Move the FIR4 from SSE2 to SSE3

SSE3 can avoid the slow _MM_TRANSPOSE_PS4 call thanks to the inclusion of
horizontal adds.

1 files changed, 25 insertions, 0 deletions

diff --git a/CMakeLists.txt b/CMakeLists.txt
index ea2d5150..af9e961a 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -325,6 +325,10 @@ IF(NOT MSVC)
     IF(HAVE_MSSE2_SWITCH)
         SET(SSE2_SWITCH "-msse2")
     ENDIF()
+    CHECK_C_COMPILER_FLAG(-msse3 HAVE_MSSE3_SWITCH)
+    IF(HAVE_MSSE3_SWITCH)
+        SET(SSE3_SWITCH "-msse3")
+    ENDIF()
     CHECK_C_COMPILER_FLAG(-msse4.1 HAVE_MSSE4_1_SWITCH)
     IF(HAVE_MSSE4_1_SWITCH)
         SET(SSE4_1_SWITCH "-msse4.1")
@@ -592,6 +596,7 @@ SET(ALC_OBJS  Alc/ALc.c
 SET(CPU_EXTS "Default")
 SET(HAVE_SSE        0)
 SET(HAVE_SSE2       0)
+SET(HAVE_SSE3       0)
 SET(HAVE_SSE4_1     0)
 SET(HAVE_NEON       0)
 
@@ -652,6 +657,26 @@ IF(ALSOFT_REQUIRE_SSE2 AND NOT HAVE_SSE2)
     MESSAGE(FATAL_ERROR "Failed to enable required SSE2 CPU extensions")
 ENDIF()
 
+OPTION(ALSOFT_REQUIRE_SSE2 "Require SSE3 support" OFF)
+CHECK_INCLUDE_FILE(pmmintrin.h HAVE_PMMINTRIN_H "${SSE3_SWITCH}")
+IF(HAVE_EMMINTRIN_H)
+    OPTION(ALSOFT_CPUEXT_SSE3 "Enable SSE3 support" ON)
+    IF(HAVE_SSE2 AND ALSOFT_CPUEXT_SSE3)
+        IF(ALIGN_DECL OR HAVE_C11_ALIGNAS)
+            SET(HAVE_SSE3 1)
+            SET(ALC_OBJS  ${ALC_OBJS} Alc/mixer_sse3.c)
+            IF(SSE2_SWITCH)
+                SET_SOURCE_FILES_PROPERTIES(Alc/mixer_sse3.c PROPERTIES
+                                            COMPILE_FLAGS "${SSE3_SWITCH}")
+            ENDIF()
+            SET(CPU_EXTS "${CPU_EXTS}, SSE3")
+        ENDIF()
+    ENDIF()
+ENDIF()
+IF(ALSOFT_REQUIRE_SSE3 AND NOT HAVE_SSE3)
+    MESSAGE(FATAL_ERROR "Failed to enable required SSE3 CPU extensions")
+ENDIF()
+
 OPTION(ALSOFT_REQUIRE_SSE4_1 "Require SSE4.1 support" OFF)
 CHECK_INCLUDE_FILE(smmintrin.h HAVE_SMMINTRIN_H "${SSE4_1_SWITCH}")
 IF(HAVE_SMMINTRIN_H)