From 74e1cb79e40d1a7c66f7ad0b942cbe654859eeda Mon Sep 17 00:00:00 2001
From: Luke Kenneth Casson Leighton <lkcl@lkcl.net>
Date: Wed, 17 Mar 2021 22:15:58 +0000
Subject: [PATCH] add twin-predicated extsw SVP64 ISACaller unit test

---
 src/soc/decoder/isa/caller.py                 |  5 ++-
 .../isa/test_caller_svp64_predication.py      | 40 ++++++++++++++++++-
 2 files changed, 42 insertions(+), 3 deletions(-)

diff --git a/src/soc/decoder/isa/caller.py b/src/soc/decoder/isa/caller.py
index 0371f691..23c1c5ae 100644
--- a/src/soc/decoder/isa/caller.py
+++ b/src/soc/decoder/isa/caller.py
@@ -927,7 +927,7 @@ class ISACaller:
             if pmode == SVP64PredMode.INT.value:
                 srcmask = dstmask = get_predint(self.gpr, dstpred)
                 if sv_ptype == SVPtype.P2.value:
-                    srcmask = get_predint(srcpred)
+                    srcmask = get_predint(self.gpr, srcpred)
             elif pmode == SVP64PredMode.CR.value:
                 srcmask = dstmask = get_predcr(self.crl, dstpred, vl)
                 if sv_ptype == SVPtype.P2.value:
@@ -943,6 +943,9 @@ class ISACaller:
             while (((1<<srcstep) & srcmask) == 0) and (srcstep != vl):
                 print ("      skip", bin(1<<srcstep))
                 srcstep += 1
+            # same for dststep
+            while (((1<<dststep) & dstmask) == 0) and (dststep != vl):
+                print ("      skip", bin(1<<dststep))
                 dststep += 1
 
             # update SVSTATE with new srcstep
diff --git a/src/soc/decoder/isa/test_caller_svp64_predication.py b/src/soc/decoder/isa/test_caller_svp64_predication.py
index 9ab7c790..66f09007 100644
--- a/src/soc/decoder/isa/test_caller_svp64_predication.py
+++ b/src/soc/decoder/isa/test_caller_svp64_predication.py
@@ -42,7 +42,43 @@ class DecoderTestCase(FHDLTestCase):
             self.assertEqual(sim.gpr(9), SelectableInt(0x1234, 64))
             self.assertEqual(sim.gpr(10), SelectableInt(0x1235, 64))
 
-    def test_sv_add_intpred(self):
+    def test_sv_extsw_intpred(self):
+        # extsb, integer twin-pred mask: source is ~r3 (0b01), dest r3 (0b10)
+        # works as follows, where any zeros indicate "skip element"
+        #       - sources are 9 and 10
+        #       - dests are 5 and 6
+        #       - source mask says "pick first element from source (5)
+        #       - dest mask says "pick *second* element from dest (10)
+        # therefore the operation that's carried out is:
+        #       GPR(10) = extsb(GPR(5))
+        # this is a type of back-to-back VGATHER and VSCATTER but it applies
+        # to *operations*, not just MVs like in traditional Vector ISAs
+
+        isa = SVP64Asm(['svextsb/sm=~r3/m=r3 5.v, 9.v'
+                       ])
+        lst = list(isa)
+        print ("listing", lst)
+
+        # initial values in GPR regfile
+        initial_regs = [0] * 32
+        initial_regs[3] = 0b10   # predicate mask
+        initial_regs[9] = 0x91
+        initial_regs[10] = 0x90
+        # SVSTATE (in this case, VL=2)
+        svstate = SVP64State()
+        svstate.vl[0:7] = 2 # VL
+        svstate.maxvl[0:7] = 2 # MAXVL
+        print ("SVSTATE", bin(svstate.spr.asint()))
+        # copy before running
+        expected_regs = deepcopy(initial_regs)
+        expected_regs[5] = 0x0
+        expected_regs[6] = 0xffff_ffff_ffff_ff91
+
+        with Program(lst, bigendian=False) as program:
+            sim = self.run_tst_program(program, initial_regs, svstate)
+            self._check_regs(sim, expected_regs)
+
+    def tst_sv_add_intpred(self):
         # adds, integer predicated mask r3=0b10
         #       1 = 5 + 9   => not to be touched (skipped)
         #       2 = 6 + 10  => 0x3334 = 0x2223+0x1111
@@ -73,7 +109,7 @@ class DecoderTestCase(FHDLTestCase):
             sim = self.run_tst_program(program, initial_regs, svstate)
             self._check_regs(sim, expected_regs)
 
-    def test_sv_add_cr_pred(self):
+    def tst_sv_add_cr_pred(self):
         # adds, CR predicated mask CR4.eq = 1, CR5.eq = 0, invert (ne)
         #       1 = 5 + 9   => not to be touched (skipped)
         #       2 = 6 + 10  => 0x3334 = 0x2223+0x1111
-- 
2.30.2