Re-enable interrupts in between writing LED data for ARM M0. (#751)

* Fix breakage in NRF51 support introduced by #dba8825

* Re-enable interrupts in between writing LED data for ARM M0.

This adds support for FASTLED_ALLOW_INTERRUPTS to ARM M0-based platforms
(e.g. SAMD21).

This fixes the clock getting off when using more than ~30 LEDs, since
Arduino uses an interrupt to increment the millis clock.

This uses SysTick->VAL to determine whether more than 45uSecs have
elapsed while interrupts were enabled. This isn't as correct as using a
dedicated timer, but it does work on all ARM M0 platforms.

2 files changed, 111 insertions, 40 deletions

diff --git a/platforms/arm/common/m0clockless.h b/platforms/arm/common/m0clockless.h
index 4798c6fa..d5a0cf6f 100644
--- a/platforms/arm/common/m0clockless.h
+++ b/platforms/arm/common/m0clockless.h
@@ -225,16 +225,17 @@ showLedData(volatile uint32_t *_port, uint32_t _bitmask, const uint8_t *_leds, u
 #define CMPLOOP5        "  cmploop5 %[counter], loop_%=;"
 #define NOTHING         ""
 
-#if !(defined(SEI_CHK) && (FASTLED_ALLOW_INTERRUPTS == 1))
-    // We're not allowing interrupts - run the entire loop in asm to keep things
-    // as tight as possible.  In an ideal world, we should be pushing out ws281x
-    // leds (or other 3-wire leds) with zero gaps between pixels.
+#if (defined(SEI_CHK) && (FASTLED_ALLOW_INTERRUPTS == 1))
+    // We're allowing interrupts and have hardware timer support defined -
+    // track the loop outside the asm code, to allow inserting the interrupt
+    // overrun checks.
     asm __volatile__ (
       // pre-load byte 0
-    LOADLEDS3(0) LOADDITHER7(0) DITHER5 SCALE4(0) ADJDITHER7(0) SWAPBBN1
+      LOADLEDS3(0) LOADDITHER7(0) DITHER5 SCALE4(0) ADJDITHER7(0) SWAPBBN1
+      M0_ASM_ARGS);
 
-    // loop over writing out the data
-    LOOP
+    do {
+      asm __volatile__ (
       // Write out byte 0, prepping byte 1
       HI2 _D1 QLO4 NOTHING         _D2(0) LO2 _D3(0)
       HI2 _D1 QLO4 LOADLEDS3(1)    _D2(3) LO2 _D3(0)
@@ -252,24 +253,26 @@ showLedData(volatile uint32_t *_port, uint32_t _bitmask, const uint8_t *_leds, u
       HI2 _D1 QLO4 DITHER5         _D2(5) LO2 _D3(0)
       HI2 _D1 QLO4 SCALE4(2)       _D2(4) LO2 _D3(0)
       HI2 _D1 QLO4 ADJDITHER7(2)   _D2(7) LO2 _D3(0)
-      HI2 _D1 QLO4 INCLEDS3        _D2(3) LO2 _D3(0)
+      HI2 _D1 QLO4 NOTHING         _D2(0) LO2 _D3(0)
       HI2 _D1 QLO4 SWAPBBN1        _D2(1) LO2 _D3(0)
 
       // Write out byte 2, prepping byte 0
-      HI2 _D1 QLO4 NOTHING         _D2(0) LO2 _D3(0)
+      HI2 _D1 QLO4 INCLEDS3        _D2(3) LO2 _D3(0)
       HI2 _D1 QLO4 LOADLEDS3(0)    _D2(3) LO2 _D3(0)
       HI2 _D1 QLO4 LOADDITHER7(0)  _D2(7) LO2 _D3(0)
       HI2 _D1 QLO4 DITHER5         _D2(5) LO2 _D3(0)
       HI2 _D1 QLO4 SCALE4(0)       _D2(4) LO2 _D3(0)
       HI2 _D1 QLO4 ADJDITHER7(0)   _D2(7) LO2 _D3(0)
       HI2 _D1 QLO4 NOTHING         _D2(0) LO2 _D3(0)
-      HI2 _D1 QLO4 SWAPBBN1        _D2(1) LO2 _D3(5) CMPLOOP5
+      HI2 _D1 QLO4 SWAPBBN1        _D2(1) LO2 _D3(5)
 
       M0_ASM_ARGS
-    );
-#else
-    // We're allowing interrupts - track the loop outside the asm code, to allow
-    // inserting the interrupt overrun checks.
+      );
+      SEI_CHK; INNER_SEI; --counter; CLI_CHK;
+    } while(counter);
+#elif (FASTLED_ALLOW_INTERRUPTS == 1)
+    // We're allowing interrupts - track the loop outside the asm code, and
+    // re-enable interrupts in between each iteration.
     asm __volatile__ (
       // pre-load byte 0
       LOADLEDS3(0) LOADDITHER7(0) DITHER5 SCALE4(0) ADJDITHER7(0) SWAPBBN1
@@ -278,39 +281,107 @@ showLedData(volatile uint32_t *_port, uint32_t _bitmask, const uint8_t *_leds, u
     do {
       asm __volatile__ (
       // Write out byte 0, prepping byte 1
-      HI2 D1 QLO4 NOTHING         D2(0) LO2 D3(0)
-      HI2 D1 QLO4 LOADLEDS3(1)    D2(3) LO2 D3(0)
-      HI2 D1 QLO4 LOADDITHER7(1)  D2(7) LO2 D3(0)
-      HI2 D1 QLO4 DITHER5         D2(5) LO2 D3(0)
-      HI2 D1 QLO4 SCALE4(1)       D2(4) LO2 D3(0)
-      HI2 D1 QLO4 ADJDITHER7(1)   D2(7) LO2 D3(0)
-      HI2 D1 QLO4 NOTHING         D2(0) LO2 D3(0)
-      HI2 D1 QLO4 SWAPBBN1        D2(1) LO2 D3(0)
+      HI2 _D1 QLO4 NOTHING         _D2(0) LO2 _D3(0)
+      HI2 _D1 QLO4 LOADLEDS3(1)    _D2(3) LO2 _D3(0)
+      HI2 _D1 QLO4 LOADDITHER7(1)  _D2(7) LO2 _D3(0)
+      HI2 _D1 QLO4 DITHER5         _D2(5) LO2 _D3(0)
+      HI2 _D1 QLO4 SCALE4(1)       _D2(4) LO2 _D3(0)
+      HI2 _D1 QLO4 ADJDITHER7(1)   _D2(7) LO2 _D3(0)
+      HI2 _D1 QLO4 NOTHING         _D2(0) LO2 _D3(0)
+      HI2 _D1 QLO4 SWAPBBN1        _D2(1) LO2 _D3(0)
 
       // Write out byte 1, prepping byte 2
-      HI2 D1 QLO4 NOTHING         D2(0) LO2 D3(0)
-      HI2 D1 QLO4 LOADLEDS3(2)    D2(3) LO2 D3(0)
-      HI2 D1 QLO4 LOADDITHER7(2)  D2(7) LO2 D3(0)
-      HI2 D1 QLO4 DITHER5         D2(5) LO2 D3(0)
-      HI2 D1 QLO4 SCALE4(2)       D2(4) LO2 D3(0)
-      HI2 D1 QLO4 ADJDITHER7(2)   D2(7) LO2 D3(0)
-      HI2 D1 QLO4 NOTHING         D2(0) LO2 D3(0)
-      HI2 D1 QLO4 SWAPBBN1        D2(1) LO2 D3(0)
+      HI2 _D1 QLO4 NOTHING         _D2(0) LO2 _D3(0)
+      HI2 _D1 QLO4 LOADLEDS3(2)    _D2(3) LO2 _D3(0)
+      HI2 _D1 QLO4 LOADDITHER7(2)  _D2(7) LO2 _D3(0)
+      HI2 _D1 QLO4 DITHER5         _D2(5) LO2 _D3(0)
+      HI2 _D1 QLO4 SCALE4(2)       _D2(4) LO2 _D3(0)
+      HI2 _D1 QLO4 ADJDITHER7(2)   _D2(7) LO2 _D3(0)
+      HI2 _D1 QLO4 INCLEDS3        _D2(3) LO2 _D3(0)
+      HI2 _D1 QLO4 SWAPBBN1        _D2(1) LO2 _D3(0)
 
       // Write out byte 2, prepping byte 0
-      HI2 D1 QLO4 INCLEDS3        D2(3) LO2 D3(0)
-      HI2 D1 QLO4 LOADLEDS3(0)    D2(3) LO2 D3(0)
-      HI2 D1 QLO4 LOADDITHER7(0)  D2(7) LO2 D3(0)
-      HI2 D1 QLO4 DITHER5         D2(5) LO2 D3(0)
-      HI2 D1 QLO4 SCALE4(0)       D2(4) LO2 D3(0)
-      HI2 D1 QLO4 ADJDITHER7(0)   D2(7) LO2 D3(0)
-      HI2 D1 QLO4 NOTHING         D2(0) LO2 D3(0)
-      HI2 D1 QLO4 SWAPBBN1        D2(1) LO2 D3(5)
+      HI2 _D1 QLO4 NOTHING         _D2(0) LO2 _D3(0)
+      HI2 _D1 QLO4 LOADLEDS3(0)    _D2(3) LO2 _D3(0)
+      HI2 _D1 QLO4 LOADDITHER7(0)  _D2(7) LO2 _D3(0)
+      HI2 _D1 QLO4 DITHER5         _D2(5) LO2 _D3(0)
+      HI2 _D1 QLO4 SCALE4(0)       _D2(4) LO2 _D3(0)
+      HI2 _D1 QLO4 ADJDITHER7(0)   _D2(7) LO2 _D3(0)
+      HI2 _D1 QLO4 NOTHING         _D2(0) LO2 _D3(0)
+      HI2 _D1 QLO4 SWAPBBN1        _D2(1) LO2 _D3(5)
 
       M0_ASM_ARGS
       );
-      SEI_CHK; INNER_SEI; --counter; CLI_CHK;
+
+      uint32_t ticksBeforeInterrupts = SysTick->VAL;
+      sei();
+      --counter;
+      cli();
+
+      // If more than 45 uSecs have elapsed, give up on this frame and start over.
+      // Note: this isn't completely correct. It's possible that more than one
+      // millisecond will elapse, and so SysTick->VAL will lap
+      // ticksBeforeInterrupts.
+      // Note: ticksBeforeInterrupts DECREASES
+      const uint32_t kTicksPerMs = VARIANT_MCK / 1000;
+      const uint32_t kTicksPerUs = kTicksPerMs / 1000;
+      const uint32_t kTicksIn45us = kTicksPerUs * 45;
+
+      const uint32_t currentTicks = SysTick->VAL;
+
+      if (ticksBeforeInterrupts < currentTicks) {
+        // Timer started over
+        if ((ticksBeforeInterrupts + (kTicksPerMs - currentTicks)) > kTicksIn45us) {
+          return 0;
+        }
+      } else {
+        if ((ticksBeforeInterrupts - currentTicks) > kTicksIn45us) {
+          return 0;
+        }
+      }
     } while(counter);
+#else
+    // We're not allowing interrupts - run the entire loop in asm to keep things
+    // as tight as possible.  In an ideal world, we should be pushing out ws281x
+    // leds (or other 3-wire leds) with zero gaps between pixels.
+    asm __volatile__ (
+      // pre-load byte 0
+    LOADLEDS3(0) LOADDITHER7(0) DITHER5 SCALE4(0) ADJDITHER7(0) SWAPBBN1
+
+    // loop over writing out the data
+    LOOP
+      // Write out byte 0, prepping byte 1
+      HI2 _D1 QLO4 NOTHING         _D2(0) LO2 _D3(0)
+      HI2 _D1 QLO4 LOADLEDS3(1)    _D2(3) LO2 _D3(0)
+      HI2 _D1 QLO4 LOADDITHER7(1)  _D2(7) LO2 _D3(0)
+      HI2 _D1 QLO4 DITHER5         _D2(5) LO2 _D3(0)
+      HI2 _D1 QLO4 SCALE4(1)       _D2(4) LO2 _D3(0)
+      HI2 _D1 QLO4 ADJDITHER7(1)   _D2(7) LO2 _D3(0)
+      HI2 _D1 QLO4 NOTHING         _D2(0) LO2 _D3(0)
+      HI2 _D1 QLO4 SWAPBBN1        _D2(1) LO2 _D3(0)
+
+      // Write out byte 1, prepping byte 2
+      HI2 _D1 QLO4 NOTHING         _D2(0) LO2 _D3(0)
+      HI2 _D1 QLO4 LOADLEDS3(2)    _D2(3) LO2 _D3(0)
+      HI2 _D1 QLO4 LOADDITHER7(2)  _D2(7) LO2 _D3(0)
+      HI2 _D1 QLO4 DITHER5         _D2(5) LO2 _D3(0)
+      HI2 _D1 QLO4 SCALE4(2)       _D2(4) LO2 _D3(0)
+      HI2 _D1 QLO4 ADJDITHER7(2)   _D2(7) LO2 _D3(0)
+      HI2 _D1 QLO4 INCLEDS3        _D2(3) LO2 _D3(0)
+      HI2 _D1 QLO4 SWAPBBN1        _D2(1) LO2 _D3(0)
+
+      // Write out byte 2, prepping byte 0
+      HI2 _D1 QLO4 NOTHING         _D2(0) LO2 _D3(0)
+      HI2 _D1 QLO4 LOADLEDS3(0)    _D2(3) LO2 _D3(0)
+      HI2 _D1 QLO4 LOADDITHER7(0)  _D2(7) LO2 _D3(0)
+      HI2 _D1 QLO4 DITHER5         _D2(5) LO2 _D3(0)
+      HI2 _D1 QLO4 SCALE4(0)       _D2(4) LO2 _D3(0)
+      HI2 _D1 QLO4 ADJDITHER7(0)   _D2(7) LO2 _D3(0)
+      HI2 _D1 QLO4 NOTHING         _D2(0) LO2 _D3(0)
+      HI2 _D1 QLO4 SWAPBBN1        _D2(1) LO2 _D3(5) CMPLOOP5
+
+      M0_ASM_ARGS
+    );
 #endif
     return num_leds;
 }
diff --git a/platforms/arm/d21/led_sysdefs_arm_d21.h b/platforms/arm/d21/led_sysdefs_arm_d21.h
index b343d3f2..a48db10a 100644
--- a/platforms/arm/d21/led_sysdefs_arm_d21.h
+++ b/platforms/arm/d21/led_sysdefs_arm_d21.h
@@ -11,7 +11,7 @@
 
 // Default to allowing interrupts
 #ifndef FASTLED_ALLOW_INTERRUPTS
-#define FASTLED_ALLOW_INTERRUPTS 0
+#define FASTLED_ALLOW_INTERRUPTS 1
 #endif
 
 #if FASTLED_ALLOW_INTERRUPTS == 1