From 852d3a691d22f664e24741ffe5cb545c96ce5cdb Mon Sep 17 00:00:00 2001
From: David Given <dg@cowlark.com>
Date: Fri, 11 Nov 2016 21:48:36 +0100
Subject: [PATCH] Update the table to return call output values in the right
 registers. Fix the register allocator so the corrupted registers only apply
 to throughs (otherwise, you can't put output registers in corrupted
 registers).

---
 mach/powerpc/mcg/table                  | 22 +++++++++++-----------
 mach/proto/mcg/pass_registerallocator.c |  8 +++-----
 2 files changed, 14 insertions(+), 16 deletions(-)

diff --git a/mach/powerpc/mcg/table b/mach/powerpc/mcg/table
index 486d983ed..a6f416b65 100644
--- a/mach/powerpc/mcg/table
+++ b/mach/powerpc/mcg/table
@@ -19,7 +19,7 @@ REGISTERS
     r6                                            int volatile;
     r5                                            int volatile;
     r4                                            int volatile;
-    r3                                            int volatile ret;
+    r3                                            int volatile iret;
 
     r31                                           int; 
     r30                                           int; 
@@ -185,7 +185,7 @@ PATTERNS
 		emit "addi sp, sp, 8"
 		cost 8;
 		
-	SETRET.I(in:(ret)reg)
+	SETRET.I(in:(iret)reg)
 		emit "! setret4"
 		cost 1;
 
@@ -392,12 +392,12 @@ PATTERNS
         emit "li32 %out.1, 0"
         cost 8;
 
-    out:(ret)reg = FROMF.I(in:(dret)reg)
+    out:(iret)reg = FROMF.I(in:(dret)reg)
         with corrupted(volatile)
         emit "bl .fromf2i"
         cost 4;
 
-    out:(ret)reg = FROMD.I(in:(dret)reg)
+    out:(iret)reg = FROMD.I(in:(dret)reg)
         with corrupted(volatile)
         emit "bl .fromd2i"
         cost 4;
@@ -407,17 +407,17 @@ PATTERNS
         emit "bl .fromf2l"
         cost 4;
 
-    out:(dret)reg = FROMSI.D(in:(ret)reg)
+    out:(dret)reg = FROMSI.D(in:(iret)reg)
         with corrupted(volatile)
         emit "bl .fromsi2d"
         cost 4;
 
-    out:(fret)reg = FROMUI.F(in:(ret)reg)
+    out:(fret)reg = FROMUI.F(in:(iret)reg)
         with corrupted(volatile)
         emit "bl .fromui2f"
         cost 4;
 
-    out:(dret)reg = FROMUI.D(in:(ret)reg)
+    out:(dret)reg = FROMUI.D(in:(iret)reg)
         with corrupted(volatile)
         emit "bl .fromui2d"
         cost 4;
@@ -493,8 +493,8 @@ PATTERNS
             cost 4;
 
     CALLLABEL(CALL)
-    out:(int)reg = CALLLABEL(CALL.I)
-    out:(long)reg = CALLLABEL(CALL.L)
+    out:(iret)reg = CALLLABEL(CALL.I)
+    out:(lret)reg = CALLLABEL(CALL.L)
 
     #define CALLINDIRECT(insn) \
         insn (dest:(int)reg) \
@@ -504,8 +504,8 @@ PATTERNS
             cost 8;
 
     CALLINDIRECT(CALL)
-    out:(int)reg = CALLINDIRECT(CALL.I)
-    out:(long)reg = CALLINDIRECT(CALL.L)
+    out:(iret)reg = CALLINDIRECT(CALL.I)
+    out:(lret)reg = CALLINDIRECT(CALL.L)
 
     JUMP(dest:LABEL.I)
         emit "b $dest"
diff --git a/mach/proto/mcg/pass_registerallocator.c b/mach/proto/mcg/pass_registerallocator.c
index 7c348f0a3..a064adb8e 100644
--- a/mach/proto/mcg/pass_registerallocator.c
+++ b/mach/proto/mcg/pass_registerallocator.c
@@ -153,8 +153,7 @@ static bool allocatable_stackable_input(struct hreg* hreg, struct vreg* vreg)
 static bool allocatable_stackable_output(struct hreg* hreg, struct vreg* vreg)
 {
     return !register_used(current_outs, hreg) &&
-        (hreg->attrs & vreg->type) &&
-        !(hreg->attrs & current_hop->insndata->corrupts);
+        (hreg->attrs & vreg->type);
 }
 
 static bool allocatable_input(struct hreg* hreg, struct vreg* vreg)
@@ -174,7 +173,8 @@ static bool allocatable_output(struct hreg* hreg, struct vreg* vreg)
 static bool allocatable_through(struct hreg* hreg, struct vreg* vreg)
 {
     return allocatable_stackable_input(hreg, vreg) &&
-        allocatable_stackable_output(hreg, vreg);
+        allocatable_stackable_output(hreg, vreg) &&
+        !(hreg->attrs & current_hop->insndata->corrupts);
 }
 
 static struct hreg* find_input_reg(struct vreg* vreg)
@@ -203,9 +203,7 @@ static struct hreg* find_output_reg(struct vreg* vreg)
     {
         hreg = hregs.item[i];
         if (allocatable_output(hreg, vreg))
-        {
             return hreg;
-        }
     }
 
     return NULL;