Move copy to the new format

benchmarks/BM_copy.cpp

@@ -43,14 +43,16 @@ BENCHMARK_DEFINE_F(CopyArray, StdCopy)(benchmark::State& state) {
 BENCHMARK_DEFINE_F(CopyArray, Scalar)(benchmark::State& state) {
     for (auto _ : state)
     {
-        sfz::copy<float, false>(input, absl::MakeSpan(output));
+        sfz::setSIMDOpStatus(sfz::SIMDOps::copy, false);
+        sfz::copy<float>(input, absl::MakeSpan(output));
     }
 }
 
 BENCHMARK_DEFINE_F(CopyArray, SIMD)(benchmark::State& state) {
     for (auto _ : state)
     {
-        sfz::copy<float, true>(input, absl::MakeSpan(output));
+        sfz::setSIMDOpStatus(sfz::SIMDOps::copy, true);
+        sfz::copy<float>(input, absl::MakeSpan(output));
     }
 }
 
@@ -64,14 +66,16 @@ BENCHMARK_DEFINE_F(CopyArray, StdCopy_Unaligned)(benchmark::State& state) {
 BENCHMARK_DEFINE_F(CopyArray, Scalar_Unaligned)(benchmark::State& state) {
     for (auto _ : state)
     {
-        sfz::copy<float, false>(absl::MakeSpan(input).subspan(1), absl::MakeSpan(output).subspan(1));
+        sfz::setSIMDOpStatus(sfz::SIMDOps::copy, false);
+        sfz::copy<float>(absl::MakeSpan(input).subspan(1), absl::MakeSpan(output).subspan(1));
     }
 }
 
 BENCHMARK_DEFINE_F(CopyArray, SIMD_Unaligned)(benchmark::State& state) {
     for (auto _ : state)
     {
-        sfz::copy<float, true>(absl::MakeSpan(input).subspan(1), absl::MakeSpan(output).subspan(1));
+        sfz::setSIMDOpStatus(sfz::SIMDOps::copy, true);
+        sfz::copy<float>(absl::MakeSpan(input).subspan(1), absl::MakeSpan(output).subspan(1));
     }
 }
 

src/sfizz/SIMDHelpers.cpp

@@ -445,4 +445,30 @@ void subtract<float>(float value, float* output, unsigned size) noexcept
         *output++ -= value;
 }
 
+template <>
+void copy<float>(const float* input, float* output, unsigned size) noexcept
+{
+    // The sentinel is the input here
+    const auto sentinel = input + size;
+
+    if (getSIMDOpStatus(SIMDOps::copy)) {
+#if SFIZZ_CPU_FAMILY_X86_64 || SFIZZ_CPU_FAMILY_I386
+        if (cpuInfo.has_sse()) {
+            const auto* lastAligned = prevAligned(sentinel);
+
+            while (unaligned(input, output) && input < lastAligned)
+                *output++ = *input++;
+
+            while (input < lastAligned) {
+                _mm_store_ps(output, _mm_load_ps(input));
+                incrementAll<4>(input, output);
+            }
+            // fallthrough from lastAligned to sentinel
+        }
+#endif
+    }
+
+    std::copy(input, sentinel, output);
+}
+
 }

src/sfizz/SIMDHelpers.h

@@ -501,39 +501,29 @@ void subtract(T value, absl::Span<T> output) noexcept
     subtract<T>(value, output.data(), output.size());
 }
 
-namespace _internals {
-    template <class T>
-    void snippetCopy(const T*& input, T*& output)
-    {
-        *output++ = *input++;
-    }
-}
-
 /**
  * @brief Copy a span in another
  *
- * The output size will be the minimum of the input span and output span sizes.
- *
  * @tparam T the underlying type
- * @tparam SIMD use the SIMD version or the scalar version
  * @param input
  * @param output
+ * @param size
  */
-template <class T, bool SIMD = SIMDConfig::copy>
+template <class T>
-void copy(absl::Span<const T> input, absl::Span<T> output) noexcept
+void copy(const T* input, T* output, unsigned size) noexcept
 {
-    CHECK(output.size() >= input.size());
+    std::copy(input, input + size, output);
-    if (output.data() == input.data() && output.size() == input.size())
-        return;
-    auto* in = input.begin();
-    auto* out = output.begin();
-    auto* sentinel = out + min(input.size(), output.size());
-    while (out < sentinel)
-        _internals::snippetCopy(in, out);
 }
 
 template <>
-void copy<float, true>(absl::Span<const float> input, absl::Span<float> output) noexcept;
+void copy<float>(const float* input, float* output, unsigned size) noexcept;
+
+template <class T>
+void copy(absl::Span<const T> input, absl::Span<T> output) noexcept
+{
+    CHECK_SPAN_SIZES(input, output);
+    copy<T>(input.data(), output.data(), minSpanSize(input, output));
+}
 
 namespace _internals {
     // Number of elements in the table, odd for equal volume at center

src/sfizz/SIMDSSE.cpp

@@ -16,27 +16,6 @@
 
 constexpr uintptr_t TypeAlignment = 4;
 
-template <>
-void sfz::copy<float, true>(absl::Span<const float> input, absl::Span<float> output) noexcept
-{
-    CHECK(output.size() >= input.size());
-    auto* in = input.begin();
-    auto* out = output.begin();
-    auto* sentinel = out + min(input.size(), output.size());
-    const auto* lastAligned = prevAligned(sentinel);
-
-    while (unaligned(in, out) && out < lastAligned)
-        _internals::snippetCopy<float>(in, out);
-
-    while (out < lastAligned) {
-        _mm_store_ps(out, _mm_load_ps(in));
-        incrementAll<TypeAlignment>(in, out);
-    }
-
-    while (out < sentinel)
-        _internals::snippetCopy<float>(in, out);
-}
-
 template <>
 void sfz::pan<float, true>(absl::Span<const float> panEnvelope, absl::Span<float> leftBuffer, absl::Span<float> rightBuffer) noexcept
 {

tests/SIMDHelpersT.cpp

@@ -620,7 +620,8 @@ TEST_CASE("[Helpers] copy")
 {
     std::array<float, 5> input { 1.0f, 2.0f, 3.0f, 4.0f, 5.0f };
     std::array<float, 5> output { 1.0f, 1.0f, 1.0f, 1.0f, 1.0f };
-    sfz::copy<float, false>(input, absl::MakeSpan(output));
+    sfz::setSIMDOpStatus(sfz::SIMDOps::copy, false);
+    sfz::copy<float>(input, absl::MakeSpan(output));
     REQUIRE(output == input);
 }
 
@@ -628,7 +629,8 @@ TEST_CASE("[Helpers] copy (SIMD)")
 {
     std::array<float, 5> input { 1.0f, 2.0f, 3.0f, 4.0f, 5.0f };
     std::array<float, 5> output { 1.0f, 1.0f, 1.0f, 1.0f, 1.0f };
-    sfz::copy<float, true>(input, absl::MakeSpan(output));
+    sfz::setSIMDOpStatus(sfz::SIMDOps::copy, true);
+    sfz::copy<float>(input, absl::MakeSpan(output));
     REQUIRE(output == input);
 }
 
@@ -641,8 +643,10 @@ TEST_CASE("[Helpers] copy (SIMD vs scalar)")
     absl::c_fill(outputScalar, 0.0f);
     absl::c_fill(outputSIMD, 0.0f);
 
-    sfz::copy<float, false>(input, absl::MakeSpan(outputScalar));
+    sfz::setSIMDOpStatus(sfz::SIMDOps::copy, false);
-    sfz::copy<float, true>(input, absl::MakeSpan(outputSIMD));
+    sfz::copy<float>(input, absl::MakeSpan(outputScalar));
+    sfz::setSIMDOpStatus(sfz::SIMDOps::copy, true);
+    sfz::copy<float>(input, absl::MakeSpan(outputSIMD));
     REQUIRE(approxEqual<float>(outputScalar, outputSIMD));
 }