From 90050202f57b22243f5d3dd434a81df2f89de9ed Mon Sep 17 00:00:00 2001
From: Quentin Carbonneaux <quentin@c9x.me>
Date: Tue, 1 Oct 2024 19:38:15 +0200
Subject: [PATCH] fix various codegen bugs on arm64

- dynamic allocations could generate
  bad 'and' instructions (for the
  and with -16 in salloc()).
- symbols used in w context would
  generate adrp and add instructions
  on wN registers while they seem to
  only work on xN registers.

Thanks to Rosie for reporting them.
---
 arm64/emit.c   | 19 ++++++++++++++-----
 test/isel5.ssa | 16 ++++++++++++++++
 2 files changed, 30 insertions(+), 5 deletions(-)
 create mode 100644 test/isel5.ssa

diff --git a/arm64/emit.c b/arm64/emit.c
index ffdc178..28cd6a5 100644
--- a/arm64/emit.c
+++ b/arm64/emit.c
@@ -160,7 +160,8 @@ emitf(char *s, Ins *i, E *e)
 	Ref r;
 	int k, c;
 	Con *pc;
-	uint n, sp;
+	uint64_t n;
+	uint sp;
 
 	fputc('\t', e->f);
 
@@ -217,10 +218,17 @@ emitf(char *s, Ins *i, E *e)
 				pc = &e->fn->con[r.val];
 				n = pc->bits.i;
 				assert(pc->type == CBits);
-				if (n & 0xfff000)
-					fprintf(e->f, "#%u, lsl #12", n>>12);
-				else
-					fprintf(e->f, "#%u", n);
+				if (n >> 24) {
+					assert(arm64_logimm(n, k));
+					fprintf(e->f, "#%"PRIu64, n);
+				} else if (n & 0xfff000) {
+					assert(!(n & ~0xfff000ull));
+					fprintf(e->f, "#%"PRIu64", lsl #12",
+						n>>12);
+				} else {
+					assert(!(n & ~0xfffull));
+					fprintf(e->f, "#%"PRIu64, n);
+				}
 				break;
 			}
 			break;
@@ -304,6 +312,7 @@ loadcon(Con *c, int r, int k, E *e)
 	rn = rname(r, k);
 	n = c->bits.i;
 	if (c->type == CAddr) {
+		rn = rname(r, Kl);
 		loadaddr(c, rn, e);
 		return;
 	}
diff --git a/test/isel5.ssa b/test/isel5.ssa
new file mode 100644
index 0000000..9c546d7
--- /dev/null
+++ b/test/isel5.ssa
@@ -0,0 +1,16 @@
+# make sure the local symbols used for
+# fp constants do not get a _ prefix
+# on apple arm hardware
+
+export function w $main() {
+@start
+	%r =d copy d_1.2
+	%x =w call $printf(l $fmt, ..., d %r)
+	ret 0
+}
+
+data $fmt = { b "%.06f\n", b 0 }
+
+# >>> output
+# 1.200000
+# <<<