Slightly restructure some loops

3 files changed, 34 insertions, 39 deletions

diff --git a/Alc/mixer/mixer_neon.c b/Alc/mixer/mixer_neon.c
index 4feb431d..db61fd41 100644
--- a/Alc/mixer/mixer_neon.c
+++ b/Alc/mixer/mixer_neon.c
@@ -19,18 +19,17 @@ const ALfloat *Resample_lerp_Neon(const InterpState* UNUSED(state),
     const int32x4_t fracMask4 = vdupq_n_s32(FRACTIONMASK);
     alignas(16) ALint pos_[4];
     alignas(16) ALsizei frac_[4];
-    int32x4_t pos4;
-    int32x4_t frac4;
-    ALsizei i;
+    int32x4_t pos4, frac4;
+    ALsizei todo, pos, i;
 
     ASSUME(numsamples > 0);
 
     InitiatePositionArrays(frac, increment, frac_, pos_, 4);
-
     frac4 = vld1q_s32(frac_);
     pos4 = vld1q_s32(pos_);
 
-    for(i = 0;numsamples-i > 3;i += 4)
+    todo = numsamples & ~3;
+    for(i = 0;i < todo;i += 4)
     {
         const int pos0 = vgetq_lane_s32(pos4, 0);
         const int pos1 = vgetq_lane_s32(pos4, 1);
@@ -51,21 +50,19 @@ const ALfloat *Resample_lerp_Neon(const InterpState* UNUSED(state),
         frac4 = vandq_s32(frac4, fracMask4);
     }
 
-    if(i < numsamples)
+    /* NOTE: These four elements represent the position *after* the last four
+     * samples, so the lowest element is the next position to resample.
+     */
+    pos = vgetq_lane_s32(pos4, 0);
+    frac = vgetq_lane_s32(frac4, 0);
+
+    for(;i < numsamples;++i)
     {
-        /* NOTE: These four elements represent the position *after* the last
-         * four samples, so the lowest element is the next position to
-         * resample.
-         */
-        int pos = vgetq_lane_s32(pos4, 0);
-        frac = vgetq_lane_s32(frac4, 0);
-        do {
-            dst[i] = lerp(src[pos], src[pos+1], frac * (1.0f/FRACTIONONE));
-
-            frac += increment;
-            pos  += frac>>FRACTIONBITS;
-            frac &= FRACTIONMASK;
-        } while(++i < numsamples);
+        dst[i] = lerp(src[pos], src[pos+1], frac * (1.0f/FRACTIONONE));
+
+        frac += increment;
+        pos  += frac>>FRACTIONBITS;
+        frac &= FRACTIONMASK;
     }
     return dst;
 }
diff --git a/Alc/mixer/mixer_sse2.c b/Alc/mixer/mixer_sse2.c
index 83aaf7f2..629e0ec7 100644
--- a/Alc/mixer/mixer_sse2.c
+++ b/Alc/mixer/mixer_sse2.c
@@ -34,20 +34,19 @@ const ALfloat *Resample_lerp_SSE2(const InterpState* UNUSED(state),
     const __m128i increment4 = _mm_set1_epi32(increment*4);
     const __m128 fracOne4 = _mm_set1_ps(1.0f/FRACTIONONE);
     const __m128i fracMask4 = _mm_set1_epi32(FRACTIONMASK);
-    union { alignas(16) ALint i[4]; float f[4]; } pos_;
-    union { alignas(16) ALsizei i[4]; float f[4]; } frac_;
+    ALint pos_[4];
+    ALsizei frac_[4];
     __m128i frac4, pos4;
-    ALint pos;
-    ALsizei i;
+    ALsizei todo, pos, i;
 
     ASSUME(numsamples > 0);
 
-    InitiatePositionArrays(frac, increment, frac_.i, pos_.i, 4);
+    InitiatePositionArrays(frac, increment, frac_, pos_, 4);
+    frac4 = _mm_setr_epi32(frac_[0], frac_[1], frac_[2], frac_[3]);
+    pos4 = _mm_setr_epi32(pos_[0], pos_[1], pos_[2], pos_[3]);
 
-    frac4 = _mm_castps_si128(_mm_load_ps(frac_.f));
-    pos4 = _mm_castps_si128(_mm_load_ps(pos_.f));
-
-    for(i = 0;numsamples-i > 3;i += 4)
+    todo = numsamples & ~3;
+    for(i = 0;i < todo;i += 4)
     {
         const int pos0 = _mm_cvtsi128_si32(_mm_shuffle_epi32(pos4, _MM_SHUFFLE(0, 0, 0, 0)));
         const int pos1 = _mm_cvtsi128_si32(_mm_shuffle_epi32(pos4, _MM_SHUFFLE(1, 1, 1, 1)));
@@ -74,7 +73,7 @@ const ALfloat *Resample_lerp_SSE2(const InterpState* UNUSED(state),
     pos = _mm_cvtsi128_si32(pos4);
     frac = _mm_cvtsi128_si32(frac4);
 
-    for(;i < numsamples;i++)
+    for(;i < numsamples;++i)
     {
         dst[i] = lerp(src[pos], src[pos+1], frac * (1.0f/FRACTIONONE));
 
diff --git a/Alc/mixer/mixer_sse41.c b/Alc/mixer/mixer_sse41.c
index 396b7c7c..128acadf 100644
--- a/Alc/mixer/mixer_sse41.c
+++ b/Alc/mixer/mixer_sse41.c
@@ -35,20 +35,19 @@ const ALfloat *Resample_lerp_SSE41(const InterpState* UNUSED(state),
     const __m128i increment4 = _mm_set1_epi32(increment*4);
     const __m128 fracOne4 = _mm_set1_ps(1.0f/FRACTIONONE);
     const __m128i fracMask4 = _mm_set1_epi32(FRACTIONMASK);
-    union { alignas(16) ALint i[4]; float f[4]; } pos_;
-    union { alignas(16) ALsizei i[4]; float f[4]; } frac_;
+    ALint pos_[4];
+    ALsizei frac_[4];
     __m128i frac4, pos4;
-    ALint pos;
-    ALsizei i;
+    ALsizei todo, pos, i;
 
     ASSUME(numsamples > 0);
 
-    InitiatePositionArrays(frac, increment, frac_.i, pos_.i, 4);
+    InitiatePositionArrays(frac, increment, frac_, pos_, 4);
+    frac4 = _mm_setr_epi32(frac_[0], frac_[1], frac_[2], frac_[3]);
+    pos4 = _mm_setr_epi32(pos_[0], pos_[1], pos_[2], pos_[3]);
 
-    frac4 = _mm_castps_si128(_mm_load_ps(frac_.f));
-    pos4 = _mm_castps_si128(_mm_load_ps(pos_.f));
-
-    for(i = 0;numsamples-i > 3;i += 4)
+    todo = numsamples & ~3;
+    for(i = 0;i < todo;i += 4)
     {
         const int pos0 = _mm_extract_epi32(pos4, 0);
         const int pos1 = _mm_extract_epi32(pos4, 1);
@@ -75,7 +74,7 @@ const ALfloat *Resample_lerp_SSE41(const InterpState* UNUSED(state),
     pos = _mm_cvtsi128_si32(pos4);
     frac = _mm_cvtsi128_si32(frac4);
 
-    for(;i < numsamples;i++)
+    for(;i < numsamples;++i)
     {
         dst[i] = lerp(src[pos], src[pos+1], frac * (1.0f/FRACTIONONE));