add neopixel_write implementation, work ok

Author: hathach

ports/nrf/common-hal/neopixel_write/__init__.c

@@ -27,6 +27,238 @@
 #include "py/mphal.h"
 #include "shared-bindings/neopixel_write/__init__.h"
 
-void common_hal_neopixel_write(const digitalio_digitalinout_obj_t* digitalinout, uint8_t *pixels, uint32_t numBytes) {
+// https://github.com/adafruit/Adafruit_NeoPixel/blob/master/Adafruit_NeoPixel.cpp
+// [[[Begin of the Neopixel NRF52 EasyDMA implementation
+//                                    by the Hackerspace San Salvador]]]
+// This technique uses the PWM peripheral on the NRF52. The PWM uses the
+// EasyDMA feature included on the chip. This technique loads the duty
+// cycle configuration for each cycle when the PWM is enabled. For this
+// to work we need to store a 16 bit configuration for each bit of the
+// RGB(W) values in the pixel buffer.
+// Comparator values for the PWM were hand picked and are guaranteed to
+// be 100% organic to preserve freshness and high accuracy. Current
+// parameters are:
+//   * PWM Clock: 16Mhz
+//   * Minimum step time: 62.5ns
+//   * Time for zero in high (T0H): 0.31ms
+//   * Time for one in high (T1H): 0.75ms
+//   * Cycle time:  1.25us
+//   * Frequency: 800Khz
+// For 400Khz we just double the calculated times.
+// ---------- BEGIN Constants for the EasyDMA implementation -----------
+// The PWM starts the duty cycle in LOW. To start with HIGH we
+// need to set the 15th bit on each register.
 
+// WS2812 (rev A) timing is 0.35 and 0.7us
+//#define MAGIC_T0H               5UL | (0x8000) // 0.3125us
+//#define MAGIC_T1H              12UL | (0x8000) // 0.75us
+
+// WS2812B (rev B) timing is 0.4 and 0.8 us
+#define MAGIC_T0H               6UL | (0x8000) // 0.375us
+#define MAGIC_T1H              13UL | (0x8000) // 0.8125us
+#define CTOPVAL                20UL            // 1.25us
+
+// ---------- END Constants for the EasyDMA implementation -------------
+//
+// If there is no device available an alternative cycle-counter
+// implementation is tried.
+// The nRF52832 runs with a fixed clock of 64Mhz. The alternative
+// implementation is the same as the one used for the Teensy 3.0/1/2 but
+// with the Nordic SDK HAL & registers syntax.
+// The number of cycles was hand picked and is guaranteed to be 100%
+// organic to preserve freshness and high accuracy.
+// ---------- BEGIN Constants for cycle counter implementation ---------
+#define CYCLES_800_T0H  18  // ~0.36 uS
+#define CYCLES_800_T1H  41  // ~0.76 uS
+#define CYCLES_800      71  // ~1.25 uS
+
+// ---------- END of Constants for cycle counter implementation --------
+
+// find a free PWM device, which is not enabled and has no connected pins
+static NRF_PWM_Type* find_free_pwm (void) {
+    NRF_PWM_Type* PWM[3] = { NRF_PWM0, NRF_PWM1, NRF_PWM2 };
+
+    for ( int device = 0; device < 3; device++ ) {
+        if ( (PWM[device]->ENABLE == 0) && (PWM[device]->PSEL.OUT[0] & PWM_PSEL_OUT_CONNECT_Msk)
+                && (PWM[device]->PSEL.OUT[1] & PWM_PSEL_OUT_CONNECT_Msk)
+                && (PWM[device]->PSEL.OUT[2] & PWM_PSEL_OUT_CONNECT_Msk)
+                && (PWM[device]->PSEL.OUT[3] & PWM_PSEL_OUT_CONNECT_Msk) ) {
+            return PWM[device];
+        }
+    }
+
+    return NULL;
+}
+
+void common_hal_neopixel_write (const digitalio_digitalinout_obj_t* digitalinout, uint8_t *pixels, uint32_t numBytes) {
+    // To support both the SoftDevice + Neopixels we use the EasyDMA
+    // feature from the NRF25. However this technique implies to
+    // generate a pattern and store it on the memory. The actual
+    // memory used in bytes corresponds to the following formula:
+    //              totalMem = numBytes*8*2+(2*2)
+    // The two additional bytes at the end are needed to reset the
+    // sequence.
+    //
+    // If there is not enough memory, we will fall back to cycle counter
+    // using DWT
+    uint32_t pattern_size = numBytes * 8 * sizeof(uint16_t) + 2 * sizeof(uint16_t);
+    uint16_t* pixels_pattern = NULL;
+
+    NRF_PWM_Type* pwm = find_free_pwm();
+
+    // only malloc if there is PWM device available
+    if ( pwm != NULL ) {
+        pixels_pattern = (uint16_t *) m_malloc(pattern_size, false);
+    }
+
+    // Use the identified device to choose the implementation
+    // If a PWM device is available use DMA
+    if ( (pixels_pattern != NULL) && (pwm != NULL) ) {
+        uint16_t pos = 0;  // bit position
+
+        for ( uint16_t n = 0; n < numBytes; n++ ) {
+            uint8_t pix = pixels[n];
+
+            for ( uint8_t mask = 0x80, i = 0; mask > 0; mask >>= 1, i++ ) {
+                pixels_pattern[pos] = (pix & mask) ? MAGIC_T1H : MAGIC_T0H;
+                pos++;
+            }
+        }
+
+        // Zero padding to indicate the end of sequence
+        pixels_pattern[++pos] = 0 | (0x8000);  // Seq end
+        pixels_pattern[++pos] = 0 | (0x8000);  // Seq end
+
+        // Set the wave mode to count UP
+        pwm->MODE = (PWM_MODE_UPDOWN_Up << PWM_MODE_UPDOWN_Pos);
+
+        // Set the PWM to use the 16MHz clock
+        pwm->PRESCALER = (PWM_PRESCALER_PRESCALER_DIV_1 << PWM_PRESCALER_PRESCALER_Pos);
+
+        // Setting of the maximum count
+        // but keeping it on 16Mhz allows for more granularity just
+        // in case someone wants to do more fine-tuning of the timing.
+        pwm->COUNTERTOP = (CTOPVAL << PWM_COUNTERTOP_COUNTERTOP_Pos);
+
+        // Disable loops, we want the sequence to repeat only once
+        pwm->LOOP = (PWM_LOOP_CNT_Disabled << PWM_LOOP_CNT_Pos);
+
+        // On the "Common" setting the PWM uses the same pattern for the
+        // for supported sequences. The pattern is stored on half-word
+        // of 16bits
+        pwm->DECODER = (PWM_DECODER_LOAD_Common << PWM_DECODER_LOAD_Pos)
+                | (PWM_DECODER_MODE_RefreshCount << PWM_DECODER_MODE_Pos);
+
+        // Pointer to the memory storing the patter
+        pwm->SEQ[0].PTR = (uint32_t) (pixels_pattern) << PWM_SEQ_PTR_PTR_Pos;
+
+        // Calculation of the number of steps loaded from memory.
+        pwm->SEQ[0].CNT = (pattern_size / sizeof(uint16_t)) << PWM_SEQ_CNT_CNT_Pos;
+
+        // The following settings are ignored with the current config.
+        pwm->SEQ[0].REFRESH = 0;
+        pwm->SEQ[0].ENDDELAY = 0;
+
+        // The Neopixel implementation is a blocking algorithm. DMA
+        // allows for non-blocking operation. To "simulate" a blocking
+        // operation we enable the interruption for the end of sequence
+        // and block the execution thread until the event flag is set by
+        // the peripheral.
+        //    pwm->INTEN |= (PWM_INTEN_SEQEND0_Enabled<<PWM_INTEN_SEQEND0_Pos);
+
+        // PSEL must be configured before enabling PWM
+        pwm->PSEL.OUT[0] = ( digitalinout->pin->port*32 + digitalinout->pin->pin );
+
+        // Enable the PWM
+        pwm->ENABLE = 1;
+
+        // After all of this and many hours of reading the documentation
+        // we are ready to start the sequence...
+        pwm->EVENTS_SEQEND[0] = 0;
+        pwm->TASKS_SEQSTART[0] = 1;
+
+        // But we have to wait for the flag to be set.
+        while ( !pwm->EVENTS_SEQEND[0] ) {
+//#ifdef MICROPY_VM_HOOK_LOOP
+//            MICROPY_VM_HOOK_LOOP
+//#endif
+        }
+
+        // Before leave we clear the flag for the event.
+        pwm->EVENTS_SEQEND[0] = 0;
+
+        // We need to disable the device and disconnect
+        // all the outputs before leave or the device will not
+        // be selected on the next call.
+        // TODO: Check if disabling the device causes performance issues.
+        pwm->ENABLE = 0;
+        pwm->PSEL.OUT[0] = 0xFFFFFFFFUL;
+
+        m_free(pixels_pattern);
+    } // End of DMA implementation
+    // ---------------------------------------------------------------------
+    else {
+        // Fall back to DWT
+        // If you are using the Bluetooth SoftDevice we advise you to not disable
+        // the interrupts. Disabling the interrupts even for short periods of time
+        // causes the SoftDevice to stop working.
+        // Disable the interrupts only in cases where you need high performance for
+        // the LEDs and if you are not using the EasyDMA feature.
+        __disable_irq();
+
+        NRF_GPIO_Type* port = ( digitalinout->pin->port ? NRF_P1 : NRF_P0 );
+        uint32_t pinMask = ( 1UL << digitalinout->pin->pin );
+
+        uint32_t CYCLES_X00 = CYCLES_800;
+        uint32_t CYCLES_X00_T1H = CYCLES_800_T1H;
+        uint32_t CYCLES_X00_T0H = CYCLES_800_T0H;
+
+        // Enable DWT in debug core
+        CoreDebug->DEMCR |= CoreDebug_DEMCR_TRCENA_Msk;
+        DWT->CTRL |= DWT_CTRL_CYCCNTENA_Msk;
+
+        // Tries to re-send the frame if is interrupted by the SoftDevice.
+        while ( 1 ) {
+            uint8_t *p = pixels;
+
+            uint32_t cycStart = DWT->CYCCNT;
+            uint32_t cyc = 0;
+
+            for ( uint16_t n = 0; n < numBytes; n++ ) {
+                uint8_t pix = *p++;
+
+                for ( uint8_t mask = 0x80; mask; mask >>= 1 ) {
+                    while ( DWT->CYCCNT - cyc < CYCLES_X00 )
+                        ;
+                    cyc = DWT->CYCCNT;
+
+                    port->OUTSET |= pinMask;
+
+                    if ( pix & mask ) {
+                        while ( DWT->CYCCNT - cyc < CYCLES_X00_T1H )
+                            ;
+                    } else {
+                        while ( DWT->CYCCNT - cyc < CYCLES_X00_T0H )
+                            ;
+                    }
+
+                    port->OUTCLR |= pinMask;
+                }
+            }
+            while ( DWT->CYCCNT - cyc < CYCLES_X00 )
+                ;
+
+            // If total time longer than 25%, resend the whole data.
+            // Since we are likely to be interrupted by SoftDevice
+            if ( (DWT->CYCCNT - cycStart) < (8 * numBytes * ((CYCLES_X00 * 5) / 4)) ) {
+                break;
+            }
+
+            // re-send need 300us delay
+            mp_hal_delay_us(300);
+        }
+
+        // Enable interrupts again
+        __enable_irq();
+    }
 }