Blame SOURCES/valgrind-3.15.0-avx-rdrand-f16c.patch

01cf8b
commit 791fe5ecf909d573bcbf353b677b9404f9da0ed4
01cf8b
Author: Mark Wielaard <mark@klomp.org>
01cf8b
Date:   Mon May 27 22:19:27 2019 +0200
01cf8b
01cf8b
    Expose rdrand and f16c through cpuid also if the host only has avx.
01cf8b
    
01cf8b
    The amd64 CPUID dirtyhelpers are mostly static since they emulate some
01cf8b
    existing CPU "family". The avx2 ("i7-4910MQ") CPUID variant however
01cf8b
    can "dynamicly" enable rdrand and/or f16c if the host supports them.
01cf8b
    Do the same for the avx_and_cx16 ("i5-2300") CPUID variant.
01cf8b
    
01cf8b
    https://bugs.kde.org/show_bug.cgi?id=408009
01cf8b
01cf8b
diff --git a/VEX/priv/guest_amd64_defs.h b/VEX/priv/guest_amd64_defs.h
01cf8b
index 4f34b41..a5de527 100644
01cf8b
--- a/VEX/priv/guest_amd64_defs.h
01cf8b
+++ b/VEX/priv/guest_amd64_defs.h
01cf8b
@@ -165,7 +165,9 @@ extern void  amd64g_dirtyhelper_storeF80le ( Addr/*addr*/, ULong/*data*/ );
01cf8b
 extern void  amd64g_dirtyhelper_CPUID_baseline ( VexGuestAMD64State* st );
01cf8b
 extern void  amd64g_dirtyhelper_CPUID_sse3_and_cx16 ( VexGuestAMD64State* st );
01cf8b
 extern void  amd64g_dirtyhelper_CPUID_sse42_and_cx16 ( VexGuestAMD64State* st );
01cf8b
-extern void  amd64g_dirtyhelper_CPUID_avx_and_cx16 ( VexGuestAMD64State* st );
01cf8b
+extern void  amd64g_dirtyhelper_CPUID_avx_and_cx16 ( VexGuestAMD64State* st,
01cf8b
+                                                     ULong hasF16C,
01cf8b
+                                                     ULong hasRDRAND );
01cf8b
 extern void  amd64g_dirtyhelper_CPUID_avx2 ( VexGuestAMD64State* st,
01cf8b
                                              ULong hasF16C, ULong hasRDRAND );
01cf8b
 
01cf8b
diff --git a/VEX/priv/guest_amd64_helpers.c b/VEX/priv/guest_amd64_helpers.c
01cf8b
index e4cf7e2..182bae0 100644
01cf8b
--- a/VEX/priv/guest_amd64_helpers.c
01cf8b
+++ b/VEX/priv/guest_amd64_helpers.c
01cf8b
@@ -3141,8 +3141,11 @@ void amd64g_dirtyhelper_CPUID_sse42_and_cx16 ( VexGuestAMD64State* st )
01cf8b
    address sizes   : 36 bits physical, 48 bits virtual
01cf8b
    power management:
01cf8b
 */
01cf8b
-void amd64g_dirtyhelper_CPUID_avx_and_cx16 ( VexGuestAMD64State* st )
01cf8b
+void amd64g_dirtyhelper_CPUID_avx_and_cx16 ( VexGuestAMD64State* st,
01cf8b
+                                             ULong hasF16C, ULong hasRDRAND )
01cf8b
 {
01cf8b
+   vassert((hasF16C >> 1) == 0ULL);
01cf8b
+   vassert((hasRDRAND >> 1) == 0ULL);
01cf8b
 #  define SET_ABCD(_a,_b,_c,_d)                \
01cf8b
       do { st->guest_RAX = (ULong)(_a);        \
01cf8b
            st->guest_RBX = (ULong)(_b);        \
01cf8b
@@ -3157,9 +3160,14 @@ void amd64g_dirtyhelper_CPUID_avx_and_cx16 ( VexGuestAMD64State* st )
01cf8b
       case 0x00000000:
01cf8b
          SET_ABCD(0x0000000d, 0x756e6547, 0x6c65746e, 0x49656e69);
01cf8b
          break;
01cf8b
-      case 0x00000001:
01cf8b
-         SET_ABCD(0x000206a7, 0x00100800, 0x1f9ae3bf, 0xbfebfbff);
01cf8b
+      case 0x00000001: {
01cf8b
+         // As a baseline, advertise neither F16C (ecx:29) nor RDRAND (ecx:30),
01cf8b
+         // but patch in support for them as directed by the caller.
01cf8b
+         UInt ecx_extra
01cf8b
+            = (hasF16C ? (1U << 29) : 0) | (hasRDRAND ? (1U << 30) : 0);
01cf8b
+         SET_ABCD(0x000206a7, 0x00100800, (0x1f9ae3bf | ecx_extra), 0xbfebfbff);
01cf8b
          break;
01cf8b
+      }
01cf8b
       case 0x00000002:
01cf8b
          SET_ABCD(0x76035a01, 0x00f0b0ff, 0x00000000, 0x00ca0000);
01cf8b
          break;
01cf8b
diff --git a/VEX/priv/guest_amd64_toIR.c b/VEX/priv/guest_amd64_toIR.c
01cf8b
index 56e992c..96dee38 100644
01cf8b
--- a/VEX/priv/guest_amd64_toIR.c
01cf8b
+++ b/VEX/priv/guest_amd64_toIR.c
01cf8b
@@ -22007,7 +22007,8 @@ Long dis_ESC_0F (
01cf8b
 
01cf8b
       vassert(fName); vassert(fAddr);
01cf8b
       IRExpr** args = NULL;
01cf8b
-      if (fAddr == &amd64g_dirtyhelper_CPUID_avx2) {
01cf8b
+      if (fAddr == &amd64g_dirtyhelper_CPUID_avx2
01cf8b
+          || fAddr == &amd64g_dirtyhelper_CPUID_avx_and_cx16) {
01cf8b
          Bool hasF16C   = (archinfo->hwcaps & VEX_HWCAPS_AMD64_F16C) != 0;
01cf8b
          Bool hasRDRAND = (archinfo->hwcaps & VEX_HWCAPS_AMD64_RDRAND) != 0;
01cf8b
          args = mkIRExprVec_3(IRExpr_GSPTR(),
01cf8b
diff --git a/coregrind/m_machine.c b/coregrind/m_machine.c
01cf8b
index 3536e57..56a28d1 100644
01cf8b
--- a/coregrind/m_machine.c
01cf8b
+++ b/coregrind/m_machine.c
01cf8b
@@ -1076,10 +1076,10 @@ Bool VG_(machine_get_hwcaps)( void )
01cf8b
         have_avx2 = (ebx & (1<<5)) != 0; /* True => have AVX2 */
01cf8b
      }
01cf8b
 
01cf8b
-     /* Sanity check for RDRAND and F16C.  These don't actually *need* AVX2, but
01cf8b
-        it's convenient to restrict them to the AVX2 case since the simulated
01cf8b
-        CPUID we'll offer them on has AVX2 as a base. */
01cf8b
-     if (!have_avx2) {
01cf8b
+     /* Sanity check for RDRAND and F16C.  These don't actually *need* AVX, but
01cf8b
+        it's convenient to restrict them to the AVX case since the simulated
01cf8b
+        CPUID we'll offer them on has AVX as a base. */
01cf8b
+     if (!have_avx) {
01cf8b
         have_f16c   = False;
01cf8b
         have_rdrand = False;
01cf8b
      }