kfr

commit 237db95b698f5afbe3702feff8531ec00647fd4d
parent 6f6f113fe92c3795504fd23d5f684f8e7105f391
Author: [email protected] <[email protected]>
Date:   Wed, 13 Mar 2019 01:10:48 +0000

short_fir: change default vector width

Diffstat:
M
CHANGELOG.md
 | 
3
++-
M
include/kfr/dsp/fir.hpp
 | 
12
++++++------
M
tests/CMakeLists.txt
 | 
2
+-

3 files changed, 9 insertions(+), 8 deletions(-)
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -2,7 +2,7 @@
 
 ## 3.0.7
 
-2019-03-11
+2019-03-13
 
 #### Added
 
@@ -12,6 +12,7 @@
 #### Changed
 
 - `mask<>` is now a specialization of `vec<>`. This allows using many `vec` functions for masks
+- `short_fir` performance has been increased by around 50%-60%
 
 #### Fixed
 
diff --git a/include/kfr/dsp/fir.hpp b/include/kfr/dsp/fir.hpp
@@ -105,7 +105,7 @@ struct expression_short_fir : expression_with_arguments<E1>
 
     template <size_t N>
     KFR_INTRINSIC friend vec<U, N> get_elements(const expression_short_fir& self, cinput_t cinput,
-                                                    size_t index, vec_shape<U, N> x)
+                                                size_t index, vec_shape<U, N> x)
     {
         vec<U, N> in = self.argument_first(cinput, index, x);
 
@@ -133,7 +133,7 @@ struct expression_fir : expression_with_arguments<E1>
 
     template <size_t N>
     KFR_INTRINSIC friend vec<U, N> get_elements(const expression_fir& self, cinput_t cinput, size_t index,
-                                                    vec_shape<U, N> x)
+                                                vec_shape<U, N> x)
     {
         const size_t tapcount = self.state.s.taps.size();
         const vec<U, N> input = self.argument_first(cinput, index, x);
@@ -184,11 +184,11 @@ KFR_INTRINSIC internal::expression_fir<T, U, E1, true> fir(fir_state<T, U>& stat
  * @param taps coefficients for the FIR filter
  */
 template <typename T, size_t TapCount, typename E1>
-KFR_INTRINSIC internal::expression_short_fir<next_poweroftwo(TapCount), T, value_type_of<E1>, E1> short_fir(
-    E1&& e1, const univector<T, TapCount>& taps)
+KFR_INTRINSIC internal::expression_short_fir<next_poweroftwo(TapCount - 1) + 1, T, value_type_of<E1>, E1>
+short_fir(E1&& e1, const univector<T, TapCount>& taps)
 {
-    static_assert(TapCount >= 2 && TapCount <= 32, "Use short_fir only for small FIR filters");
-    return internal::expression_short_fir<next_poweroftwo(TapCount), T, value_type_of<E1>, E1>(
+    static_assert(TapCount >= 2 && TapCount <= 33, "Use short_fir only for small FIR filters");
+    return internal::expression_short_fir<next_poweroftwo(TapCount - 1) + 1, T, value_type_of<E1>, E1>(
         std::forward<E1>(e1), taps);
 }
 
diff --git a/tests/CMakeLists.txt b/tests/CMakeLists.txt
@@ -33,7 +33,7 @@ set(CMAKE_MODULE_PATH ${CMAKE_MODULE_PATH} "${CMAKE_SOURCE_DIR}/tests/cmake/")
 if (ENABLE_ASMTEST)
     add_executable(asm_test asm_test.cpp)
     target_link_libraries(asm_test kfr)
-    target_set_arch(asm_test PRIVATE sse2)
+    target_set_arch(asm_test PRIVATE avx)
     target_compile_definitions(asm_test PRIVATE KFR_SHOW_NOT_OPTIMIZED)
     if (CMAKE_CXX_COMPILER_ID STREQUAL "GNU")
         target_compile_options(asm_test PRIVATE -fno-stack-protector)