Improvements to heap-memory and PSRAM handling

usermods/audioreactive/audio_reactive.cpp

@@ -224,8 +224,8 @@ void FFTcode(void * parameter)
   DEBUGSR_PRINT("FFT started on core: "); DEBUGSR_PRINTLN(xPortGetCoreID());
 
   // allocate FFT buffers on first call
-  if (vReal == nullptr) vReal = (float*) calloc(sizeof(float), samplesFFT);
-  if (vImag == nullptr) vImag = (float*) calloc(sizeof(float), samplesFFT);
+  if (vReal == nullptr) vReal = (float*) calloc(samplesFFT, sizeof(float));
+  if (vImag == nullptr) vImag = (float*) calloc(samplesFFT, sizeof(float));
   if ((vReal == nullptr) || (vImag == nullptr)) {
     // something went wrong
     if (vReal) free(vReal); vReal = nullptr;

wled00/FX.h

@@ -88,23 +88,22 @@ extern byte realtimeMode;           // used in getMappedPixelIndex()
 #endif
 #define FPS_CALC_SHIFT 7 // bit shift for fixed point math
 
-/* each segment uses 82 bytes of SRAM memory, so if you're application fails because of
-  insufficient memory, decreasing MAX_NUM_SEGMENTS may help */
+// heap memory limit for effects data, pixel buffers try to reserve it if PSRAM is available
 #ifdef ESP8266
   #define MAX_NUM_SEGMENTS  16
   /* How much data bytes all segments combined may allocate */
-  #define MAX_SEGMENT_DATA  5120
+  #define MAX_SEGMENT_DATA  (6*1024) // 6k by default
 #elif defined(CONFIG_IDF_TARGET_ESP32S2)
   #define MAX_NUM_SEGMENTS  20
-  #define MAX_SEGMENT_DATA  (MAX_NUM_SEGMENTS*512)  // 10k by default (S2 is short on free RAM)
+  #define MAX_SEGMENT_DATA  (20*1024) // 20k by default (S2 is short on free RAM), limit does not apply if PSRAM is available
 #else
   #define MAX_NUM_SEGMENTS  32  // warning: going beyond 32 may consume too much RAM for stable operation
-  #define MAX_SEGMENT_DATA  (MAX_NUM_SEGMENTS*1280) // 40k by default
+  #define MAX_SEGMENT_DATA  (64*1024) // 64k by default, limit does not apply if PSRAM is available
 #endif
 
 /* How much data bytes each segment should max allocate to leave enough space for other segments,
   assuming each segment uses the same amount of data. 256 for ESP8266, 640 for ESP32. */
-#define FAIR_DATA_PER_SEG (MAX_SEGMENT_DATA / WS2812FX::getMaxSegments())
+#define FAIR_DATA_PER_SEG (MAX_SEGMENT_DATA / MAX_NUM_SEGMENTS)
 
 #define MIN_SHOW_DELAY   (_frametime < 16 ? 8 : 15)
 
@@ -533,7 +532,6 @@ class Segment {
 
   protected:
 
-    inline static unsigned getUsedSegmentData()            { return Segment::_usedSegmentData; }
     inline static void     addUsedSegmentData(int len)     { Segment::_usedSegmentData += len; }
 
     inline uint32_t *getPixels() const                              { return pixels; }
@@ -600,8 +598,8 @@ class Segment {
     , _t(nullptr)
     {
       DEBUGFX_PRINTF_P(PSTR("-- Creating segment: %p [%d,%d:%d,%d]\n"), this, (int)start, (int)stop, (int)startY, (int)stopY);
-      // allocate render buffer (always entire segment)
-      pixels = static_cast<uint32_t*>(d_calloc(sizeof(uint32_t), length())); // error handling is also done in isActive()
+      // allocate render buffer (always entire segment), prefer PSRAM if DRAM is running low. Note: impact on FPS with PSRAM buffer is low (<2% with QSPI PSRAM)
+      pixels = static_cast<uint32_t*>(allocate_buffer(length() * sizeof(uint32_t), BFRALLOC_PREFER_PSRAM | BFRALLOC_NOBYTEACCESS | BFRALLOC_CLEAR));
       if (!pixels) {
         DEBUGFX_PRINTLN(F("!!! Not enough RAM for pixel buffer !!!"));
         extern byte errorFlag;
@@ -672,6 +670,7 @@ class Segment {
     inline uint16_t dataSize() const { return _dataLen; }
     bool allocateData(size_t len);  // allocates effect data buffer in heap and clears it
     void deallocateData();          // deallocates (frees) effect data buffer from heap
+    inline static unsigned getUsedSegmentData()            { return Segment::_usedSegmentData; }
     /**
       * Flags that before the next effect is calculated,
       * the internal segment state should be reset.

wled00/FX_fcn.cpp

@@ -68,10 +68,10 @@ Segment::Segment(const Segment &orig) {
   if (!stop) return;  // nothing to do if segment is inactive/invalid
   if (orig.pixels) {
     // allocate pixel buffer: prefer IRAM/PSRAM
-    pixels = static_cast<uint32_t*>(d_malloc(sizeof(uint32_t) * orig.length()));
+    pixels = static_cast<uint32_t*>(allocate_buffer(orig.length() * sizeof(uint32_t), BFRALLOC_PREFER_PSRAM | BFRALLOC_NOBYTEACCESS));
     if (pixels) {
       memcpy(pixels, orig.pixels, sizeof(uint32_t) * orig.length());
-      if (orig.name) { name = static_cast<char*>(d_malloc(strlen(orig.name)+1)); if (name) strcpy(name, orig.name); }
+      if (orig.name) { name = static_cast<char*>(allocate_buffer(strlen(orig.name)+1, BFRALLOC_PREFER_PSRAM)); if (name) strcpy(name, orig.name); }
       if (orig.data) { if (allocateData(orig._dataLen)) memcpy(data, orig.data, orig._dataLen); }
     } else {
       DEBUGFX_PRINTLN(F("!!! Not enough RAM for pixel buffer !!!"));
@@ -97,10 +97,10 @@ Segment& Segment::operator= (const Segment &orig) {
   //DEBUG_PRINTF_P(PSTR("-- Copying segment: %p -> %p\n"), &orig, this);
   if (this != &orig) {
     // clean destination
-    if (name) { d_free(name); name = nullptr; }
+    if (name) { p_free(name); name = nullptr; }
     if (_t) stopTransition(); // also erases _t
     deallocateData();
-    d_free(pixels);
+    p_free(pixels);
     // copy source
     memcpy((void*)this, (void*)&orig, sizeof(Segment));
     // erase pointers to allocated data
@@ -111,10 +111,10 @@ Segment& Segment::operator= (const Segment &orig) {
     // copy source data
     if (orig.pixels) {
       // allocate pixel buffer: prefer IRAM/PSRAM
-      pixels = static_cast<uint32_t*>(d_malloc(sizeof(uint32_t) * orig.length()));
+      pixels = static_cast<uint32_t*>(allocate_buffer(orig.length() * sizeof(uint32_t), BFRALLOC_PREFER_PSRAM | BFRALLOC_NOBYTEACCESS));
       if (pixels) {
         memcpy(pixels, orig.pixels, sizeof(uint32_t) * orig.length());
-        if (orig.name) { name = static_cast<char*>(d_malloc(strlen(orig.name)+1)); if (name) strcpy(name, orig.name); }
+        if (orig.name) { name = static_cast<char*>(allocate_buffer(strlen(orig.name)+1, BFRALLOC_PREFER_PSRAM)); if (name) strcpy(name, orig.name); }
         if (orig.data) { if (allocateData(orig._dataLen)) memcpy(data, orig.data, orig._dataLen); }
       } else {
         DEBUG_PRINTLN(F("!!! Not enough RAM for pixel buffer !!!"));
@@ -130,10 +130,10 @@ Segment& Segment::operator= (const Segment &orig) {
 Segment& Segment::operator= (Segment &&orig) noexcept {
   //DEBUG_PRINTF_P(PSTR("-- Moving segment: %p -> %p\n"), &orig, this);
   if (this != &orig) {
-    if (name) { d_free(name); name = nullptr; } // free old name
+    if (name) { p_free(name); name = nullptr; } // free old name
     if (_t) stopTransition(); // also erases _t
     deallocateData(); // free old runtime data
-    d_free(pixels);   // free old pixel buffer
+    p_free(pixels);   // free old pixel buffer
     // move source data
     memcpy((void*)this, (void*)&orig, sizeof(Segment));
     orig.name = nullptr;
@@ -147,35 +147,38 @@ Segment& Segment::operator= (Segment &&orig) noexcept {
 
 // allocates effect data buffer on heap and initialises (erases) it
 bool Segment::allocateData(size_t len) {
-  if (len == 0) return false; // nothing to do
-  if (data && _dataLen >= len) {          // already allocated enough (reduce fragmentation)
+  if (len == 0) return false;    // nothing to do
+  if (data && _dataLen >= len) { // already allocated enough (reduce fragmentation)
     if (call == 0) {
-      //DEBUG_PRINTF_P(PSTR("--   Clearing data (%d): %p\n"), len, this);
-      memset(data, 0, len);  // erase buffer if called during effect initialisation
+      if (_dataLen < FAIR_DATA_PER_SEG) { // segment data is small
+        //DEBUG_PRINTF_P(PSTR("--   Clearing data (%d): %p\n"), len, this);
+        memset(data, 0, len);  // erase buffer if called during effect initialisation
+        return true; // no need to reallocate
+      }
     }
-    return true;
+    else
+      return true;
   }
   //DEBUG_PRINTF_P(PSTR("--   Allocating data (%d): %p\n"), len, this);
+  // limit to MAX_SEGMENT_DATA if there is no PSRAM, otherwise prefer functionality over speed
+  #ifndef BOARD_HAS_PSRAM
   if (Segment::getUsedSegmentData() + len - _dataLen > MAX_SEGMENT_DATA) {
     // not enough memory
-    DEBUG_PRINTF_P(PSTR("!!! Not enough RAM: %d/%d !!!\n"), len, Segment::getUsedSegmentData());
+    DEBUG_PRINTF_P(PSTR("SegmentData limit reached: %d/%d\n"), len, Segment::getUsedSegmentData());
     errorFlag = ERR_NORAM;
     return false;
   }
-  // prefer DRAM over SPI RAM on ESP32 since it is slow
+  #endif
+
   if (data) {
-    data = (byte*)d_realloc_malloc(data, len); // realloc with malloc fallback
-    if (!data) {
-      data = nullptr;
-      Segment::addUsedSegmentData(-_dataLen); // subtract original buffer size
-      _dataLen = 0;   // reset data length
-    }
+    d_free(data); // free data and try to allocate again (segment buffer may be blocking contiguous heap)
+    Segment::addUsedSegmentData(-_dataLen); // subtract buffer size
   }
-  else data = (byte*)d_malloc(len);
+
+  data = static_cast<byte*>(allocate_buffer(len, BFRALLOC_PREFER_DRAM | BFRALLOC_CLEAR)); // prefer DRAM over PSRAM for speed
 
   if (data) {
-    memset(data, 0, len);  // erase buffer
-    Segment::addUsedSegmentData(len - _dataLen);
+    Segment::addUsedSegmentData(len);
     _dataLen = len;
     //DEBUG_PRINTF_P(PSTR("---  Allocated data (%p): %d/%d -> %p\n"), this, len, Segment::getUsedSegmentData(), data);
     return true;
@@ -209,7 +212,11 @@ void Segment::deallocateData() {
 void Segment::resetIfRequired() {
   if (!reset || !isActive()) return;
   //DEBUG_PRINTF_P(PSTR("-- Segment reset: %p\n"), this);
-  if (data && _dataLen > 0) memset(data, 0, _dataLen);  // prevent heap fragmentation (just erase buffer instead of deallocateData())
+  if (data && _dataLen > 0) {
+    if (_dataLen > FAIR_DATA_PER_SEG) deallocateData(); // do not keep large allocations
+    else memset(data, 0, _dataLen);  // can prevent heap fragmentation
+    DEBUG_PRINTF_P(PSTR("-- Segment %p reset, data cleared\n"), this);
+  }
   if (pixels) for (size_t i = 0; i < length(); i++) pixels[i] = BLACK; // clear pixel buffer
   next_time = 0; step = 0; call = 0; aux0 = 0; aux1 = 0;
   reset = false;
@@ -466,7 +473,7 @@ void Segment::setGeometry(uint16_t i1, uint16_t i2, uint8_t grp, uint8_t spc, ui
   if (length() != oldLength) {
     // allocate render buffer (always entire segment), prefer IRAM/PSRAM. Note: impact on FPS with PSRAM buffer is low (<2% with QSPI PSRAM) on S2/S3
     p_free(pixels);
-    pixels = static_cast<uint32_t*>(d_malloc(sizeof(uint32_t) * length()));
+    pixels = static_cast<uint32_t*>(allocate_buffer(length() * sizeof(uint32_t), BFRALLOC_PREFER_PSRAM | BFRALLOC_NOBYTEACCESS));
     if (!pixels) {
       DEBUGFX_PRINTLN(F("!!! Not enough RAM for pixel buffer !!!"));
       deallocateData();
@@ -581,8 +588,8 @@ Segment &Segment::setName(const char *newName) {
   if (newName) {
     const int newLen = min(strlen(newName), (size_t)WLED_MAX_SEGNAME_LEN);
     if (newLen) {
-      if (name) d_free(name); // free old name
-      name = static_cast<char*>(d_malloc(newLen+1));
+      if (name) p_free(name); // free old name
+      name = static_cast<char*>(allocate_buffer(newLen+1, BFRALLOC_PREFER_PSRAM));
       if (mode == FX_MODE_2DSCROLLTEXT) startTransition(strip.getTransition(), true); // if the name changes in scrolling text mode, we need to copy the segment for blending
       if (name) strlcpy(name, newName, newLen+1);
       return *this;
@@ -1177,7 +1184,10 @@ void WS2812FX::finalizeInit() {
     mem += bus.memUsage(Bus::isDigital(bus.type) && !Bus::is2Pin(bus.type) ? digitalCount++ : 0); // includes global buffer
     if (mem <= MAX_LED_MEMORY) {
       if (BusManager::add(bus) == -1) break;
-    } else DEBUG_PRINTF_P(PSTR("Out of LED memory! Bus %d (%d) #%u not created."), (int)bus.type, (int)bus.count, digitalCount);
+    } else {
+      errorFlag = ERR_NORAM_PX; // alert UI
+      DEBUG_PRINTF_P(PSTR("Out of LED memory! Bus %d (%d) #%u not created."), (int)bus.type, (int)bus.count, digitalCount);
+    }
   }
   busConfigs.clear();
   busConfigs.shrink_to_fit();
@@ -1196,7 +1206,7 @@ void WS2812FX::finalizeInit() {
     bus->begin();
     bus->setBrightness(bri);
   }
-  DEBUG_PRINTF_P(PSTR("Heap after buses: %d\n"), ESP.getFreeHeap());
+  DEBUG_PRINTF_P(PSTR("Heap after buses: %d\n"), getFreeHeapSize());
 
   Segment::maxWidth  = _length;
   Segment::maxHeight = 1;
@@ -1208,10 +1218,11 @@ void WS2812FX::finalizeInit() {
   deserializeMap();     // (re)load default ledmap (will also setUpMatrix() if ledmap does not exist)
 
   // allocate frame buffer after matrix has been set up (gaps!)
-  d_free(_pixels); // using realloc on large buffers can cause additional fragmentation instead of reducing it
-  _pixels = static_cast<uint32_t*>(d_malloc(getLengthTotal() * sizeof(uint32_t)));
+  p_free(_pixels); // using realloc on large buffers can cause additional fragmentation instead of reducing it
+  // use PSRAM if available: there is no measurable perfomance impact between PSRAM and DRAM on S2/S3 with QSPI PSRAM for this buffer
+  _pixels = static_cast<uint32_t*>(allocate_buffer(getLengthTotal() * sizeof(uint32_t), BFRALLOC_ENFORCE_PSRAM | BFRALLOC_NOBYTEACCESS | BFRALLOC_CLEAR));
   DEBUG_PRINTF_P(PSTR("strip buffer size: %uB\n"), getLengthTotal() * sizeof(uint32_t));
-  DEBUG_PRINTF_P(PSTR("Heap after strip init: %uB\n"), ESP.getFreeHeap());
+  DEBUG_PRINTF_P(PSTR("Heap after strip init: %uB\n"), getFreeHeapSize());
 }
 
 void WS2812FX::service() {
@@ -1611,7 +1622,11 @@ static uint8_t estimateCurrentAndLimitBri(uint8_t brightness, uint32_t *pixels)
 }
 
 void WS2812FX::show() {
-  if (!_pixels) return; // no pixels allocated, nothing to show
+  if (!_pixels) {
+    DEBUGFX_PRINTLN(F("Error: no _pixels!"));
+    errorFlag = ERR_NORAM;
+    return; // no pixels allocated, nothing to show
+  }
 
   unsigned long showNow = millis();
   size_t diff = showNow - _lastShow;
@@ -1621,7 +1636,7 @@ void WS2812FX::show() {
   // we need to keep track of each pixel's CCT when blending segments (if CCT is present)
   // and then set appropriate CCT from that pixel during paint (see below).
   if ((hasCCTBus() || correctWB) && !cctFromRgb)
-    _pixelCCT = static_cast<uint8_t*>(d_malloc(totalLen * sizeof(uint8_t))); // allocate CCT buffer if necessary
+    _pixelCCT = static_cast<uint8_t*>(allocate_buffer(totalLen * sizeof(uint8_t), BFRALLOC_PREFER_PSRAM)); // allocate CCT buffer if necessary, prefer PSRAM
   if (_pixelCCT) memset(_pixelCCT, 127, totalLen); // set neutral (50:50) CCT
 
   if (realtimeMode == REALTIME_MODE_INACTIVE || useMainSegmentOnly || realtimeOverride > REALTIME_OVERRIDE_NONE) {
@@ -1655,7 +1670,7 @@ void WS2812FX::show() {
   }
   Bus::setCCT(oldCCT);  // restore old CCT for ABL adjustments
 
-  d_free(_pixelCCT);
+  p_free(_pixelCCT);
   _pixelCCT = nullptr;
 
   // some buses send asynchronously and this method will return before

wled00/bus_manager.cpp

@@ -38,35 +38,29 @@ uint32_t colorBalanceFromKelvin(uint16_t kelvin, uint32_t rgb);
 uint8_t realtimeBroadcast(uint8_t type, IPAddress client, uint16_t length, const byte *buffer, uint8_t bri=255, bool isRGBW=false);
 
 //util.cpp
-// PSRAM allocation wrappers
-#if !defined(ESP8266) && !defined(CONFIG_IDF_TARGET_ESP32C3)
+// memory allocation wrappers
 extern "C" {
-  void *p_malloc(size_t);           // prefer PSRAM over DRAM
-  void *p_calloc(size_t, size_t);   // prefer PSRAM over DRAM
-  void *p_realloc(void *, size_t);  // prefer PSRAM over DRAM
-  void *p_realloc_malloc(void *ptr, size_t size); // realloc with malloc fallback, prefer PSRAM over DRAM
-  inline void p_free(void *ptr) { heap_caps_free(ptr); }
-  void *d_malloc(size_t);           // prefer DRAM over PSRAM
-  void *d_calloc(size_t, size_t);   // prefer DRAM over PSRAM
-  void *d_realloc(void *, size_t);  // prefer DRAM over PSRAM
-  void *d_realloc_malloc(void *ptr, size_t size); // realloc with malloc fallback, prefer DRAM over PSRAM
+  // prefer DRAM over PSRAM (if available) in d_ alloc functions
+  void *d_malloc(size_t);
+  void *d_calloc(size_t, size_t);
+  void *d_realloc_malloc(void *ptr, size_t size);
+  #ifndef ESP8266
   inline void d_free(void *ptr) { heap_caps_free(ptr); }
+  #else
+  inline void d_free(void *ptr) { free(ptr); }
+  #endif
+  #if defined(BOARD_HAS_PSRAM)
+  // prefer PSRAM over DRAM in p_ alloc functions
+  void *p_malloc(size_t);
+  void *p_calloc(size_t, size_t);
+  void *p_realloc_malloc(void *ptr, size_t size);
+  inline void p_free(void *ptr) { heap_caps_free(ptr); }
+  #else
+  #define p_malloc d_malloc
+  #define p_calloc d_calloc
+  #define p_free d_free
+  #endif
 }
-#else
-extern "C" {
-  void *realloc_malloc(void *ptr, size_t size);
-}
-#define p_malloc malloc
-#define p_calloc calloc
-#define p_realloc realloc
-#define p_realloc_malloc realloc_malloc
-#define p_free free
-#define d_malloc malloc
-#define d_calloc calloc
-#define d_realloc realloc
-#define d_realloc_malloc realloc_malloc
-#define d_free free
-#endif
 
 //color mangling macros
 #define RGBW32(r,g,b,w) (uint32_t((byte(w) << 24) | (byte(r) << 16) | (byte(g) << 8) | (byte(b))))

wled00/cfg.cpp

@@ -201,7 +201,7 @@ bool deserializeConfig(JsonObject doc, bool fromFS) {
   }
   #endif
 
-  DEBUG_PRINTF_P(PSTR("Heap before buses: %d\n"), ESP.getFreeHeap());
+  DEBUG_PRINTF_P(PSTR("Heap before buses: %d\n"), getFreeHeapSize());
   JsonArray ins = hw_led["ins"];
   if (!ins.isNull()) {
     int s = 0;  // bus iterator

wled00/const.h

@@ -546,8 +546,21 @@ static_assert(WLED_MAX_BUSSES <= 32, "WLED_MAX_BUSSES exceeds hard limit");
   #endif
 #endif
 
-// minimum heap size required to process web requests
-#define MIN_HEAP_SIZE 8192
+// minimum heap size required to process web requests: try to keep free heap above this value
+#ifdef ESP8266
+  #define MIN_HEAP_SIZE (9*1024)
+#else
+  #define MIN_HEAP_SIZE (15*1024) // WLED allocation functions (util.cpp) try to keep this much contiguous heap free for other tasks
+#endif
+// threshold for PSRAM use: if heap is running low, requests to allocate_buffer(prefer DRAM) above PSRAM_THRESHOLD may be put in PSRAM
+// if heap is depleted, PSRAM will be used regardless of threshold
+#if defined(CONFIG_IDF_TARGET_ESP32S3)
+  #define PSRAM_THRESHOLD (12*1024) // S3 has plenty of DRAM
+#elif defined(CONFIG_IDF_TARGET_ESP32)
+  #define PSRAM_THRESHOLD (5*1024)
+#else
+  #define PSRAM_THRESHOLD (2*1024) // S2 does not have a lot of RAM. C3 and ESP8266 do not support PSRAM: the value is not used
+#endif
 
 // Web server limits
 #ifdef ESP8266

wled00/data/settings_leds.htm

@@ -195,7 +195,6 @@
 			if (isAna(t)) return 5;	// analog
 			let len = parseInt(d.getElementsByName("LC"+n)[0].value);
 			len += parseInt(d.getElementsByName("SL"+n)[0].value); // skipped LEDs are allocated too
-			let dbl = 0;
 			let ch = 3*hasRGB(t) + hasW(t) + hasCCT(t);
 			let mul = 1;
 			if (isDig(t)) {
@@ -207,7 +206,7 @@
 					mul = 2;
 				}
 			}
-			return len * ch * mul + dbl;
+			return len * ch * mul + len * 4; // add 4 bytes per LED for segment buffer (TODO: how to account for global buffer?)
 		}
 
 		function UI(change=false)