From 0bb801c11404225735c7f25b1ecc6ea70db349e8 Mon Sep 17 00:00:00 2001
From: Jacob Lifshay <programmerjake@gmail.com>
Date: Mon, 19 Feb 2024 21:34:23 -0800
Subject: [PATCH] mul_remap: WIP -- fill in SVSHAPE[0-2], SVSHAPE3 needs work
 still

see: https://bugs.libre-soc.org/show_bug.cgi?id=1155#c72
---
 src/openpower/test/bigint/mul_remap.py | 21 ++++++++++++++++-----
 1 file changed, 16 insertions(+), 5 deletions(-)

diff --git a/src/openpower/test/bigint/mul_remap.py b/src/openpower/test/bigint/mul_remap.py
index 4baaba8f..d11ef619 100644
--- a/src/openpower/test/bigint/mul_remap.py
+++ b/src/openpower/test/bigint/mul_remap.py
@@ -42,20 +42,29 @@ def load_u32(reg, value):
 
 def mul_remap_256_x_256_to_512_asm():
     # manually create SVSHAPE until we have svshape[34] instructions working
-    # FIXME: fill in correct SVSHAPE[0-2] values
     # x for y in range(4) for x in range(4)
     SVSHAPE0 = SVSHAPE(0)
+    SVSHAPE0.lims = [4, 4, 1]
+    SVSHAPE0.skip = 0b10
 
     # y for y in range(4) for x in range(4)
     SVSHAPE1 = SVSHAPE(0)
+    SVSHAPE1.lims = [4, 4, 1]
+    SVSHAPE1.skip = 0b01
 
     # x + y for y in range(4) for x in range(4)
     # *must not* be limited mod 4, maximum value is 6 (3 + 3)
     SVSHAPE2 = SVSHAPE(0)
+    SVSHAPE0.lims = [4, 4, 1]
+    SVSHAPE0.mode = 0b11
+    SVSHAPE0.submode = 0b01      # rhombus mode
 
-    # 0 for y in range(4) for x in range(4)
+    # -1 for y in range(4) for x in range(4)
     # yes, this is a constant
     SVSHAPE3 = SVSHAPE(0)
+    SVSHAPE3.order = (0, 2, 1)  # so something is non-zero
+    # unsigned -- see https://bugs.libre-soc.org/show_bug.cgi?id=1155#c72
+    SVSHAPE3.offset = -1  # FIXME: offset is unsigned
     retval = [
         "mul_256_to_512:",
         # a is in r4-7, b is in r8-11, output (y) in r4-11
@@ -74,17 +83,19 @@ def mul_remap_256_x_256_to_512_asm():
         "mtspr %d, 0" % (SPRfull.SVSHAPE1.value,),
         *load_u32(0, SVSHAPE2),
         "mtspr %d, 0" % (SPRfull.SVSHAPE2.value,),
+        *load_u32(0, SVSHAPE3),
+        "mtspr %d, 0" % (SPRfull.SVSHAPE3.value,),
     ]
     retval += [
         "loop:",
         # use SVSHAPE0 for RA, SVSHAPE1 for RB,
         # SVSHAPE2 for RC/RT, SVSHAPE3 for RS
         "svremap 0o37, 0, 1, 2, 2, 3, 0",
-        # RS is scalar by using constant remap
-        "sv.maddedu *4, *32, *36, *4",  # FIXME: need to set EXTRA2_MODE to 0
+        # RS is scalar r3 by using constant remap with offset -1
+        "sv.maddedu *4, *32, *36, *4",
         # use SVSHAPE2 for RT/RA, all others are not remapped
         "svremap 0o11, 2, 0, 0, 2, 0, 0",
-        "sv.adde *5, *5, 20",  # FIXME: maddedu's RS is in r20, right?
+        "sv.adde *5, *5, 3",  # FIXME: maddedu's RS is in r3, right?
         "svstep 0, 0, 1",
         "bc 16, 0, loop # bdnz loop",
         "bclr 20, 0, 0 # blr",
-- 
2.30.2