From e01dc7a5887757a5e1d6894cc1fd1305f97a3187 Mon Sep 17 00:00:00 2001
From: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Date: Wed, 20 Nov 2024 21:57:56 -0400
Subject: [PATCH] asahi/genxml: optimize out masking with shr

noticed in the agx asm.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32320>
---
 src/asahi/genxml/gen_pack.py | 21 +++++++++++++++++----
 1 file changed, 17 insertions(+), 4 deletions(-)

diff --git a/src/asahi/genxml/gen_pack.py b/src/asahi/genxml/gen_pack.py
index c2dbf987456..1d711a06e31 100644
--- a/src/asahi/genxml/gen_pack.py
+++ b/src/asahi/genxml/gen_pack.py
@@ -292,9 +292,22 @@ class Group(object):
                 if field.exact:
                     value = field.default
 
+                # These types all use util_bitpack_uint
+                pack_as_uint = field.type in ["uint", "hex", "address", "bool"]
+                pack_as_uint |= field.type in self.parser.enums
+                start_adjusted = start
+                value_unshifted = None
+
                 if field.modifier is not None:
                     if field.modifier[0] == "shr":
-                        value = f"{value} >> {field.modifier[1]}"
+                        if pack_as_uint and start >= field.modifier[1]:
+                            # For uint, we fast path.  If we do `(a >> 2) << 2`,
+                            # clang will generate a mask in release builds, even
+                            # though we know we're aligned. So don't generate
+                            # that to avoid the masking.
+                            start_adjusted = start - field.modifier[1]
+                        else:
+                            value = f"{value} >> {field.modifier[1]}"
                     elif field.modifier[0] == "minus":
                         value = f"{value} - {field.modifier[1]}"
                     elif field.modifier[0] == "align":
@@ -305,15 +318,15 @@ class Group(object):
                         value = "__gen_to_groups({}, {}, {})".format(value,
                                 field.modifier[1], end - start + 1)
 
-                if field.type in ["uint", "hex", "address", "bool"] or field.type in self.parser.enums:
-                    bits = (end - start + 1)
+                if pack_as_uint:
+                    bits = (end - start_adjusted + 1)
                     if bits < 64 and not field.exact:
                         # Add some nicer error checking
                         label = f"{self.label}::{name}"
                         bound = hex(1 << bits)
                         print(f"   agx_genxml_validate_bounds(\"{label}\", {value}, {bound}ull);")
 
-                    s = f"util_bitpack_uint({value}, {start}, {end})"
+                    s = f"util_bitpack_uint({value}, {start_adjusted}, {end})"
                 elif field.type == "int":
                     s = "util_bitpack_sint(%s, %d, %d)" % \
                         (value, start, end)