From 2a953464fd5a399d5f728de7ac72b647ced41e7f Mon Sep 17 00:00:00 2001
From: radex <radex@hackerspace.pl>
Date: Sun, 26 May 2024 23:00:49 +0200
Subject: [PATCH] optimize: use irq to signal readiness

---
 firmware/src/gfx_decoder.cpp |  1 +
 firmware/src/leds.cpp        | 19 ++++++++-----------
 firmware/src/leds.h          |  1 +
 firmware/src/leds.pio        |  5 ++---
 firmware/src/leds.pio.h      |  6 +++---
 5 files changed, 15 insertions(+), 17 deletions(-)

diff --git a/firmware/src/gfx_decoder.cpp b/firmware/src/gfx_decoder.cpp
index eaff6a0..b438366 100644
--- a/firmware/src/gfx_decoder.cpp
+++ b/firmware/src/gfx_decoder.cpp
@@ -78,6 +78,7 @@ int32_t gfx_decoder_loadNextFrame() {
       }
     }
   }
+  ledBufferReady = true;
 
   // copy to framebuffer
   // TODO: mutex? double buffer? or something...
diff --git a/firmware/src/leds.cpp b/firmware/src/leds.cpp
index c881153..7b5c23b 100644
--- a/firmware/src/leds.cpp
+++ b/firmware/src/leds.cpp
@@ -36,6 +36,7 @@ uint8_t brightnessPhaseDelays[COLOR_BITS] = {0, 1, 6, 20, 60};
 
 // NOTE: Alignment required to allow 4-byte reads
 uint8_t framebuffer[ROW_COUNT * COL_COUNT]  __attribute__((aligned(32))) = {0};
+bool ledBufferReady = false;
 uint32_t ledBuffer[8][ROW_COUNT * COL_MODULES] = {0};
 
 void leds_init() {
@@ -93,6 +94,11 @@ void leds_initRenderer() {
 }
 
 void leds_render() {
+  if (!ledBufferReady) {
+    outputEnable(ROW_OE, false);
+    return;
+  }
+
   // brightness phase
   bool brightPhase = brightnessPhase >= 3;
   auto buffer = ledBuffer[brightnessPhase + 3];
@@ -134,25 +140,16 @@ void leds_render() {
       // - see if we can disable px pusher delays on improved electric interface
       // - improve outer loop which adds 2us of processing on each loop
       // - change busy wait into some kind of interrupt-based thing so that processing can continue
-      // - latch row and clock simultaneously, avoid disabling output
       // - DMA?
       for (int xModule = 0; xModule < COL_MODULES; xModule++) {
         uint32_t pxValues = buffer[bufferOffset + xModule];
         pio_sm_put_blocking(pusher_pio, pusher_sm, pxValues);
       }
 
-      // wait for all data to be shifted out
-      while (!pio_sm_is_tx_fifo_empty(pusher_pio, pusher_sm)) {
+      // wait until pushing and RCLK latch are done
+      while (!pio_interrupt_get(pusher_pio, 0)) {
         tight_loop_contents();
       }
-      // TODO: Is there an API to wait for PIO to actually become idle?
-      // pio_sm_drain_tx_fifo doesn't seem to do the trick
-      // if not, we might need to use irqs or something
-      busy_wait_us(4);
-      // TODO: Why doesn't this work?!
-      // while (!pio_interrupt_get(pusher_pio, pusher_sm)) {
-      //   tight_loop_contents();
-      // }
       pio_interrupt_clear(pusher_pio, pusher_sm);
 
       // show for a certain period
diff --git a/firmware/src/leds.h b/firmware/src/leds.h
index f443b63..7ea0485 100644
--- a/firmware/src/leds.h
+++ b/firmware/src/leds.h
@@ -33,6 +33,7 @@ void leds_loop();
 void leds_render();
 
 extern uint8_t framebuffer[ROW_COUNT * COL_COUNT];
+extern bool ledBufferReady;
 extern uint32_t ledBuffer[8][ROW_COUNT * COL_MODULES];
 
 #endif
diff --git a/firmware/src/leds.pio b/firmware/src/leds.pio
index b7299b9..b3c4b5e 100644
--- a/firmware/src/leds.pio
+++ b/firmware/src/leds.pio
@@ -6,7 +6,7 @@ public entry_point:
   ; pull
   ; push 24 bits to the shift registers
   ; also, return latch bit to 0
-  set x, 23   side 0
+  set x, 23      side 0
 loop:
   ; TODO: check if delays can be lowered with a PCB
   ; set bit; lower clock edge
@@ -23,10 +23,9 @@ end:
   .wrap
 end_of_row:
   ; indicate to main processor that row was processed
+  irq set 0
   ; clock RCLK (latch onto register output stage)
   set pins, 1 [3]
   set pins, 0
-  irq wait 0 rel
-
   ; wait for next row
   jmp entry_point
diff --git a/firmware/src/leds.pio.h b/firmware/src/leds.pio.h
index d7b559d..c5eb55d 100644
--- a/firmware/src/leds.pio.h
+++ b/firmware/src/leds.pio.h
@@ -25,9 +25,9 @@ static const uint16_t leds_px_pusher_program_instructions[] = {
     0x7028, //  3: out    x, 8            side 0     
     0x0045, //  4: jmp    x--, 5                     
             //     .wrap
-    0xe301, //  5: set    pins, 1                [3] 
-    0xe000, //  6: set    pins, 0                    
-    0xc030, //  7: irq    wait 0 rel                 
+    0xc000, //  5: irq    nowait 0                   
+    0xe301, //  6: set    pins, 1                [3] 
+    0xe000, //  7: set    pins, 0                    
     0x0000, //  8: jmp    0                          
 };