+index 43cb561..ef85625 100644
+--- a/absl/hash/internal/hash.cc
++++ b/absl/hash/internal/hash.cc
+@@ -28,7 +28,8 @@
+ 
+ #ifdef ABSL_AES_INTERNAL_HAVE_X86_SIMD
+ #error ABSL_AES_INTERNAL_HAVE_X86_SIMD cannot be directly set
+-#elif defined(__SSE4_2__) && defined(__AES__)
++#elif defined(__SSE4_2__) && defined(__AES__) && \
++    (defined(__x86_64__) || defined(_M_X64))
+ #define ABSL_AES_INTERNAL_HAVE_X86_SIMD
+ #endif
+ 
Signature · 6 lines
2.34.1

_______________________________________________
buildroot mailing list
[email protected]
https://lists.buildroot.org/mailman/listinfo/buildroot
#1 lore raw
Quoted 10 lines · Yegor Yefremov via buildroot

From: Yegor Yefremov <[email protected]>

Since the bump to version 20260817.0, the AES-based hash for long strings is used whenever __SSE4_2__ and __AES__ are defined. It relies on _mm_cvtsi128_si64() and _mm_extract_epi64(), which are only available on x86_64, so the build fails on 32-bit x86 for CPUs that support SSE4.2 and AES:

hash.cc:115:32: error: '_mm_cvtsi128_si64' was not declared in this scope

Quoted 8 lines

Add a patch from a pending upstream pull request restricting this code path to x86_64.

Fixes:
https://autobuild.buildroot.org/results/48198506ef76c6ebad0a802cb1fdd5270494eae1

Assisted-by Claude:claude-opus-5

Applied to master, thanks!

Best regards,

Thomas Petazzoni

Signature · 7 lines
Thomas Petazzoni, co-owner and CEO, Bootlin
Embedded Linux and Kernel engineering and training
https://bootlin.com
_______________________________________________
buildroot mailing list
[email protected]
https://lists.buildroot.org/mailman/listinfo/buildroot