From: Kevin Brothaler <redacted>
Date: Tue, 20 Dec 2022 18:33:33 +0000 (-0500)
Subject: Check for both __ARM_NEON and __ARM_FEATURE_FMA so that the project can be compiled... 
X-Git-Tag: upstream/1.7.4~1688
X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=e1432dd91add9ba93fdbb7afedb52f224410f9dd;p=pkg%2Fggml%2Fsources%2Fwhisper.cpp

Check for both __ARM_NEON and __ARM_FEATURE_FMA so that the project can be compiled for armv7a.

Android armeabi-v7a's NEON support doesn't support FMA unless configured with `-mfpu=neon-fp-armv8`, which would need runtime checks.
* Also removed ABI filter from Android project.
---

diff --git a/examples/whisper.android/.idea/gradle.xml b/examples/whisper.android/.idea/gradle.xml
index a9f4e522..a2d7c213 100644
--- a/examples/whisper.android/.idea/gradle.xml
+++ b/examples/whisper.android/.idea/gradle.xml
@@ -1,5 +1,6 @@
 <?xml version="1.0" encoding="UTF-8"?>
 <project version="4">
+  <component name="GradleMigrationSettings" migrationVersion="1" />
   <component name="GradleSettings">
     <option name="linkedExternalProjectsSettings">
       <GradleProjectSettings>
diff --git a/examples/whisper.android/app/build.gradle b/examples/whisper.android/app/build.gradle
index 64a43bbb..66add991 100644
--- a/examples/whisper.android/app/build.gradle
+++ b/examples/whisper.android/app/build.gradle
@@ -14,10 +14,6 @@ android {
         versionCode 1
         versionName "1.0"
 
-        ndk {
-            abiFilters 'arm64-v8a', 'x86_64'
-        }
-
         testInstrumentationRunner "androidx.test.runner.AndroidJUnitRunner"
         vectorDrawables {
             useSupportLibrary true
diff --git a/ggml.c b/ggml.c
index f1d2b256..df4a779f 100644
--- a/ggml.c
+++ b/ggml.c
@@ -333,7 +333,7 @@ inline static void ggml_vec_div_f32 (const int n, float * z, const float * x, co
 
 inline static void ggml_vec_dot_f32(const int n, float * restrict s, const float * restrict x, const float * restrict y) {
     ggml_float sumf = 0.0;
-#ifdef __ARM_NEON
+#if defined(__ARM_NEON) && defined(__ARM_FEATURE_FMA)
     // NEON 128-bit
     const int n16 = (n & ~15);
 
@@ -511,7 +511,7 @@ inline static void ggml_vec_dot_f32(const int n, float * restrict s, const float
 
 inline static void ggml_vec_dot_f16(const int n, float * restrict s, ggml_fp16_t * restrict x, ggml_fp16_t * restrict y) {
     ggml_float sumf = 0.0;
-#ifdef __ARM_NEON
+#if defined(__ARM_NEON) && defined(__ARM_FEATURE_FMA)
     const int n32 = (n & ~31);
 
 #if defined(__ARM_FEATURE_FP16_VECTOR_ARITHMETIC)
@@ -760,7 +760,7 @@ inline static void ggml_vec_dot_f16(const int n, float * restrict s, ggml_fp16_t
 }
 
 inline static void ggml_vec_mad_f32(const int n, float * restrict y, const float * restrict x, const float v) {
-#ifdef __ARM_NEON
+#if defined(__ARM_NEON) && defined(__ARM_FEATURE_FMA)
     // NEON 128-bit
     const int n16 = (n & ~15);
 
@@ -909,7 +909,7 @@ inline static void ggml_vec_mad_f32(const int n, float * restrict y, const float
 }
 
 inline static void ggml_vec_mad_f16(const int n, ggml_fp16_t * restrict y, ggml_fp16_t * restrict x, const float v) {
-#ifdef __ARM_NEON
+#if defined(__ARM_NEON) && defined(__ARM_FEATURE_FMA)
     // NEON 128-bit
     const int n32 = (n & ~31);
 
@@ -8432,6 +8432,14 @@ int ggml_cpu_has_neon(void) {
 #endif
 }
 
+int ggml_cpu_has_arm_fma(void) {
+#if defined(__ARM_FEATURE_FMA)
+    return 1;
+#else
+    return 0;
+#endif
+}
+
 int ggml_cpu_has_f16c(void) {
 #if defined(__F16C__)
     return 1;
diff --git a/ggml.h b/ggml.h
index 7f24e531..4bb6118f 100644
--- a/ggml.h
+++ b/ggml.h
@@ -725,6 +725,7 @@ int ggml_cpu_has_avx(void);
 int ggml_cpu_has_avx2(void);
 int ggml_cpu_has_avx512(void);
 int ggml_cpu_has_neon(void);
+int ggml_cpu_has_arm_fma(void);
 int ggml_cpu_has_f16c(void);
 int ggml_cpu_has_fp16_va(void);
 int ggml_cpu_has_wasm_simd(void);
diff --git a/whisper.cpp b/whisper.cpp
index 0aaaea0d..fb507053 100644
--- a/whisper.cpp
+++ b/whisper.cpp
@@ -2555,6 +2555,7 @@ const char * whisper_print_system_info(void) {
     s += "AVX2 = "      + std::to_string(ggml_cpu_has_avx2())      + " | ";
     s += "AVX512 = "    + std::to_string(ggml_cpu_has_avx512())    + " | ";
     s += "NEON = "      + std::to_string(ggml_cpu_has_neon())      + " | ";
+    s += "ARM FMA = "   + std::to_string(ggml_cpu_has_arm_fma())   + " | ";
     s += "F16C = "      + std::to_string(ggml_cpu_has_f16c())      + " | ";
     s += "FP16_VA = "   + std::to_string(ggml_cpu_has_fp16_va())   + " | ";
     s += "WASM_SIMD = " + std::to_string(ggml_cpu_has_wasm_simd()) + " | ";