From 8eef50287e5d991d6368d457e9fff5929b293254 Mon Sep 17 00:00:00 2001
From: Luke Kenneth Casson Leighton <lkcl@lkcl.net>
Date: Sat, 23 Dec 2023 19:13:03 +0000
Subject: [PATCH] first cut at a triangular mode

---
 src/openpower/decoder/isa/remapyield.py | 51 +++++++++++++++++++------
 src/openpower/decoder/isa/svshape.py    |  5 +++
 2 files changed, 44 insertions(+), 12 deletions(-)

diff --git a/src/openpower/decoder/isa/remapyield.py b/src/openpower/decoder/isa/remapyield.py
index 82eb13b8..389cfb6b 100644
--- a/src/openpower/decoder/isa/remapyield.py
+++ b/src/openpower/decoder/isa/remapyield.py
@@ -4,13 +4,15 @@
 # python "yield" can be iterated. use this to make it clear how
 # the indices are generated by using natural-looking nested loops
 def iterate_indices(SVSHAPE):
+    # establish if this is triangular mode
+    triangle = SVSHAPE.mode == 0b11 and SVSHAPE.submode == 0b00
     # get indices to iterate over, in the required order
     xd = SVSHAPE.lims[0]
     yd = SVSHAPE.lims[1]
     zd = SVSHAPE.lims[2]
     # create lists of indices to iterate over in each dimension
-    x_r = list(range(xd))
-    y_r = list(range(yd))
+    _x_r = list(range(xd))
+    _y_r = list(range(yd))
     z_r = list(range(zd))
     # invert the indices if needed
     if SVSHAPE.invxyz[0]: x_r.reverse()
@@ -19,11 +21,16 @@ def iterate_indices(SVSHAPE):
     # start an infinite (wrapping) loop
     step = 0 # track src/dst step
     while True:
-        for z in z_r:   # loop over 1st order dimension
+        for zi, z in enumerate(z_r):   # loop over 1st order dimension
             z_end = z == z_r[-1]
-            for y in y_r:       # loop over 2nd order dimension
+            # triangle uses a truncated copy
+            y_r = list(_y_r[:zi+1] if triangle else _y_r)
+            y_end = True # if list is empty
+            for yi, y in enumerate(y_r): # loop over 2nd order dimension
                 y_end = y == y_r[-1]
-                for x in x_r:           # loop over 3rd order dimension
+                x_r = list(_x_r[:yi+1] if triangle else _x_r)
+                x_end = True # if list is empty
+                for xi, x in enumerate(x_r): # loop over 3rd order dimension
                     x_end = x == x_r[-1]
                     # ok work out which order to construct things in.
                     # start by creating a list of tuples of the dimension
@@ -49,12 +56,12 @@ def iterate_indices(SVSHAPE):
                         # was actually selected above on all 3 dimensions,
                         # e.g. [z][x][y] or [y][z][x].  "skip" allows one of
                         # those to be knocked out
-                        if SVSHAPE.skip == i+1: continue
                         #print ("select %d %s" % (i, dbg))
-                        if SVSHAPE.mode == 0b00:
+                        if SVSHAPE.mode == 0b00: # Matrix Mode
+                            if SVSHAPE.skip == i+1: continue
                             idx *= mult   # shifts up by previous dimension(s)
                         result += idx # adds on this dimension
-                        if SVSHAPE.mode == 0b00:
+                        if SVSHAPE.mode == 0b00: # Matrix Mode
                             mult *= lim   # for the next dimension
 
                     loopends = (x_end |
@@ -71,10 +78,16 @@ def dump_shape(SVSHAPE0):
     for idx, (new_idx, end) in enumerate(iterate_indices(SVSHAPE0)):
         estring = bin(end)[2:] if end else ''
         print ("%2d->%-2d" % (idx, new_idx), "%2s" % estring, end=' ')
-        if end != 0:
-            print()
         if end == 0b111:
+            print()
             break
+        triangle = SVSHAPE0.mode == 0b11 and SVSHAPE0.submode == 0b00
+        if triangle:
+            if SVSHAPE0.lims[0] == 1: end &= ~0b001
+        if triangle:
+            if SVSHAPE0.lims[1] == 1: end &= ~0b010
+        if end != 0:
+            print()
 
 
 def demo():
@@ -96,12 +109,26 @@ def demo():
     dump_shape(SVSHAPE0)
 
     # rhombus - see https://bugs.libre-soc.org/show_bug.cgi?id=1155#c17
-    print("\ntriangular, skip-x", xdim, ydim, zdim)
+    xdim, ydim, zdim = 3, 4, 1 # set the dimension sizes here
+    print("\nrhombus", xdim, ydim, zdim)
+    SVSHAPE0 = SVSHAPE()
+    SVSHAPE0.lims = [xdim, ydim, zdim]
+    SVSHAPE0.order = [0,1,2]  # experiment with different permutations, here
+    SVSHAPE0.mode = 0b11
+    SVSHAPE0.submode = 0b01      # rhombus mode
+    SVSHAPE0.offset = 0       # experiment with different offset, here
+    SVSHAPE0.invxyz = [0,0,0] # inversion if desired
+
+    dump_shape(SVSHAPE0)
+
+    # triangle
+    print("\ntriangle", xdim, ydim, zdim)
+    xdim, ydim, zdim = 1, 5, 5 # set the dimension sizes here
     SVSHAPE0 = SVSHAPE()
     SVSHAPE0.lims = [xdim, ydim, zdim]
     SVSHAPE0.order = [0,1,2]  # experiment with different permutations, here
     SVSHAPE0.mode = 0b11
-    SVSHAPE0.skip = 0b01      # try skipping x-dimension (for fun)
+    SVSHAPE0.submode = 0b00      # triangle mode
     SVSHAPE0.offset = 0       # experiment with different offset, here
     SVSHAPE0.invxyz = [0,0,0] # inversion if desired
 
diff --git a/src/openpower/decoder/isa/svshape.py b/src/openpower/decoder/isa/svshape.py
index e34c0842..7876fda4 100644
--- a/src/openpower/decoder/isa/svshape.py
+++ b/src/openpower/decoder/isa/svshape.py
@@ -133,6 +133,11 @@ class SVSHAPE(SelectableInt):
     def elwid(self):
         return self.fsi['skip'].asint(msb0=True)
 
+    # submode is the same bits as skip (like elwidth)
+    @property
+    def submode(self):
+        return self.fsi['skip'].asint(msb0=True)
+
     @property
     def skip(self):
         if self.is_indexed():
-- 
2.30.2