frameline GW: fix roi

frameline GW: remove unused import

frameline GW: add econ roi

src/gateware/cxp_frame_pipeline.py

@@ -1,5 +1,4 @@
 from migen import *
-from migen.genlib.coding import PriorityEncoder
 from misoc.interconnect.csr import *
 from misoc.interconnect import stream
 from misoc.cores.liteeth_mini.mac.crc import LiteEthMACCRCEngine
@@ -631,31 +630,26 @@ class ROI(Module):
 
         # # #
 
-        # TODO: remove the self. from self.roi_4x
-        self.roi_4x = []
-        for _ in range(4):
-            self.roi_4x.append(Record([
+        roi_4x = [
+            Record([
                 ("x_good", 1),
                 ("y_good", 1),
                 ("gray", len(pixel_4x[0].gray)),
                 ("stb", 1),
                 ("count", count_width),
-            ]))
+            ]) for _ in range(4)
             
-        for pix, roi in zip(pixel_4x, self.roi_4x):
+        ]
+
+        for pix, roi in zip(pixel_4x, roi_4x):
             self.sync += [
-                # TODO: replace the comparision with preprocess equal
-                # e.g. pix.x == self.cfg.x0 - i
                 # stage 1 - generate "good" (in-ROI) signals
-                If(pix.x <= self.cfg.x0,
+                roi.x_good.eq(0),
+                If((self.cfg.x0 <= pix.x) & (pix.x < self.cfg.x1),
                     roi.x_good.eq(1)
                 ),
-                # NOTE: this gate doens't work as 4 pixes are coming in
-                If(pix.x >= self.cfg.x1,
-                    roi.x_good.eq(0)
-                ),
 
-                # This is fine because 4x pixel are on the same line
+                # the 4 pixels are on the same y level, no need for extra calculation
                 If(pix.y == self.cfg.y0,
                     roi.y_good.eq(1)
                 ),
@@ -683,12 +677,115 @@ class ROI(Module):
         self.sync += [
             eof.eq(reduce(or_, [pix.eof for pix in pixel_4x])),
             eof_buf.eq(eof),
-            count_buf[0].eq(self.roi_4x[0].count + self.roi_4x[1].count),
-            count_buf[1].eq(self.roi_4x[2].count + self.roi_4x[3].count),
+            count_buf[0].eq(roi_4x[0].count + roi_4x[1].count),
+            count_buf[1].eq(roi_4x[2].count + roi_4x[3].count),
 
             self.out.update.eq(0),
             If(eof_buf,
-                [roi.count.eq(0) for roi in self.roi_4x],
+                [roi.count.eq(0) for roi in roi_4x],
+                self.out.update.eq(1),
+                self.out.count.eq(reduce(add, count_buf))
+            ),
+        ]
+
+class Economical_ROI(Module):
+    """
+    ROI Engine. For each frame, accumulates pixels values within a
+    rectangular region of interest, and reports the total.
+
+    This econ version limits the distance between x0 and x1 need to be at least 4 pixel long
+    """
+    def __init__(self, pixel_4x, count_width):
+        assert len(pixel_4x) == 4
+
+        self.cfg = Record([
+            ("x0", len(pixel_4x[0].x)),
+            ("y0", len(pixel_4x[0].y)),
+            ("x1", len(pixel_4x[0].x)),
+            ("y1", len(pixel_4x[0].y)),
+        ])
+
+        self.out = Record([
+            ("update", 1),
+            # registered output - can be used as CDC input
+            ("count", count_width),
+        ])
+
+        # # #
+
+        roi_4x = [
+            Record([
+                ("x_good", 1),
+                ("y_good", 1),
+                ("gray", len(pixel_4x[0].gray)),
+                ("stb", 1),
+                ("count", count_width),
+            ]) for _ in range(4)
+            
+        ]
+
+        # Pipeline the offset calculation
+        x0_offset = [Signal.like(self.cfg.x0) for _ in range(4)]
+        x1_offset = [Signal.like(self.cfg.x1) for _ in range(4)]
+
+        for offset, (x0, x1) in enumerate(zip(x0_offset, x1_offset)):
+            self.sync += [
+                x0.eq(self.cfg.x0 + offset),
+                x1.eq(self.cfg.x1 + offset),
+            ]
+
+        for pix, roi in zip(pixel_4x, roi_4x):
+            # stage 1 - generate "good" (in-ROI) signals
+             
+            # if p in [x0, x0+3] => x_good == 1
+            # if p in [x1, x1+3] => x_good == 0
+            # when [x0, x0+3] overlap with [x1, x1+3]:
+            # some x_good won't have the chance to fall and keep accmulated extra data
+            # Thus, x1-x0 >= 4 need to be match for pixel to accmulated correctly
+            for x0, x1 in zip(x0_offset, x1_offset):
+                self.sync += [
+                    If(pix.x == x0,
+                       roi.x_good.eq(1),
+                    ),
+                    If(pix.x == x1,
+                       roi.x_good.eq(0),
+                    ),
+                ]
+            self.sync += [
+                # the 4 pixels are on the same y level, no need for extra calculation
+                If(pix.y == self.cfg.y0,
+                    roi.y_good.eq(1)
+                ),
+                If(pix.y == self.cfg.y1,
+                    roi.y_good.eq(0)
+                ),
+                If(pix.eof,
+                    roi.x_good.eq(0),
+                    roi.y_good.eq(0)
+                ),
+                roi.gray.eq(pix.gray),
+                roi.stb.eq(pix.stb),
+
+                # stage 2 - accumulate
+                If((roi.stb & roi.x_good & roi.y_good),
+                    roi.count.eq(roi.count + roi.gray)
+                )
+            ]
+
+        eof = Signal()
+        eof_buf = Signal()
+        count_buf = [Signal(count_width), Signal(count_width)]
+        
+        # stage 3 - update
+        self.sync += [
+            eof.eq(reduce(or_, [pix.eof for pix in pixel_4x])),
+            eof_buf.eq(eof),
+            count_buf[0].eq(roi_4x[0].count + roi_4x[1].count),
+            count_buf[1].eq(roi_4x[2].count + roi_4x[3].count),
+
+            self.out.update.eq(0),
+            If(eof_buf,
+                [roi.count.eq(0) for roi in roi_4x],
                 self.out.update.eq(1),
                 self.out.count.eq(reduce(add, count_buf))
             ),