PPC assebmly pack: initial POWER9 support tidbits.

Reviewed-by: Richard Levitte <levitte@openssl.org>

diff --git a/crypto/perlasm/ppc-xlate.pl b/crypto/perlasm/ppc-xlate.pl
index 0f46cf06bcb859f50f140443bd1711e91886c768..bc79b4626c4d6b0dfa5a530a02163c2e966cdd9f 100755 (executable)
--- a/crypto/perlasm/ppc-xlate.pl
+++ b/crypto/perlasm/ppc-xlate.pl
@@ -207,6 +207,21 @@ my $mtsle  = sub {
     "  .long   ".sprintf "0x%X",(31<<26)|($arg<<21)|(147*2);
 };
 
     "  .long   ".sprintf "0x%X",(31<<26)|($arg<<21)|(147*2);
 };
 

+# PowerISA 3.0 stuff
+my $maddhdu = sub {
+    my ($f, $rt, $ra, $rb, $rc) = @_;
+    "  .long   ".sprintf "0x%X",(4<<26)|($rt<<21)|($ra<<16)|($rb<<11)|($rc<<6)|49;
+};
+my $maddld = sub {
+    my ($f, $rt, $ra, $rb, $rc) = @_;
+    "  .long   ".sprintf "0x%X",(4<<26)|($rt<<21)|($ra<<16)|($rb<<11)|($rc<<6)|51;
+};
+
+my $darn = sub {
+    my ($f, $rt, $l) = @_;
+    "  .long   ".sprintf "0x%X",(31<<26)|($rt<<21)|($l<<16)|(755<<1);
+};
+

 while($line=<>) {
 
     $line =~ s|[#!;].*$||;     # get rid of asm-style comments...
 while($line=<>) {
 
     $line =~ s|[#!;].*$||;     # get rid of asm-style comments...

diff --git a/crypto/ppc_arch.h b/crypto/ppc_arch.h
index c0b4f1846bbd1110cb6c510f7fe642f267e4f494..a27fb728df6e27130d985daebb8aaacc50329856 100644 (file)
--- a/crypto/ppc_arch.h
+++ b/crypto/ppc_arch.h
@@ -12,5 +12,6 @@ extern unsigned int OPENSSL_ppccap_P;
 # define PPC_ALTIVEC     (1<<1)
 # define PPC_CRYPTO207   (1<<2)
 # define PPC_FPU         (1<<3)
 # define PPC_ALTIVEC     (1<<1)
 # define PPC_CRYPTO207   (1<<2)
 # define PPC_FPU         (1<<3)

+# define PPC_MADD300     (1<<4)

 
 #endif
 
 #endif

diff --git a/crypto/ppccap.c b/crypto/ppccap.c
index 76cfdbde8b7ab3f192008dd4af5fb720aeca8c14..2c98daf5b4df785d88269b8519dd9cbcc2942e18 100644 (file)
--- a/crypto/ppccap.c
+++ b/crypto/ppccap.c
@@ -151,6 +151,7 @@ static unsigned long (*getauxval) (unsigned long) = NULL;
 
 #define HWCAP2                  26      /* AT_HWCAP2 */
 #define HWCAP_VEC_CRYPTO        (1U << 25)
 
 #define HWCAP2                  26      /* AT_HWCAP2 */
 #define HWCAP_VEC_CRYPTO        (1U << 25)

+#define HWCAP_ARCH_3_00         (1U << 23)

 
 # if defined(__GNUC__) && __GNUC__>=2
 __attribute__ ((constructor))
 
 # if defined(__GNUC__) && __GNUC__>=2
 __attribute__ ((constructor))


@@ -208,6 +209,9 @@ void OPENSSL_cpuid_setup(void)
     if (__power_set(0xffffffffU<<16))           /* POWER8 and later */
         OPENSSL_ppccap_P |= PPC_CRYPTO207;
 
     if (__power_set(0xffffffffU<<16))           /* POWER8 and later */
         OPENSSL_ppccap_P |= PPC_CRYPTO207;
 

+    if (__power_set(0xffffffffU<<17))           /* POWER9 and later */
+        OPENSSL_ppccap_P |= PPC_MADD300;
+

     return;
 # endif
 #endif
     return;
 # endif
 #endif


@@ -236,6 +240,10 @@ void OPENSSL_cpuid_setup(void)
                 OPENSSL_ppccap_P |= PPC_CRYPTO207;
         }
 
                 OPENSSL_ppccap_P |= PPC_CRYPTO207;
         }
 

+        if (hwcap & HWCAP_ARCH_3_00) {
+            OPENSSL_ppccap_P |= PPC_MADD300;
+        }
+

         return;
     }
 
         return;
     }
 


@@ -285,6 +293,11 @@ void OPENSSL_cpuid_setup(void)
         }
     }
 
         }
     }
 

+    if (sigsetjmp(ill_jmp, 1) == 0) {
+        OPENSSL_madd300_probe();
+        OPENSSL_ppccap_P |= PPC_MADD300;
+    }
+

     sigaction(SIGILL, &ill_oact, NULL);
     sigprocmask(SIG_SETMASK, &oset, NULL);
 }
     sigaction(SIGILL, &ill_oact, NULL);
     sigprocmask(SIG_SETMASK, &oset, NULL);
 }

diff --git a/crypto/ppccpuid.pl b/crypto/ppccpuid.pl
index 4c2530d147ae9aa61c15a6a94e77b02909de482c..9de1355b2f8f0d4a15bdc0e386bc89fd3b96b4c6 100755 (executable)
--- a/crypto/ppccpuid.pl
+++ b/crypto/ppccpuid.pl
@@ -60,6 +60,16 @@ $code=<<___;
        .byte   0,12,0x14,0,0,0,0,0
 .size  .OPENSSL_crypto207_probe,.-.OPENSSL_crypto207_probe
 
        .byte   0,12,0x14,0,0,0,0,0
 .size  .OPENSSL_crypto207_probe,.-.OPENSSL_crypto207_probe
 

+.globl .OPENSSL_madd300_probe
+.align 4
+.OPENSSL_madd300_probe:
+       xor     r0,r0,r0
+       maddld  r3,r0,r0,r0
+       maddhdu r3,r0,r0,r0
+       blr
+       .long   0
+       .byte   0,12,0x14,0,0,0,0,0
+

 .globl .OPENSSL_wipe_cpu
 .align 4
 .OPENSSL_wipe_cpu:
 .globl .OPENSSL_wipe_cpu
 .align 4
 .OPENSSL_wipe_cpu: