From 1d9489a76ea8085640fb1835aae709a912eb964c Mon Sep 17 00:00:00 2001
From: Joel Dice <joel.dice@gmail.com>
Date: Tue, 30 Nov 2010 16:58:51 -0700
Subject: [PATCH] fix a couple of 64-bit-op-on-32-bit-system bugs

We weren't properly handling the case where a 64-bit value is
multipled with itself in multiplyRR, leading to wrong code.  Also,
addCarryCR didn't know how to handle constants more than 8-bits wide.
---
 src/x86.cpp     | 28 ++++++++++++++++++++++------
 test/Longs.java | 17 +++++++++++++++++
 2 files changed, 39 insertions(+), 6 deletions(-)

diff --git a/src/x86.cpp b/src/x86.cpp
index 4e67576f1f..65ac67ce30 100644
--- a/src/x86.cpp
+++ b/src/x86.cpp
@@ -1403,17 +1403,18 @@ addRR(Context* c, unsigned aSize, Assembler::Register* a,
 }
 
 void
-addCarryCR(Context* c, unsigned size UNUSED, Assembler::Constant* a,
+addCarryCR(Context* c, unsigned size, Assembler::Constant* a,
            Assembler::Register* b)
 {
   
   int64_t v = a->value->value();
+  maybeRex(c, size, b);
   if (isInt8(v)) {
-    maybeRex(c, size, b);
     opcode(c, 0x83, 0xd0 + regCode(b));
     c->code.append(v);
   } else {
-    abort(c);
+    opcode(c, 0x81, 0xd0 + regCode(b));
+    c->code.append4(v);
   }
 }
 
@@ -1740,16 +1741,31 @@ multiplyRR(Context* c, unsigned aSize, Assembler::Register* a,
     Assembler::Register ah(a->high);
     Assembler::Register bh(b->high);
 
+    Assembler::Register tmp(-1);
+    Assembler::Register* scratch;
+    if (a->low == b->low) {
+      tmp.low = c->client->acquireTemporary
+        (GeneralRegisterMask & ~(1 << rax));
+      scratch = &tmp;
+      moveRR(c, 4, b, 4, scratch);
+    } else {
+      scratch = b;
+    }
+
     moveRR(c, 4, b, 4, &axdx);
-    multiplyRR(c, 4, &ah, 4, b);
+    multiplyRR(c, 4, &ah, 4, scratch);
     multiplyRR(c, 4, a, 4, &bh);
-    addRR(c, 4, &bh, 4, b);
+    addRR(c, 4, &bh, 4, scratch);
     
     // mul a->low,%eax%edx
     opcode(c, 0xf7, 0xe0 + a->low);
     
-    addRR(c, 4, b, 4, &bh);
+    addRR(c, 4, scratch, 4, &bh);
     moveRR(c, 4, &axdx, 4, b);
+
+    if (tmp.low != -1) {
+      c->client->releaseTemporary(tmp.low);
+    }
   } else {
     maybeRex(c, aSize, b, a);
     opcode(c, 0x0f, 0xaf);
diff --git a/test/Longs.java b/test/Longs.java
index 7565b42121..8c34fddce9 100644
--- a/test/Longs.java
+++ b/test/Longs.java
@@ -243,6 +243,23 @@ public class Longs {
     expect(buffer.getLong() == 231);
 
     expect(unsignedShiftRight32(231) == 0);
+
+    { int[] x = new int[] { 1701899151 };
+      int[] z = new int[x.length * 2];
+      final long LONG_MASK = 0xffffffffL;
+
+      int lastProductLowWord = 0;
+      for (int j=0, i=0; j<x.length; j++) {
+        long piece = (x[j] & LONG_MASK);
+        long product = piece * piece;
+        z[i++] = (lastProductLowWord << 31) | (int) (product >>> 33);
+        z[i++] = (int) (product >>> 1);
+        lastProductLowWord = (int) product;
+      }
+
+      expect(z[0] == 337192406);
+      expect(z[1] == -437261072);
+    }
   }
 
 }