From b0cfcca9904a6c3d106edb2de2e8d14b805dda08 Mon Sep 17 00:00:00 2001
From: Slava Pestov <slava@factorcode.org>
Date: Sun, 11 Dec 2005 03:48:09 +0000
Subject: [PATCH] fix AMD64 %fixnum* overflow

---
 library/compiler/basic-blocks.factor    |  8 ++++++++
 library/compiler/ppc/fixnum.factor      |  2 +-
 library/compiler/x86/fixnum.factor      |  2 +-
 library/test/compiler/intrinsics.factor | 10 ++++++----
 native/s48_bignum.c                     | 15 +++++++++++++--
 native/s48_bignum.h                     |  1 +
 6 files changed, 30 insertions(+), 8 deletions(-)

diff --git a/library/compiler/basic-blocks.factor b/library/compiler/basic-blocks.factor
index d9488972ad..b1c3c5ad89 100644
--- a/library/compiler/basic-blocks.factor
+++ b/library/compiler/basic-blocks.factor
@@ -5,6 +5,14 @@ USING: arrays hashtables kernel lists math namespaces sequences ;
 
 ! Optimizations performed here:
 ! - combining %inc-d/%inc-r within a single basic block
+! - if a literal is loaded into a vreg but the vreg is
+!   overwritten before being read, the literal load is deleted
+! - if a %replace is writing a vreg to a stack location already
+!   holding that vreg, or a stack location that is not read
+!   before being popped, the %replace is deleted
+! - if a %peek is reading a stack location into a vreg that
+!   already holds that vreg, or if the vreg is overwritten
+!   before being read, the %peek is deleted
 ! - removing dead loads of stack locations into vregs
 ! - removing dead stores of vregs into stack locations
 
diff --git a/library/compiler/ppc/fixnum.factor b/library/compiler/ppc/fixnum.factor
index 1aa2d39eb7..8d26dc9118 100644
--- a/library/compiler/ppc/fixnum.factor
+++ b/library/compiler/ppc/fixnum.factor
@@ -36,7 +36,7 @@ M: %fixnum* generate-node ( vop -- )
     "end" get BNO
     >3-vop< MULHW
     4 0 scratch MR
-    "s48_long_long_to_bignum" f compile-c-call
+    "s48_long_pair_to_bignum" f compile-c-call
     ! now we have to shift it by three bits to remove the second
     ! tag
     tag-bits neg 4 LI
diff --git a/library/compiler/x86/fixnum.factor b/library/compiler/x86/fixnum.factor
index 28c5495a9b..8081cd6d6f 100644
--- a/library/compiler/x86/fixnum.factor
+++ b/library/compiler/x86/fixnum.factor
@@ -44,7 +44,7 @@ M: %fixnum* generate-node ( vop -- )
     0 input-operand IMUL
     <label> "end" set
     "end" get JNO
-    "s48_long_long_to_bignum" f
+    "s48_long_pair_to_bignum" f
     1 input-operand remainder-reg 2array compile-c-call*
     ! now we have to shift it by three bits to remove the second
     ! tag
diff --git a/library/test/compiler/intrinsics.factor b/library/test/compiler/intrinsics.factor
index e30d57ccee..ab4519f587 100644
--- a/library/test/compiler/intrinsics.factor
+++ b/library/test/compiler/intrinsics.factor
@@ -194,6 +194,12 @@ cell 8 = [
     [ t ] [ 1 59 fixnum-shift dup [ fixnum+ ] compile-1 1 60 fixnum-shift = ] unit-test
     [ -1152921504606846977 ] [ 1 60 shift neg >fixnum [ -1 fixnum+ ] compile-1 ] unit-test
     
+    [ t ] [ 1 40 shift 1 40 shift [ fixnum* ] compile-1 1 80 shift = ] unit-test
+    [ t ] [ 1 40 shift neg 1 40 shift [ fixnum* ] compile-1 1 80 shift neg = ] unit-test
+    [ t ] [ 1 40 shift neg 1 40 shift neg [ fixnum* ] compile-1 1 80 shift = ] unit-test
+    [ t ] [ 1 30 shift neg 1 50 shift neg [ fixnum* ] compile-1 1 80 shift = ] unit-test
+    [ t ] [ 1 50 shift neg 1 30 shift neg [ fixnum* ] compile-1 1 80 shift = ] unit-test
+    
     [ 18446744073709551616 ] [ 1 64 [ fixnum-shift ] compile-1 ] unit-test
     [ 18446744073709551616 ] [ 1 [ 64 fixnum-shift ] compile-1 ] unit-test
     [ 18446744073709551616 ] [ 1 [ 32 fixnum-shift 32 fixnum-shift ] compile-1 ] unit-test
@@ -201,10 +207,6 @@ cell 8 = [
     [ -18446744073709551616 ] [ -1 [ 64 fixnum-shift ] compile-1 ] unit-test
     [ -18446744073709551616 ] [ -1 [ 32 fixnum-shift 32 fixnum-shift ] compile-1 ] unit-test
     
-    [ t ] [ 1 40 shift 1 40 shift [ fixnum* ] compile-1 1 80 shift = ] unit-test
-    [ t ] [ 1 40 shift neg 1 40 shift [ fixnum* ] compile-1 1 80 shift neg = ] unit-test
-    [ t ] [ 1 40 shift neg 1 40 shift neg [ fixnum* ] compile-1 1 80 shift = ] unit-test
-    
     [ 1152921504606846976 ] [ -1152921504606846976 >fixnum -1 [ fixnum/i ] compile-1 ] unit-test
 
     [ 1152921504606846976 0 ] [ -1152921504606846976 >fixnum -1 [ fixnum/mod ] compile-1 ] unit-test
diff --git a/native/s48_bignum.c b/native/s48_bignum.c
index ba801294d8..f1ac989d96 100644
--- a/native/s48_bignum.c
+++ b/native/s48_bignum.c
@@ -381,6 +381,17 @@ FOO_TO_BIGNUM(ulong,unsigned long,unsigned long)
 FOO_TO_BIGNUM(long_long,s64,u64)
 FOO_TO_BIGNUM(ulong_long,u64,u64)
 
+/* this is inefficient; its only used for fixnum multiplication overflow so
+it probaly does not matter */
+bignum_type s48_long_pair_to_bignum(unsigned long x, long y)
+{
+  return s48_bignum_add(
+    s48_bignum_arithmetic_shift(
+      s48_long_to_bignum(y),
+      sizeof(unsigned long) * 8),
+    s48_ulong_to_bignum(x));
+}
+
 #define BIGNUM_TO_FOO(name,type,utype) \
   type s48_bignum_to_##name(bignum_type bignum)                                     \
   {                                                                                 \
@@ -941,7 +952,7 @@ bignum_divide_unsigned_normalized(bignum_type u, bignum_type v, bignum_type q)
   bignum_digit_type * u_scan_start = (u_scan - v_length);
   bignum_digit_type * v_start = (BIGNUM_START_PTR (v));
   bignum_digit_type * v_end = (v_start + v_length);
-  bignum_digit_type * q_scan;
+  bignum_digit_type * q_scan = NULL;
   bignum_digit_type v1 = (v_end[-1]);
   bignum_digit_type v2 = (v_end[-2]);
   bignum_digit_type ph;        /* high half of double-digit product */
@@ -1615,7 +1626,7 @@ s48_bignum_bitwise_xor(bignum_type arg1, bignum_type arg2)
 bignum_type
 bignum_magnitude_ash(bignum_type arg1, long n)
 {
-  bignum_type result;
+  bignum_type result = NULL;
   bignum_digit_type *scan1;
   bignum_digit_type *scanr;
   bignum_digit_type *end;
diff --git a/native/s48_bignum.h b/native/s48_bignum.h
index d28a485a01..fc0b099414 100644
--- a/native/s48_bignum.h
+++ b/native/s48_bignum.h
@@ -69,6 +69,7 @@ DLLEXPORT bignum_type s48_long_to_bignum(long);
 DLLEXPORT bignum_type s48_long_long_to_bignum(s64 n);
 DLLEXPORT bignum_type s48_ulong_long_to_bignum(u64 n);
 DLLEXPORT bignum_type s48_ulong_to_bignum(unsigned long);
+DLLEXPORT bignum_type s48_long_pair_to_bignum(unsigned long x, long y);
 long s48_bignum_to_long(bignum_type);
 unsigned long s48_bignum_to_ulong(bignum_type);
 s64 s48_bignum_to_long_long(bignum_type);