GH1141: Different fix, preferred by Richard.

[openssl.git] / crypto / sparcv9cap.c

diff --git a/crypto/sparcv9cap.c b/crypto/sparcv9cap.c
index e1e6d7395559c5787f419bd202c45cf628178bba..92841ce76a3a8c6bc4d77e91f6c8a08bef7f5973 100644 (file)
--- a/crypto/sparcv9cap.c
+++ b/crypto/sparcv9cap.c
@@ -1,3 +1,12 @@
+/*
+ * Copyright 2005-2016 The OpenSSL Project Authors. All Rights Reserved.
+ *
+ * Licensed under the OpenSSL license (the "License").  You may not use
+ * this file except in compliance with the License.  You can obtain a copy
+ * in the file LICENSE in the source distribution or at
+ * https://www.openssl.org/source/license.html
+ */
+
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
@@ -60,9 +69,18 @@ int bn_mul_mont(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp,
         if ((OPENSSL_sparcv9cap_P[0] & SPARCV9_VIS3))
             return bn_mul_mont_vis3(rp, ap, bp, np, n0, num);
         else if (num >= 8 &&
-                 (OPENSSL_sparcv9cap_P[0] &
-                  (SPARCV9_PREFER_FPU | SPARCV9_VIS1)) ==
-                 (SPARCV9_PREFER_FPU | SPARCV9_VIS1))
+                 /*
+                  * bn_mul_mont_fpu doesn't use FMADD, we just use the
+                  * flag to detect when FPU path is preferable in cases
+                  * when current heuristics is unreliable. [it works
+                  * out because FMADD-capable processors where FPU
+                  * code path is undesirable are also VIS3-capable and
+                  * VIS3 code path takes precedence.]
+                  */
+                 ( (OPENSSL_sparcv9cap_P[0] & SPARCV9_FMADD) ||
+                   (OPENSSL_sparcv9cap_P[0] &
+                    (SPARCV9_PREFER_FPU | SPARCV9_VIS1)) ==
+                   (SPARCV9_PREFER_FPU | SPARCV9_VIS1) ))
             return bn_mul_mont_fpu(rp, ap, bp, np, n0, num);
     }
     return bn_mul_mont_int(rp, ap, bp, np, n0, num);
@@ -149,17 +167,24 @@ void OPENSSL_cpuid_setup(void)
         unsigned int vec[1];
 
         if (getisax (vec,1)) {
-            if (vec[0]&0x0020)  OPENSSL_sparcv9cap_P[0] |= SPARCV9_VIS1;
-            if (vec[0]&0x0040)  OPENSSL_sparcv9cap_P[0] |= SPARCV9_VIS2;
-            if (vec[0]&0x0080)  OPENSSL_sparcv9cap_P[0] |= SPARCV9_BLK;
-            if (vec[0]&0x0100)  OPENSSL_sparcv9cap_P[0] |= SPARCV9_FMADD;
-            if (vec[0]&0x0400)  OPENSSL_sparcv9cap_P[0] |= SPARCV9_VIS3;
+            if (vec[0]&0x00020) OPENSSL_sparcv9cap_P[0] |= SPARCV9_VIS1;
+            if (vec[0]&0x00040) OPENSSL_sparcv9cap_P[0] |= SPARCV9_VIS2;
+            if (vec[0]&0x00080) OPENSSL_sparcv9cap_P[0] |= SPARCV9_BLK;
+            if (vec[0]&0x00100) OPENSSL_sparcv9cap_P[0] |= SPARCV9_FMADD;
+            if (vec[0]&0x00400) OPENSSL_sparcv9cap_P[0] |= SPARCV9_VIS3;
+            if (vec[0]&0x01000) OPENSSL_sparcv9cap_P[0] |= SPARCV9_FJHPCACE;
+            if (vec[0]&0x02000) OPENSSL_sparcv9cap_P[0] |= SPARCV9_FJDESX;
+            if (vec[0]&0x08000) OPENSSL_sparcv9cap_P[0] |= SPARCV9_IMA;
             if (vec[0]&0x10000) OPENSSL_sparcv9cap_P[0] |= SPARCV9_FJAESX;
+            if (vec[1]&0x00008) OPENSSL_sparcv9cap_P[0] |= SPARCV9_VIS4;
 
             /* reconstruct %cfr copy */
             OPENSSL_sparcv9cap_P[1] = (vec[0]>>17)&0x3ff;
             OPENSSL_sparcv9cap_P[1] |= (OPENSSL_sparcv9cap_P[1]&CFR_MONTMUL)<<1;
             if (vec[0]&0x20000000) OPENSSL_sparcv9cap_P[1] |= CFR_CRC32C;
+            if (vec[1]&0x00000020) OPENSSL_sparcv9cap_P[1] |= CFR_XMPMUL;
+            if (vec[1]&0x00000040)
+                OPENSSL_sparcv9cap_P[1] |= CFR_XMONTMUL|CFR_XMONTSQR;
 
             /* Some heuristics */
             /* all known VIS2-capable CPUs have unprivileged tick counter */