firmware: zero-copy JPEG → BGR888 → DSI hot path; Scrypted pre-rotates

Architectural rework of the /frame hot path. Scrypted now ships every
JPEG already scaled + rotated to the panel's native dimensions (read
from the firmware's /state response so nothing is hardcoded on the
Scrypted side); the firmware decodes the JPEG straight into a BGR888
buffer that's directly draw_bitmap'able by the DSI driver, with zero
CPU pixel work and zero rotation work in between.

Firmware
- jpeg_decoder now uses JPEG_DECODE_OUT_FORMAT_RGB888 +
  JPEG_DEC_RGB_ELEMENT_ORDER_BGR. Output buffer sized for 800*480*3.
- New display_present_bgr888() is a one-liner that hands the decoder's
  output straight to esp_lcd_panel_draw_bitmap.
- /frame handler validates dimensions against the panel-native
  VIEWPORT_PANEL_WIDTH x VIEWPORT_PANEL_HEIGHT (was effective_dims
  branching on orientation). Returns 400 if it's anything else.
- /state JSON adds panel_width + panel_height so Scrypted can read them
  without hardcoding board-specific knowledge.
- display_present_rgb565 + s_rot_buf stay for the local-screens cold
  path (info screen, loading) which still does its own CPU conversion +
  rotation — infrequent enough that it's not worth the rewrite.

Scrypted
- startStream() GETs /state at stream-start time, caches panel_width
  and panel_height in viewport storage, and uses them as the ffmpeg
  scale target. Falls back to cached or 800x480 if /state is mid-reboot.
- For portrait viewports the ffmpeg pipeline now does
  scale=H:W:flags=lanczos,transpose=1 so the JPEG arrives pre-rotated
  90° CW into panel-native dimensions. Landscape is just scale=W:H.
- No more in-firmware rotation; Scrypted is the single source of truth
  for "how do I get this camera frame into a panel-shaped JPEG".

Expected ceiling lift: ~5 fps → ~10 fps, gated by the JPEG decoder
hardware throughput instead of the CPU rgb565→bgr888 + rotation loop.

5 files changed, 48 insertions, 19 deletions

diff --git a/main/display.c b/main/display.c
index a704891..1acddba 100644
--- a/main/display.c
+++ b/main/display.c
@@ -429,6 +429,14 @@ static inline void rgb565_to_rgb888(uint16_t px, uint8_t *out)
     out[2] = (uint8_t)(((px >> 11) & 0x1F) * 255 / 31);  // R
 }
 
+esp_err_t display_present_bgr888(const void *bgr888)
+{
+    if (!s_up) return ESP_ERR_INVALID_STATE;
+    return esp_lcd_panel_draw_bitmap(s_panel, 0, 0,
+                                     PANEL_H_ACTIVE, PANEL_V_ACTIVE,
+                                     bgr888);
+}
+
 esp_err_t display_present_rgb565(const uint16_t *src,
                                  uint16_t        src_w,
                                  uint16_t        src_h)
diff --git a/main/display.h b/main/display.h
index 9092efb..5f939b9 100644
--- a/main/display.h
+++ b/main/display.h
@@ -31,7 +31,16 @@ esp_err_t display_wake(void);
 // orientation. Source dimensions must match the effective resolution:
 //   portrait  -> src is 480x800 (rotated 90° CW into the 800x480 panel)
 //   landscape -> src is 800x480 (copied 1:1)
-// Used by /frame after JPEG decode.
+// CPU-rotation + format-conversion path used by local_screens for the
+// info / loading screens (cold path). /frame uses the zero-copy
+// BGR888 path below.
 esp_err_t display_present_rgb565(const uint16_t *src,
                                  uint16_t        src_w,
                                  uint16_t        src_h);
+
+// Zero-copy hot path. Source is already 800x480 with bytes in [B, G, R]
+// memory order (i.e. the format the panel pipeline natively wants);
+// hand it straight to esp_lcd_panel_draw_bitmap. No CPU pixel work, no
+// format conversion, no rotation — Scrypted is responsible for sending
+// the buffer pre-rotated and pre-scaled to panel-native dimensions.
+esp_err_t display_present_bgr888(const void *bgr888);
diff --git a/main/http_api.c b/main/http_api.c
index 2c5ce87..db317e4 100644
--- a/main/http_api.c
+++ b/main/http_api.c
@@ -53,6 +53,12 @@ static esp_err_t state_get_handler(httpd_req_t *req)
     cJSON_AddNumberToObject(root, "decode_errors", (double)st->decode_errors);
     cJSON_AddNumberToObject(root, "state_post_failures", (double)st->state_post_failures);
     cJSON_AddStringToObject(root, "resolution", viewport_state_resolution_str());
+    // Panel-native dimensions are stable per-board (Scrypted uses them as
+    // the ffmpeg scale target; orientation drives whether to transpose
+    // before scaling). Separate from "resolution" which is the effective
+    // dimensions after the orientation rotation.
+    cJSON_AddNumberToObject(root, "panel_width",  (double)VIEWPORT_PANEL_WIDTH);
+    cJSON_AddNumberToObject(root, "panel_height", (double)VIEWPORT_PANEL_HEIGHT);
     cJSON_AddStringToObject(root, "ip", net_eth_get_ip_str());
     cJSON_AddNumberToObject(root, "free_heap",
         (double)heap_caps_get_free_size(MALLOC_CAP_INTERNAL));
@@ -362,19 +368,22 @@ static esp_err_t frame_post_handler(httpd_req_t *req)
         return respond_status(req, "400 Bad Request", "JPEG decode failed");
     }
 
-    uint16_t want_w, want_h;
-    viewport_state_effective_dims(&want_w, &want_h);
-    if (w != want_w || h != want_h) {
+    // /frame always expects the panel-native 800x480 BGR888 layout —
+    // Scrypted does the rotation + scale, the firmware just decodes and
+    // paints. Orientation lives in viewport_state for /state reporting
+    // and the Scrypted-side ffmpeg pipeline only.
+    if (w != VIEWPORT_PANEL_WIDTH || h != VIEWPORT_PANEL_HEIGHT) {
         viewport_state_lock();
         viewport_state_get()->decode_errors++;
         viewport_state_unlock();
         jpeg_decoder_unlock();
         char msg[80];
-        snprintf(msg, sizeof(msg), "expected %ux%u, got %ux%u", want_w, want_h, w, h);
+        snprintf(msg, sizeof(msg), "expected %ux%u, got %ux%u",
+                 VIEWPORT_PANEL_WIDTH, VIEWPORT_PANEL_HEIGHT, w, h);
         return respond_status(req, "400 Bad Request", msg);
     }
 
-    esp_err_t paint_err = display_present_rgb565((const uint16_t *)rgb, w, h);
+    esp_err_t paint_err = display_present_bgr888(rgb);
     if (paint_err != ESP_OK) {
         jpeg_decoder_unlock();
         return respond_status(req, "500 Internal Server Error", "display paint failed");
diff --git a/main/jpeg_decoder.c b/main/jpeg_decoder.c
index f2a7d92..2538f69 100644
--- a/main/jpeg_decoder.c
+++ b/main/jpeg_decoder.c
@@ -62,18 +62,20 @@ void jpeg_decoder_unlock(void)
 void *jpeg_decoder_input_buffer(void) { return s_in_buf; }
 
 esp_err_t jpeg_decoder_decode(size_t   jpeg_len,
-                              void   **out_rgb565,
+                              void   **out_bgr888,
                               uint16_t *out_width,
                               uint16_t *out_height)
 {
     if (jpeg_len == 0 || jpeg_len > s_in_cap) return ESP_ERR_INVALID_SIZE;
 
+    // Hardware decode directly into a panel-native BGR888 buffer.
+    // _BGR rgb_order swaps the channel layout so memory ends up as
+    // [B, G, R] per pixel — exactly what the ESP32-P4 DSI engine +
+    // TC358762 + Pi panel pipeline wants. Total firmware-side cost in
+    // the hot /frame path: this hardware decode + one DMA hand-off to
+    // the panel. No CPU pixel work.
     jpeg_decode_cfg_t dec_cfg = {
-        .output_format = JPEG_DECODE_OUT_FORMAT_RGB565,
-        // Empirically: _RGB emits the RGB565 word in big-endian byte order
-        // (red 0xF800 → bytes f8 00) which our LE uint16 painter reads
-        // swapped. _BGR emits little-endian to match (despite the
-        // misleading "BGR" name) — verified by the decode-byte dump.
+        .output_format = JPEG_DECODE_OUT_FORMAT_RGB888,
         .rgb_order     = JPEG_DEC_RGB_ELEMENT_ORDER_BGR,
         .conv_std      = JPEG_YUV_RGB_CONV_STD_BT601,
     };
@@ -94,7 +96,7 @@ esp_err_t jpeg_decoder_decode(size_t   jpeg_len,
         return err;
     }
 
-    *out_rgb565 = s_out_buf;
+    *out_bgr888 = s_out_buf;
     *out_width  = info.width;
     *out_height = info.height;
     return ESP_OK;
diff --git a/main/jpeg_decoder.h b/main/jpeg_decoder.h
index 9c5bf89..2ac436d 100644
--- a/main/jpeg_decoder.h
+++ b/main/jpeg_decoder.h
@@ -7,7 +7,7 @@
 #include "esp_err.h"
 
 #define JPEG_DECODER_MAX_INPUT_BYTES   (1024 * 1024)        // 1 MB
-#define JPEG_DECODER_MAX_OUTPUT_BYTES  (800 * 480 * 2)      // RGB565 panel native
+#define JPEG_DECODER_MAX_OUTPUT_BYTES  (800 * 480 * 3)      // BGR888 panel native
 
 // One-time setup of the ESP32-P4 hardware JPEG decoder. Allocates reusable
 // DMA-aligned input + output buffers in PSRAM.
@@ -23,11 +23,12 @@ void jpeg_decoder_unlock(void);
 // JPEG_DECODER_MAX_INPUT_BYTES. Caller fills before calling decode.
 void *jpeg_decoder_input_buffer(void);
 
-// Decode the JPEG sitting in the input buffer. Fills out_rgb565 with a
-// pointer to the decoded RGB565 image and reports its width/height in
-// pixels.  The output buffer is owned by the decoder — valid only until
-// the next jpeg_decoder_unlock().
+// Decode the JPEG sitting in the input buffer. Fills out_bgr888 with a
+// pointer to the decoded 24-bit image (3 bytes/pixel, BGR memory order so
+// the DSI engine + TC358762 + Pi panel render channels correctly) and
+// reports the image's width/height in pixels. The output buffer is owned
+// by the decoder — valid only until the next jpeg_decoder_unlock().
 esp_err_t jpeg_decoder_decode(size_t   jpeg_len,
-                              void   **out_rgb565,
+                              void   **out_bgr888,
                               uint16_t *out_width,
                               uint16_t *out_height);