From b52fde7c0e57b4965e3b87d9cc2d98611064b7e5 Mon Sep 17 00:00:00 2001
From: Chris Robinson <chris.kcat@gmail.com>
Date: Sat, 2 May 2020 20:27:05 -0700
Subject: Simplify SIMD linear resampler loop count

---
 alc/mixer/mixer_neon.cpp  | 7 +++----
 alc/mixer/mixer_sse2.cpp  | 7 +++----
 alc/mixer/mixer_sse41.cpp | 7 +++----
 3 files changed, 9 insertions(+), 12 deletions(-)

diff --git a/alc/mixer/mixer_neon.cpp b/alc/mixer/mixer_neon.cpp
index 83388b68..2cf7e2e8 100644
--- a/alc/mixer/mixer_neon.cpp
+++ b/alc/mixer/mixer_neon.cpp
@@ -64,8 +64,7 @@ const float *Resample_<LerpTag,NEONTag>(const InterpState*, const float *RESTRIC
     pos4 = vld1q_s32(reinterpret_cast<int*>(pos_));
 
     auto dst_iter = dst.begin();
-    const auto aligned_end = (dst.size()&~3u) + dst_iter;
-    while(dst_iter != aligned_end)
+    for(size_t todo{dst.size()>>2};todo;--todo)
     {
         const int pos0{vgetq_lane_s32(pos4, 0)};
         const int pos1{vgetq_lane_s32(pos4, 1)};
@@ -87,7 +86,7 @@ const float *Resample_<LerpTag,NEONTag>(const InterpState*, const float *RESTRIC
         frac4 = vandq_s32(frac4, fracMask4);
     }
 
-    if(dst_iter != dst.end())
+    if(size_t todo{dst.size()&3})
     {
         src += static_cast<ALuint>(vgetq_lane_s32(pos4, 0));
         frac = static_cast<ALuint>(vgetq_lane_s32(frac4, 0));
@@ -98,7 +97,7 @@ const float *Resample_<LerpTag,NEONTag>(const InterpState*, const float *RESTRIC
             frac += increment;
             src  += frac>>FRACTIONBITS;
             frac &= FRACTIONMASK;
-        } while(dst_iter != dst.end());
+        } while(--todo);
     }
     return dst.data();
 }
diff --git a/alc/mixer/mixer_sse2.cpp b/alc/mixer/mixer_sse2.cpp
index 3558b55e..088284a7 100644
--- a/alc/mixer/mixer_sse2.cpp
+++ b/alc/mixer/mixer_sse2.cpp
@@ -46,8 +46,7 @@ const float *Resample_<LerpTag,SSE2Tag>(const InterpState*, const float *RESTRIC
         static_cast<int>(pos_[2]), static_cast<int>(pos_[3]))};
 
     auto dst_iter = dst.begin();
-    const auto aligned_end = (dst.size()&~3u) + dst_iter;
-    while(dst_iter != aligned_end)
+    for(size_t todo{dst.size()>>2};todo;--todo)
     {
         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)))};
@@ -69,7 +68,7 @@ const float *Resample_<LerpTag,SSE2Tag>(const InterpState*, const float *RESTRIC
         frac4 = _mm_and_si128(frac4, fracMask4);
     }
 
-    if(dst_iter != dst.end())
+    if(size_t todo{dst.size()&3})
     {
         src += static_cast<ALuint>(_mm_cvtsi128_si32(pos4));
         frac = static_cast<ALuint>(_mm_cvtsi128_si32(frac4));
@@ -80,7 +79,7 @@ const float *Resample_<LerpTag,SSE2Tag>(const InterpState*, const float *RESTRIC
             frac += increment;
             src  += frac>>FRACTIONBITS;
             frac &= FRACTIONMASK;
-        } while(dst_iter != dst.end());
+        } while(--todo);
     }
     return dst.data();
 }
diff --git a/alc/mixer/mixer_sse41.cpp b/alc/mixer/mixer_sse41.cpp
index b5aebe7d..f18cd6b4 100644
--- a/alc/mixer/mixer_sse41.cpp
+++ b/alc/mixer/mixer_sse41.cpp
@@ -47,8 +47,7 @@ const float *Resample_<LerpTag,SSE4Tag>(const InterpState*, const float *RESTRIC
         static_cast<int>(pos_[2]), static_cast<int>(pos_[3]))};
 
     auto dst_iter = dst.begin();
-    const auto aligned_end = (dst.size()&~3u) + dst_iter;
-    while(dst_iter != aligned_end)
+    for(size_t todo{dst.size()>>2};todo;--todo)
     {
         const int pos0{_mm_extract_epi32(pos4, 0)};
         const int pos1{_mm_extract_epi32(pos4, 1)};
@@ -70,7 +69,7 @@ const float *Resample_<LerpTag,SSE4Tag>(const InterpState*, const float *RESTRIC
         frac4 = _mm_and_si128(frac4, fracMask4);
     }
 
-    if(dst_iter != dst.end())
+    if(size_t todo{dst.size()&3})
     {
         /* NOTE: These four elements represent the position *after* the last
          * four samples, so the lowest element is the next position to
@@ -85,7 +84,7 @@ const float *Resample_<LerpTag,SSE4Tag>(const InterpState*, const float *RESTRIC
             frac += increment;
             src  += frac>>FRACTIONBITS;
             frac &= FRACTIONMASK;
-        } while(dst_iter != dst.end());
+        } while(--todo);
     }
     return dst.data();
 }
-- 
cgit v1.2.3