From 8bbd793ae67c36a16ae3a2a59fed19b4f23a77a9 Mon Sep 17 00:00:00 2001
From: - <>
Date: Thu, 23 Jun 2022 01:32:08 -0400
Subject: [PATCH] rounding through intrinsics

---
 src/cubeb_mixer.cpp  | 9 ++++++---
 src/cubeb_sndio.c    | 4 +++-
 src/cubeb_wasapi.cpp | 4 +++-
 3 files changed, 12 insertions(+), 5 deletions(-)

diff --git a/src/cubeb_mixer.cpp b/src/cubeb_mixer.cpp
index 74bab713..0a3366ed 100644
--- a/src/cubeb_mixer.cpp
+++ b/src/cubeb_mixer.cpp
@@ -17,6 +17,7 @@
 #include <climits>
 #include <cmath>
 #include <cstdlib>
+#include <emmintrin.h>
 #include <memory>
 #include <type_traits>
 
@@ -377,22 +378,24 @@ MixerContext::init()
 
       for (uint32_t j = 0; j < _in_ch_count; j++) {
         double target = _matrix[i][j] * 32768 + rem;
-        int value = lrintf(target);
+        int value = _mm_cvtsd_si32(_mm_set_sd(target)); // round double to int
         rem += target - value;
         sum += std::abs(value);
       }
       maxsum = std::max(maxsum, sum);
     }
-    if (maxsum > 32768) {
+    if (maxsum > 32768) { //> or >=?
       _clipping = true;
     }
   }
 
   // FIXME quantize for integers
+  // June 23, 2022: I made a change but have no idea what was wanted
   for (uint32_t i = 0; i < CHANNELS_MAX; i++) {
     int ch_in = 0;
     for (uint32_t j = 0; j < CHANNELS_MAX; j++) {
-      _matrix32[i][j] = lrintf(_matrix[i][j] * 32768);
+      // round double to int
+      _matrix32[i][j] = _mm_cvtsd_si32(_mm_set_sd(_matrix[i][j] * 32768));
       if (_matrix[i][j]) {
         _matrix_ch[i][++ch_in] = j;
       }
diff --git a/src/cubeb_sndio.c b/src/cubeb_sndio.c
index 5e11725e..d928d0c3 100644
--- a/src/cubeb_sndio.c
+++ b/src/cubeb_sndio.c
@@ -16,6 +16,7 @@
 #include <stdbool.h>
 #include <stdio.h>
 #include <stdlib.h>
+#include <xmmintrin.h>
 
 #if defined(CUBEB_SNDIO_DEBUG)
 #define DPR(...) fprintf(stderr, __VA_ARGS__);
@@ -106,7 +107,8 @@ float_to_s16(void * ptr, long nsamp, float volume)
   int s;
 
   while (nsamp-- > 0) {
-    s = lrintf(*(src++) * mult);
+    // round float to int
+    s = _mm_cvt_ss2si(_mm_set_ss(*(src++) * mult));
     if (s < -32768)
       s = -32768;
     else if (s > 32767)
diff --git a/src/cubeb_wasapi.cpp b/src/cubeb_wasapi.cpp
index 039ad76c..1e16d1fa 100644
--- a/src/cubeb_wasapi.cpp
+++ b/src/cubeb_wasapi.cpp
@@ -24,6 +24,7 @@
 #include <vector>
 #include <windef.h>
 #include <windows.h>
+#include <xmmintrin.h>
 
 #include "cubeb-internal.h"
 #include "cubeb/cubeb.h"
@@ -938,7 +939,8 @@ refill(cubeb_stream * stm, void * input_buffer, long input_frames_count,
       case CUBEB_SAMPLE_S16NE: {
         short * buf = static_cast<short *>(dest);
         for (long i = 0; i < out_samples; ++i) {
-          buf[i] = static_cast<short>(static_cast<float>(buf[i]) * volume);
+          // round float to int
+          buf[i] = _mm_cvt_ss2si(_mm_set_ss(buf[i] * volume));
         }
         break;
       }