Unified - adapt the generation of cpuid, uplink and buildinf to use GENERATE

[openssl.git] / crypto / x86cpuid.pl

diff --git a/crypto/x86cpuid.pl b/crypto/x86cpuid.pl
index e8eaef758262343b541942476157b644eac2e7d8..48c4cf204b73b9d849c97b482d5f3edaa6232c40 100644 (file)
--- a/crypto/x86cpuid.pl
+++ b/crypto/x86cpuid.pl
@@ -4,6 +4,10 @@ $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
 push(@INC, "${dir}perlasm", "perlasm");
 require "x86asm.pl";
 
+$output = pop;
+open OUT,">$output";
+*STDOUT=*OUT;
+
 &asm_init($ARGV[0],"x86cpuid");
 
 for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
@@ -19,9 +23,11 @@ for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
        &pushf  ();
        &pop    ("eax");
        &xor    ("ecx","eax");
-       &bt     ("ecx",21);
-       &jnc    (&label("generic"));
        &xor    ("eax","eax");
+       &bt     ("ecx",21);
+       &jnc    (&label("nocpuid"));
+       &mov    ("esi",&wparam(0));
+       &mov    (&DWP(8,"esi"),"eax");  # clear 3rd word
        &cpuid  ();
        &mov    ("edi","eax");          # max value for standard query level
 
@@ -67,6 +73,7 @@ for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
        &inc    ("esi");                # number of cores
 
        &mov    ("eax",1);
+       &xor    ("ecx","ecx");
        &cpuid  ();
        &bt     ("edx",28);
        &jnc    (&label("generic"));
@@ -78,6 +85,16 @@ for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
        &jmp    (&label("generic"));
        
 &set_label("intel");
+       &cmp    ("edi",7);
+       &jb     (&label("cacheinfo"));
+
+       &mov    ("esi",&wparam(0));
+       &mov    ("eax",7);
+       &xor    ("ecx","ecx");
+       &cpuid  ();
+       &mov    (&DWP(8,"esi"),"ebx");
+
+&set_label("cacheinfo");
        &cmp    ("edi",4);
        &mov    ("edi",-1);
        &jb     (&label("nocacheinfo"));
@@ -91,6 +108,7 @@ for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
 
 &set_label("nocacheinfo");
        &mov    ("eax",1);
+       &xor    ("ecx","ecx");
        &cpuid  ();
        &and    ("edx",0xbfefffff);     # force reserved bits #20, #30 to 0
        &cmp    ("ebp",0);
@@ -133,9 +151,12 @@ for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
        &and    ("esi",0xfeffffff);     # clear FXSR
 &set_label("clear_avx");
        &and    ("ebp",0xefffe7ff);     # clear AVX, FMA and AMD XOP bits
+       &mov    ("edi",&wparam(0));
+       &and    (&DWP(8,"edi"),0xffffffdf);     # clear AVX2
 &set_label("done");
        &mov    ("eax","esi");
        &mov    ("edx","ebp");
+&set_label("nocpuid");
 &function_end("OPENSSL_ia32_cpuid");
 
 &external_label("OPENSSL_ia32cap_P");
@@ -164,7 +185,7 @@ for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
        &jnz    (&label("nohalt"));     # not enough privileges
 
        &pushf  ();
-       &pop    ("eax")
+       &pop    ("eax");
        &bt     ("eax",9);
        &jnc    (&label("nohalt"));     # interrupts are disabled
 
@@ -195,7 +216,7 @@ for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
 
 &function_begin_B("OPENSSL_far_spin");
        &pushf  ();
-       &pop    ("eax")
+       &pop    ("eax");
        &bt     ("eax",9);
        &jnc    (&label("nospin"));     # interrupts are disabled
 
@@ -279,7 +300,7 @@ for (@ARGV) { $sse2=1 if (/-DOPENSSL_IA32_SSE2/); }
 #      arguments is 1 or 2!
 &function_begin_B("OPENSSL_indirect_call");
        {
-       my $i,$max=7;           # $max has to be chosen as 4*n-1
+       my ($max,$i)=(7,);      # $max has to be chosen as 4*n-1
                                # in order to preserve eventual
                                # stack alignment
        &push   ("ebp");
@@ -452,6 +473,23 @@ my $max = "ebp";
        &ret    ();
 &function_end_B("OPENSSL_ia32_rdrand");
 
+&function_begin_B("OPENSSL_ia32_rdseed");
+       &mov    ("ecx",8);
+&set_label("loop");
+       &rdseed ("eax");
+       &jc     (&label("break"));
+       &loop   (&label("loop"));
+&set_label("break");
+       &cmp    ("eax",0);
+       &cmove  ("eax","ecx");
+       &ret    ();
+&function_end_B("OPENSSL_ia32_rdseed");
+
 &initseg("OPENSSL_cpuid_setup");
 
+&hidden("OPENSSL_cpuid_setup");
+&hidden("OPENSSL_ia32cap_P");
+
 &asm_finish();
+
+close STDOUT;