Added SSE, AVX and AVX512 optimizations for depan_lin and depan_eqpow…

… functions

include/private/dsp/arch/generic/pan.h

@@ -40,7 +40,7 @@ namespace lsp
                 const float sl  = fabsf(l[i]);
                 const float sr  = fabsf(r[i]);
                 const float den = sl + sr;
-                dst[i]          = (den > 1e-18f) ? sr / den : dfl;
+                dst[i]          = (den >= 1e-18f) ? sr / den : dfl;
             }
         }
 
@@ -51,7 +51,7 @@ namespace lsp
                 const float sl  = l[i] * l[i];
                 const float sr  = r[i] * r[i];
                 const float den = sl + sr;
-                dst[i]          = (den > 1e-36f) ? sr / den : dfl;
+                dst[i]          = (den >= 1e-36f) ? sr / den : dfl;
             }
         }
 

src/main/x86/avx.cpp

@@ -55,6 +55,7 @@
         #include <private/dsp/arch/x86/avx/hmath/hdotp.h>
 
         #include <private/dsp/arch/x86/avx/mix.h>
+        #include <private/dsp/arch/x86/avx/pan.h>
         #include <private/dsp/arch/x86/avx/search/minmax.h>
 
         #include <private/dsp/arch/x86/avx/fft.h>
@@ -333,6 +334,9 @@
                 CEXPORT1(favx, mix_copy4);
                 CEXPORT1(favx, mix_add4);
 
+                CEXPORT1(favx, depan_lin);
+                CEXPORT1(favx, depan_eqpow);
+
                 CEXPORT1(favx, min);
                 CEXPORT1(favx, max);
                 CEXPORT1(favx, minmax);
@@ -506,6 +510,8 @@
                     CEXPORT2(favx, dyn_biquad_process_x2, dyn_biquad_process_x2_fma3);
                     CEXPORT2(favx, dyn_biquad_process_x4, dyn_biquad_process_x4_fma3);
                     CEXPORT2(ffma, dyn_biquad_process_x8, dyn_biquad_process_x8_fma3);
+
+                    CEXPORT2(favx, depan_eqpow, depan_eqpow_fma3);
                 }
             }
 

src/main/x86/avx512.cpp

@@ -51,6 +51,7 @@
         #include <private/dsp/arch/x86/avx512/pcomplex.h>
         #include <private/dsp/arch/x86/avx512/pmath.h>
         #include <private/dsp/arch/x86/avx512/search.h>
+        #include <private/dsp/arch/x86/avx512/pan.h>
 
         #include <private/dsp/arch/x86/avx512/correlation.h>
     #undef PRIVATE_DSP_ARCH_X86_AVX512_IMPL
@@ -303,6 +304,9 @@
 
                 CEXPORT1(vl, corr_init);
                 CEXPORT1(vl, corr_incr);
+
+                CEXPORT1(vl, depan_lin);
+                CEXPORT1(vl, depan_eqpow);
             }
         } /* namespace avx2 */
     } /* namespace lsp */

src/main/x86/sse.cpp

@@ -51,6 +51,7 @@
         #include <private/dsp/arch/x86/sse/hmath/hdotp.h>
 
         #include <private/dsp/arch/x86/sse/mix.h>
+        #include <private/dsp/arch/x86/sse/pan.h>
 
         #include <private/dsp/arch/x86/sse/search/minmax.h>
 
@@ -282,6 +283,9 @@
                 EXPORT1(mix_copy4);
                 EXPORT1(mix_add4);
 
+                EXPORT1(depan_lin);
+                EXPORT1(depan_eqpow);
+
                 EXPORT1(reverse1);
                 EXPORT1(reverse2);
 

src/test/ptest/pan/depan_eqpow.cpp

@@ -0,0 +1,125 @@
+/*
+ * Copyright (C) 2024 Linux Studio Plugins Project <https://lsp-plug.in/>
+ *           (C) 2024 Vladimir Sadovnikov <sadko4u@gmail.com>
+ *
+ * This file is part of lsp-dsp-lib
+ * Created on: 27 нояб. 2024 г.
+ *
+ * lsp-dsp-lib is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU Lesser General Public License as published by
+ * the Free Software Foundation, either version 3 of the License, or
+ * any later version.
+ *
+ * lsp-dsp-lib is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public License
+ * along with lsp-dsp-lib. If not, see <https://www.gnu.org/licenses/>.
+ */
+
+#include <lsp-plug.in/common/alloc.h>
+#include <lsp-plug.in/common/types.h>
+#include <lsp-plug.in/dsp/dsp.h>
+#include <lsp-plug.in/test-fw/helpers.h>
+#include <lsp-plug.in/test-fw/ptest.h>
+
+#define MIN_RANK 8
+#define MAX_RANK 16
+
+namespace lsp
+{
+    namespace generic
+    {
+        void depan_eqpow(float *dst, const float *l, const float *r, float dfl, size_t count);
+    }
+
+    IF_ARCH_X86(
+        namespace sse
+        {
+            void depan_eqpow(float *dst, const float *l, const float *r, float dfl, size_t count);
+        }
+
+        namespace avx
+        {
+            void depan_eqpow(float *dst, const float *l, const float *r, float dfl, size_t count);
+            void depan_eqpow_fma3(float *dst, const float *l, const float *r, float dfl, size_t count);
+        }
+
+        namespace avx512
+        {
+            void depan_eqpow(float *dst, const float *l, const float *r, float dfl, size_t count);
+        }
+    )
+
+    IF_ARCH_ARM(
+        namespace neon_d32
+        {
+//            void depan_eqpow(float *dst, const float *l, const float *r, float dfl, size_t count);
+        }
+    )
+
+    IF_ARCH_AARCH64(
+        namespace asimd
+        {
+//            void depan_eqpow(float *dst, const float *l, const float *r, float dfl, size_t count);
+        }
+    )
+
+    typedef void (* depan_t) (float *dst, const float *l, const float *r, float dfl, size_t count);
+}
+
+//-----------------------------------------------------------------------------
+// Performance test
+PTEST_BEGIN("dsp.pan", depan_eqpow, 5, 1000)
+
+    void call(const char *label, float *dst, const float *src1, const float *src2, size_t count, depan_t func)
+    {
+        if (!PTEST_SUPPORTED(func))
+            return;
+
+        char buf[80];
+        sprintf(buf, "%s x %d", label, int(count));
+        printf("Testing %s numbers...\n", buf);
+
+        PTEST_LOOP(buf,
+            func(dst, src1, src2, 0.5f, count);
+        );
+    }
+
+    PTEST_MAIN
+    {
+        size_t buf_size = 1 << MAX_RANK;
+        uint8_t *data   = NULL;
+        float *dst      = alloc_aligned<float>(data, buf_size * 3, 64);
+        float *src1     = &dst[buf_size];
+        float *src2     = &src1[buf_size];
+
+        randomize_sign(src1, buf_size);
+        randomize_sign(src2, buf_size);
+
+        #define CALL(func) \
+            call(#func, dst, src1, src2, count, func);
+
+        for (size_t i=MIN_RANK; i <= MAX_RANK; ++i)
+        {
+            size_t count = 1 << i;
+
+            CALL(generic::depan_eqpow);
+            IF_ARCH_X86(CALL(sse::depan_eqpow));
+            IF_ARCH_X86(CALL(avx::depan_eqpow));
+            IF_ARCH_X86(CALL(avx::depan_eqpow_fma3));
+            IF_ARCH_X86(CALL(avx512::depan_eqpow));
+//            IF_ARCH_ARM(CALL(neon_d32::depan_eqpow));
+//            IF_ARCH_AARCH64(CALL(asimd::depan_eqpow));
+            PTEST_SEPARATOR;
+        }
+
+        free_aligned(data);
+    }
+PTEST_END
+
+
+
+

src/test/ptest/pan/depan_lin.cpp

@@ -0,0 +1,123 @@
+/*
+ * Copyright (C) 2024 Linux Studio Plugins Project <https://lsp-plug.in/>
+ *           (C) 2024 Vladimir Sadovnikov <sadko4u@gmail.com>
+ *
+ * This file is part of lsp-dsp-lib
+ * Created on: 27 нояб. 2024 г.
+ *
+ * lsp-dsp-lib is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU Lesser General Public License as published by
+ * the Free Software Foundation, either version 3 of the License, or
+ * any later version.
+ *
+ * lsp-dsp-lib is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public License
+ * along with lsp-dsp-lib. If not, see <https://www.gnu.org/licenses/>.
+ */
+
+#include <lsp-plug.in/common/alloc.h>
+#include <lsp-plug.in/common/types.h>
+#include <lsp-plug.in/dsp/dsp.h>
+#include <lsp-plug.in/test-fw/helpers.h>
+#include <lsp-plug.in/test-fw/ptest.h>
+
+#define MIN_RANK 8
+#define MAX_RANK 16
+
+namespace lsp
+{
+    namespace generic
+    {
+        void depan_lin(float *dst, const float *l, const float *r, float dfl, size_t count);
+    }
+
+    IF_ARCH_X86(
+        namespace sse
+        {
+            void depan_lin(float *dst, const float *l, const float *r, float dfl, size_t count);
+        }
+
+        namespace avx
+        {
+            void depan_lin(float *dst, const float *l, const float *r, float dfl, size_t count);
+        }
+
+        namespace avx512
+        {
+            void depan_lin(float *dst, const float *l, const float *r, float dfl, size_t count);
+        }
+    )
+
+    IF_ARCH_ARM(
+        namespace neon_d32
+        {
+//            void depan_lin(float *dst, const float *l, const float *r, float dfl, size_t count);
+        }
+    )
+
+    IF_ARCH_AARCH64(
+        namespace asimd
+        {
+//            void depan_lin(float *dst, const float *l, const float *r, float dfl, size_t count);
+        }
+    )
+
+    typedef void (* depan_t) (float *dst, const float *l, const float *r, float dfl, size_t count);
+}
+
+//-----------------------------------------------------------------------------
+// Performance test
+PTEST_BEGIN("dsp.pan", depan_lin, 5, 1000)
+
+    void call(const char *label, float *dst, const float *src1, const float *src2, size_t count, depan_t func)
+    {
+        if (!PTEST_SUPPORTED(func))
+            return;
+
+        char buf[80];
+        sprintf(buf, "%s x %d", label, int(count));
+        printf("Testing %s numbers...\n", buf);
+
+        PTEST_LOOP(buf,
+            func(dst, src1, src2, 0.5f, count);
+        );
+    }
+
+    PTEST_MAIN
+    {
+        size_t buf_size = 1 << MAX_RANK;
+        uint8_t *data   = NULL;
+        float *dst      = alloc_aligned<float>(data, buf_size * 3, 64);
+        float *src1     = &dst[buf_size];
+        float *src2     = &src1[buf_size];
+
+        randomize_sign(src1, buf_size);
+        randomize_sign(src2, buf_size);
+
+        #define CALL(func) \
+            call(#func, dst, src1, src2, count, func);
+
+        for (size_t i=MIN_RANK; i <= MAX_RANK; ++i)
+        {
+            size_t count = 1 << i;
+
+            CALL(generic::depan_lin);
+            IF_ARCH_X86(CALL(sse::depan_lin));
+            IF_ARCH_X86(CALL(avx::depan_lin));
+            IF_ARCH_X86(CALL(avx512::depan_lin));
+//            IF_ARCH_ARM(CALL(neon_d32::depan_lin));
+//            IF_ARCH_AARCH64(CALL(asimd::depan_lin));
+            PTEST_SEPARATOR;
+        }
+
+        free_aligned(data);
+    }
+PTEST_END
+
+
+
+