upstream: https://github.com/ggml-org/llama.cpp/pull/26860 From acb20b139639ec1b7c78739d98f3b04f500ff689 Mon Sep 17 00:00:00 2001 From: Pascal Date: Mon, 10 Aug 2026 20:51:29 +0200 Subject: [PATCH 1/2] ggml-cpu: gate __fp16 on __ARM_FP16_FORMAT_IEEE __ARM_NEON only signals NEON availability. The __fp16 type also needs the IEEE half format, implied on AArch64 but selected with -mfp16-format=ieee on 32 bit Arm, where the compiler otherwise rejects the type. The guard keeps every toolchain that provides the type on the same code and sends that one configuration to the generic lookup path. --- ggml/src/ggml-cpu/simd-mappings.h | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) diff --git a/third_party/llama.cpp/ggml/src/ggml-cpu/simd-mappings.h b/third_party/llama.cpp/ggml/src/ggml-cpu/simd-mappings.h index fca5119e1..4179605d5 100644 --- a/third_party/llama.cpp/ggml/src/ggml-cpu/simd-mappings.h +++ b/third_party/llama.cpp/ggml/src/ggml-cpu/simd-mappings.h @@ -29,13 +29,14 @@ extern "C" { // FP16 to FP32 conversion // 16-bit float -// on Arm, we use __fp16 +// on Arm, we use __fp16, which requires the IEEE fp16 format: implied on AArch64, selected by +// -mfp16-format=ieee on 32-bit Arm, where the compiler may otherwise reject the type // on x86, we use uint16_t // // for old CUDA compilers (<= 11), we use uint16_t: ref https://github.com/ggml-org/llama.cpp/pull/10616 // for MUSA compilers , we use uint16_t: ref https://github.com/ggml-org/llama.cpp/pull/11843 // -#if defined(__ARM_NEON) && !(defined(__CUDACC__) && __CUDACC_VER_MAJOR__ <= 11) && !defined(__MUSACC__) +#if defined(__ARM_NEON) && defined(__ARM_FP16_FORMAT_IEEE) && !(defined(__CUDACC__) && __CUDACC_VER_MAJOR__ <= 11) && !defined(__MUSACC__) #define GGML_CPU_COMPUTE_FP16_TO_FP32(x) neon_compute_fp16_to_fp32(x) #define GGML_CPU_COMPUTE_FP32_TO_FP16(x) neon_compute_fp32_to_fp16(x) From cf1070103a1d10e4f73d258f51f121021b05919c Mon Sep 17 00:00:00 2001 From: Pascal Date: Tue, 11 Aug 2026 13:09:21 +0200 Subject: [PATCH 2/2] ggml-cpu: gate the NEON+FMA block on __ARM_FP16_FORMAT_IEEE Both halves of the F16 section dereference __fp16, so armv7 with neon-vfpv4 hits the same unknown type error. Without the IEEE format the configuration now falls back to the scalar path. Address review from @JonathanC-ARM --- ggml/src/ggml-cpu/simd-mappings.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/third_party/llama.cpp/ggml/src/ggml-cpu/simd-mappings.h b/third_party/llama.cpp/ggml/src/ggml-cpu/simd-mappings.h index 8c6a15680096..10ce4bfc593b 100644 --- a/third_party/llama.cpp/ggml/src/ggml-cpu/simd-mappings.h +++ b/third_party/llama.cpp/ggml/src/ggml-cpu/simd-mappings.h @@ -328,7 +328,7 @@ inline static float ggml_lookup_fp16_to_fp32(ggml_fp16_t f) { #define GGML_F16_VEC_REDUCE GGML_F32Cx4_REDUCE #endif -#elif defined(__ARM_NEON) && defined(__ARM_FEATURE_FMA) +#elif defined(__ARM_NEON) && defined(__ARM_FEATURE_FMA) && defined(__ARM_FP16_FORMAT_IEEE) #define GGML_SIMD