From 97f167f227933d7fc64eaea1fd0f3c576de2dc33 Mon Sep 17 00:00:00 2001
From: Simon Perretta <simon.perretta@imgtec.com>
Date: Wed, 1 Jan 2025 21:53:54 +0000
Subject: [PATCH] pco, pygen: expose enhanced logical ops with optional mask

Signed-off-by: Simon Perretta <simon.perretta@imgtec.com>
Acked-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/36412>
---
 src/imagination/pco/pco_map.py      | 25 +++++++++++++++++++------
 src/imagination/pco/pco_ops.py      |  3 ++-
 src/imagination/pco/pco_trans_nir.c |  8 +++++++-
 src/imagination/pco/pco_validate.c  |  1 +
 4 files changed, 29 insertions(+), 8 deletions(-)

diff --git a/src/imagination/pco/pco_map.py b/src/imagination/pco/pco_map.py
index 73aa7c8d2d8..fdf04ef3135 100644
--- a/src/imagination/pco/pco_map.py
+++ b/src/imagination/pco/pco_map.py
@@ -27,6 +27,7 @@ REF_MAP = enum_type('ref_map', [
 
    ('ft0', 'ft0'),
    ('ft1', 'ft1'),
+   ('ft1_invert', '~ft1'),
    ('ft2', 'ft2'),
    ('fte', 'fte'),
 
@@ -1258,15 +1259,27 @@ encode_map(O_BBYP0S1,
    op_ref_maps=[('0', ['ft2'], ['s2'])]
 )
 
+encode_map(O_MSK_BBYP0S1,
+   encodings=[
+      (I_PHASE0_SRC, [
+         ('count_src', 's2'),
+         ('count_op', 'byp'),
+         ('bitmask_src_op', 'msk'),
+         ('shift1_op', 'byp')
+      ])
+   ],
+   op_ref_maps=[('0', ['ft0', 'ft1', 'ft2'], ['s0', 's1', 's2'])]
+)
+
 encode_map(O_LOGICAL,
    encodings=[
       (I_PHASE1, [
-         ('mskb', False),
-         ('mska', False),
+         ('mskb', ('!pco_ref_is_null', SRC(2))),
+         ('mska', ('!pco_ref_is_null', SRC(0))),
          ('logical_op', OM_LOGIOP),
       ])
    ],
-   op_ref_maps=[('1', ['ft4'], ['ft2', 's3'])]
+   op_ref_maps=[('1', ['ft4'], [['ft1', '_'], 'ft2', ['ft1_invert', '_'], 's3'])]
 )
 
 encode_map(O_SHIFT,
@@ -2105,12 +2118,12 @@ group_map(O_LOGICAL,
       ('rpt', OM_RPT)
    ]),
    enc_ops=[
-      ('0', O_BBYP0S1, ['ft2'], [SRC(0)]),
-      ('1', O_LOGICAL, [DEST(0)], ['ft2', SRC(1)], [(OM_LOGIOP, OM_LOGIOP)])
+      ('0', O_BBYP0S1, ['ft2'], [SRC(1)]),
+      ('1', O_LOGICAL, [DEST(0)], [SRC(0), 'ft2', SRC(2), SRC(3)], [(OM_LOGIOP, OM_LOGIOP)])
    ],
    srcs=[
       ('s[2]', ('0', SRC(0)), 's2'),
-      ('s[3]', ('1', SRC(1)), 's3')
+      ('s[3]', ('1', SRC(3)), 's3')
    ],
    dests=[('w[0]', ('1', DEST(0)), 'ft4')]
 )
diff --git a/src/imagination/pco/pco_ops.py b/src/imagination/pco/pco_ops.py
index 94d25c51e38..6bf3ccaef39 100644
--- a/src/imagination/pco/pco_ops.py
+++ b/src/imagination/pco/pco_ops.py
@@ -347,12 +347,13 @@ O_ATOMIC = hw_op('atomic', [OM_OLCHK, OM_EXEC_CND, OM_END, OM_ATOM_OP], 1, 2)
 ## Bitwise.
 O_MOVI32 = hw_op('movi32', OM_ALU, 1, 1)
 
-O_LOGICAL = hw_op('logical', OM_ALU + [OM_LOGIOP], 1, 2)
+O_LOGICAL = hw_op('logical', OM_ALU + [OM_LOGIOP], 1, 4)
 O_SHIFT = hw_op('shift', OM_ALU + [OM_SHIFTOP], 1, 3)
 
 O_BBYP0BM = hw_direct_op('bbyp0bm', [], 2, 2)
 O_BBYP0BM_IMM32 = hw_direct_op('bbyp0bm_imm32', [], 2, 2)
 O_BBYP0S1 = hw_direct_op('bbyp0s1', [], 1, 1)
+O_MSK_BBYP0S1 = hw_direct_op('msk_bbyp0s1', [], 3, 3)
 
 ## Control.
 O_WOP = hw_op('wop')
diff --git a/src/imagination/pco/pco_trans_nir.c b/src/imagination/pco/pco_trans_nir.c
index 0fe2d78837e..f65dc2130cc 100644
--- a/src/imagination/pco/pco_trans_nir.c
+++ b/src/imagination/pco/pco_trans_nir.c
@@ -1041,7 +1041,13 @@ static pco_instr *trans_logical(trans_ctx *tctx,
       UNREACHABLE("");
    }
 
-   return pco_logical(&tctx->b, dest, src0, src1, .logiop = logiop);
+   return pco_logical(&tctx->b,
+                      dest,
+                      pco_ref_null(),
+                      src0,
+                      pco_ref_null(),
+                      src1,
+                      .logiop = logiop);
 }
 
 /**
diff --git a/src/imagination/pco/pco_validate.c b/src/imagination/pco/pco_validate.c
index ce055e8dbd7..4ec56f04e2d 100644
--- a/src/imagination/pco/pco_validate.c
+++ b/src/imagination/pco/pco_validate.c
@@ -303,6 +303,7 @@ static inline bool ref_is_in_map(pco_ref ref, enum pco_ref_map ref_maps)
 
          CHECK_IO(FT0, ref)
          CHECK_IO(FT1, ref)
+         CHECK_IO(FT1_INVERT, ref)
          CHECK_IO(FT2, ref)
          CHECK_IO(FTE, ref)
          CHECK_IO(FT3, ref)