Fixed bug 3876 - Resampling of certain sounds adds heavy distortion

Simon Hug

Patch that adds [-1, 1] clamping to the scalar audio type conversions.

This may come from the SDL_Convert_F32_to_X_Scalar functions. They don't clamp the float value to [-1, 1] and when they cast it to the target integer it may be too large or too small for the type and get truncated, causing horrible noise.

The attached patch throws clamping in, but I don't know if that's the preferred way to fix this. For x86 (without SSE) the compiler (I tested MSVC) seems to throw a horrible amount of x87 code in it. It's a bit better with SSE, but probably still quite the performance hit. And SSE2 uses a branchless approach with maxss and minss.

src/audio/SDL_audiotypecvt.c

@@ -170,7 +170,14 @@ SDL_Convert_F32_to_S8_Scalar(SDL_AudioCVT *cvt, SDL_AudioFormat format)
     LOG_DEBUG_CONVERT("AUDIO_F32", "AUDIO_S8");
 
     for (i = cvt->len_cvt / sizeof (float); i; --i, ++src, ++dst) {
-        *dst = (Sint8) (*src * 127.0f);
+        const float sample = *src;
+        if (sample > 1.0f) {
+            *dst = 127;
+        } else if (sample < -1.0f) {
+            *dst = -127;
+        } else {
+            *dst = (Sint8)(sample * 127.0f);
+        }
     }
 
     cvt->len_cvt /= 4;
@@ -189,7 +196,14 @@ SDL_Convert_F32_to_U8_Scalar(SDL_AudioCVT *cvt, SDL_AudioFormat format)
     LOG_DEBUG_CONVERT("AUDIO_F32", "AUDIO_U8");
 
     for (i = cvt->len_cvt / sizeof (float); i; --i, ++src, ++dst) {
-        *dst = (Uint8) ((*src + 1.0f) * 127.0f);
+        const float sample = *src;
+        if (sample > 1.0f) {
+            *dst = 255;
+        } else if (sample < -1.0f) {
+            *dst = 0;
+        } else {
+            *dst = (Uint8)((sample + 1.0f) * 127.0f);
+        }
     }
 
     cvt->len_cvt /= 4;
@@ -208,7 +222,14 @@ SDL_Convert_F32_to_S16_Scalar(SDL_AudioCVT *cvt, SDL_AudioFormat format)
     LOG_DEBUG_CONVERT("AUDIO_F32", "AUDIO_S16");
 
     for (i = cvt->len_cvt / sizeof (float); i; --i, ++src, ++dst) {
-        *dst = (Sint16) (*src * 32767.0f);
+        const float sample = *src;
+        if (sample > 1.0f) {
+            *dst = 32767;
+        } else if (sample < -1.0f) {
+            *dst = -32767;
+        } else {
+            *dst = (Sint16)(sample * 32767.0f);
+        }
     }
 
     cvt->len_cvt /= 2;
@@ -227,7 +248,14 @@ SDL_Convert_F32_to_U16_Scalar(SDL_AudioCVT *cvt, SDL_AudioFormat format)
     LOG_DEBUG_CONVERT("AUDIO_F32", "AUDIO_U16");
 
     for (i = cvt->len_cvt / sizeof (float); i; --i, ++src, ++dst) {
-        *dst = (Uint16) ((*src + 1.0f) * 32767.0f);
+        const float sample = *src;
+        if (sample > 1.0f) {
+            *dst = 65534;
+        } else if (sample < -1.0f) {
+            *dst = 0;
+        } else {
+            *dst = (Uint16)((sample + 1.0f) * 32767.0f);
+        }
     }
 
     cvt->len_cvt /= 2;
@@ -246,7 +274,14 @@ SDL_Convert_F32_to_S32_Scalar(SDL_AudioCVT *cvt, SDL_AudioFormat format)
     LOG_DEBUG_CONVERT("AUDIO_F32", "AUDIO_S32");
 
     for (i = cvt->len_cvt / sizeof (float); i; --i, ++src, ++dst) {
-        *dst = (Sint32) (((double) *src) * 2147483647.0);
+        const float sample = *src;
+        if (sample > 1.0f) {
+            *dst = 2147483647;
+        } else if (sample < -1.0f) {
+            *dst = -2147483647;
+        } else {
+            *dst = (Sint32)((double)sample * 2147483647.0);
+        }
     }
 
     if (cvt->filters[++cvt->filter_index]) {