From 01935e33388a2c649ca65e8a6ad80c30ad07fd7c Mon Sep 17 00:00:00 2001
From: Tom Poole <thomas.poole@roli.com>
Date: Tue, 11 Jun 2019 21:23:56 +0100
Subject: [PATCH] DSP:  Guarded a rogue FMA instruction in the AVX native ops

---
 modules/juce_dsp/native/juce_avx_SIMDNativeOps.h | 10 +++++++++-
 1 file changed, 9 insertions(+), 1 deletion(-)

diff --git a/modules/juce_dsp/native/juce_avx_SIMDNativeOps.h b/modules/juce_dsp/native/juce_avx_SIMDNativeOps.h
index 7ecf95b5cc..270660b442 100644
--- a/modules/juce_dsp/native/juce_avx_SIMDNativeOps.h
+++ b/modules/juce_dsp/native/juce_avx_SIMDNativeOps.h
@@ -91,7 +91,6 @@ struct SIMDNativeOps<float>
     static forcedinline __m256 JUCE_VECTOR_CALLTYPE greaterThan (__m256 a, __m256 b) noexcept            { return _mm256_cmp_ps (a, b, _CMP_GT_OQ); }
     static forcedinline __m256 JUCE_VECTOR_CALLTYPE greaterThanOrEqual (__m256 a, __m256 b) noexcept     { return _mm256_cmp_ps (a, b, _CMP_GE_OQ); }
     static forcedinline bool   JUCE_VECTOR_CALLTYPE allEqual (__m256 a, __m256 b) noexcept               { return (_mm256_movemask_ps (equal (a, b)) == 0xff); }
-    static forcedinline __m256 JUCE_VECTOR_CALLTYPE multiplyAdd (__m256 a, __m256 b, __m256 c) noexcept  { return _mm256_fmadd_ps (b, c, a); }
     static forcedinline __m256 JUCE_VECTOR_CALLTYPE dupeven (__m256 a) noexcept                          { return _mm256_shuffle_ps (a, a, _MM_SHUFFLE (2, 2, 0, 0)); }
     static forcedinline __m256 JUCE_VECTOR_CALLTYPE dupodd (__m256 a) noexcept                           { return _mm256_shuffle_ps (a, a, _MM_SHUFFLE (3, 3, 1, 1)); }
     static forcedinline __m256 JUCE_VECTOR_CALLTYPE swapevenodd (__m256 a) noexcept                      { return _mm256_shuffle_ps (a, a, _MM_SHUFFLE (2, 3, 0, 1)); }
@@ -99,6 +98,15 @@ struct SIMDNativeOps<float>
     static forcedinline __m256 JUCE_VECTOR_CALLTYPE set (__m256 v, size_t i, float s) noexcept           { return SIMDFallbackOps<float, __m256>::set (v, i, s); }
     static forcedinline __m256 JUCE_VECTOR_CALLTYPE truncate (__m256 a) noexcept                         { return _mm256_cvtepi32_ps (_mm256_cvttps_epi32 (a)); }
 
+    static forcedinline __m256 JUCE_VECTOR_CALLTYPE multiplyAdd (__m256 a, __m256 b, __m256 c) noexcept
+    {
+       #if __FMA__
+        return _mm256_fmadd_ps (b, c, a);
+       #else
+        return add (a, mul (b, c));
+       #endif
+    }
+
     static forcedinline __m256 JUCE_VECTOR_CALLTYPE oddevensum (__m256 a) noexcept
     {
         a = _mm256_add_ps (_mm256_shuffle_ps (a, a, _MM_SHUFFLE (1, 0, 3, 2)), a);