Fixes

Author: jatinchowdhury18

chowdsp_fft.cpp

@@ -90,6 +90,7 @@ FFT_Setup* fft_new_setup (int N, fft_transform_t transform);
 void fft_destroy_setup (FFT_Setup* s);
 void pffft_transform_internal (FFT_Setup* setup, const float* finput, float* foutput, void* scratch, fft_direction_t direction, int ordered);
 void pffft_convolve_internal (FFT_Setup* setup, const float* a, const float* b, float* ab, float scaling);
+void fft_accumulate_internal (FFT_Setup* setup, const float* a, const float* b, float* ab, int N);
 } // namespace chowdsp::fft::avx
 static constexpr uintptr_t address_mask = ~static_cast<uintptr_t> (3);
 static constexpr uintptr_t typeid_mask = static_cast<uintptr_t> (3);
@@ -409,15 +410,15 @@ void fft_accumulate (void* setup, const float* a, const float* b, float* ab, int
 #if CHOWDSP_FFT_COMPILER_SUPPORTS_AVX
     if (check_is_pointer_sse_setup (setup))
     {
-        sse::fft_accumulate_internal (reinterpret_cast<sse::FFT_Setup*> (setup),
+        sse::fft_accumulate_internal (reinterpret_cast<sse::FFT_Setup*> (get_setup_pointer (setup)),
                                       a,
                                       b,
                                       ab,
                                       N);
     }
     else
     {
-        avx::fft_accumulate_internal (reinterpret_cast<avx::FFT_Setup*> (setup),
+        avx::fft_accumulate_internal (reinterpret_cast<avx::FFT_Setup*> (get_setup_pointer (setup)),
                                       a,
                                       b,
                                       ab,

simd/chowdsp_fft_impl_avx.cpp

@@ -2029,9 +2029,9 @@ void pffft_convolve_internal (FFT_Setup* setup, const float* a, const float* b,
     }
 }
 
-void fft_accumulate_internal (FFT_Setup* setup, const float* a, const float* b, float* ab, int N)
+void fft_accumulate_internal ([[maybe_unused]] FFT_Setup* setup, const float* a, const float* b, float* ab, int N)
 {
-    assert (N % SIMD_SZ == 0);
+    assert (N % (SIMD_SZ * 2) == 0);
     const auto Ncvec = N / (int) SIMD_SZ;
     auto* va = (const __m256*) a;
     auto* vb = (const __m256*) b;

simd/chowdsp_fft_impl_neon.cpp

@@ -1677,7 +1677,7 @@ void pffft_convolve_internal (FFT_Setup* setup, const float* a, const float* b,
 
 void fft_accumulate_internal (FFT_Setup* setup, const float* a, const float* b, float* ab, int N)
 {
-    assert (N % SIMD_SZ == 0);
+    assert (N % (SIMD_SZ * 2) == 0);
     const auto Ncvec = N / (int) SIMD_SZ;
     auto* va = (const float32x4_t*) a;
     auto* vb = (const float32x4_t*) b;

simd/chowdsp_fft_impl_sse.cpp

@@ -1692,7 +1692,7 @@ void pffft_convolve_internal (FFT_Setup* setup, const float* a, const float* b,
 
 void fft_accumulate_internal (FFT_Setup* setup, const float* a, const float* b, float* ab, int N)
 {
-    assert (N % SIMD_SZ == 0);
+    assert (N % (SIMD_SZ * 2) == 0);
     const auto Ncvec = N / (int) SIMD_SZ;
     auto* va = (const __m128*) a;
     auto* vb = (const __m128*) b;

test/test.cpp

@@ -30,10 +30,7 @@ void test_fft_complex (int N, bool use_avx = false)
     auto* fft_setup = chowdsp::fft::fft_new_setup (N, chowdsp::fft::FFT_COMPLEX, use_avx);
     REQUIRE (fft_setup != nullptr);
     auto* pffft_setup = pffft_new_setup (N, PFFFT_COMPLEX);
-
-    if (use_avx)
-        REQUIRE (chowdsp::fft::fft_simd_width_bytes (fft_setup) == 32);
-    else
+    if (! use_avx)
         REQUIRE (chowdsp::fft::fft_simd_width_bytes (fft_setup) == 16);
 
     chowdsp::fft::fft_transform (fft_setup, data, data, work_data, chowdsp::fft::FFT_FORWARD);