Upstreaming avx512 register support in debugserver. These changes

were originally written by Chris Bieneman, they've undergone a
number of changes since then.

Also including the debugserver bridgeos support, another arm
environment that runs Darwin akin to ios.  These codepaths are
activated when running in a bridgeos environment which we're not
set up to test today.

There's additional (small) lldb changes to handle bridgeos binaries
that still need to be merged up.

Tested on a darwin system with avx512 hardware and without.

<rdar://problem/36424951> 

llvm-svn: 326756

4 files changed, 241 insertions, 19 deletions

diff --git a/lldb/packages/Python/lldbsuite/test/decorators.py b/lldb/packages/Python/lldbsuite/test/decorators.py
index f67681d7a0f..38379379504 100644
--- a/lldb/packages/Python/lldbsuite/test/decorators.py
+++ b/lldb/packages/Python/lldbsuite/test/decorators.py
@@ -10,6 +10,7 @@ import platform
 import re
 import sys
 import tempfile
+import subprocess
 
 # Third-party modules
 import six
@@ -741,3 +742,20 @@ def skipIfXmlSupportMissing(func):
     fail_value = True # More likely to notice if something goes wrong
     have_xml = xml.GetValueForKey("value").GetBooleanValue(fail_value)
     return unittest2.skipIf(not have_xml, "requires xml support")(func)
+
+# Call sysctl on darwin to see if a specified hardware feature is available on this machine.
+def skipUnlessFeature(feature):
+    def is_feature_enabled(self):
+        if platform.system() == 'Darwin':
+            try:
+                DEVNULL = open(os.devnull, 'w')
+                output = subprocess.check_output(["/usr/sbin/sysctl", feature], stderr=DEVNULL)
+                # If 'feature: 1' was output, then this feature is available and
+                # the test should not be skipped.
+                if re.match('%s: 1\s*' % feature, output):
+                    return None
+                else:
+                    return "%s is not supported on this system." % feature
+            except subprocess.CalledProcessError:
+                return "%s is not supported on this system." % feature
+    return skipTestIfFn(is_feature_enabled)
diff --git a/lldb/packages/Python/lldbsuite/test/functionalities/register/intel_avx/TestYMMRegister.py b/lldb/packages/Python/lldbsuite/test/functionalities/register/intel_avx/TestYMMRegister.py
index d4249694c9b..0f497536e41 100644
--- a/lldb/packages/Python/lldbsuite/test/functionalities/register/intel_avx/TestYMMRegister.py
+++ b/lldb/packages/Python/lldbsuite/test/functionalities/register/intel_avx/TestYMMRegister.py
@@ -21,8 +21,9 @@ class TestYMMRegister(TestBase):
     @skipIfiOSSimulator
     @skipIfTargetAndroid()
     @skipIf(archs=no_match(['i386', 'x86_64']))
+    @expectedFailureAll(oslist=["linux"], bugnumber="rdar://30523153")
     def test(self):
-        self.build()
+        self.build(dictionary={"CFLAGS_EXTRAS": "-march=haswell"})
         self.setTearDownCleanup()
 
         exe = self.getBuildArtifact("a.out")
@@ -56,9 +57,10 @@ class TestYMMRegister(TestBase):
         else:
             register_range = 8
         for i in range(register_range):
+            j = i - ((i / 8) * 8)
             self.runCmd("thread step-inst")
 
-            register_byte = (byte_pattern1 | i)
+            register_byte = (byte_pattern1 | j)
             pattern = "ymm" + str(i) + " = " + str('{') + (
                 str(hex(register_byte)) + ' ') * 31 + str(hex(register_byte)) + str('}')
 
@@ -66,7 +68,7 @@ class TestYMMRegister(TestBase):
                 "register read ymm" + str(i),
                 substrs=[pattern])
 
-            register_byte = (byte_pattern2 | i)
+            register_byte = (byte_pattern2 | j)
             pattern = "ymm" + str(i) + " = " + str('{') + (
                 str(hex(register_byte)) + ' ') * 31 + str(hex(register_byte)) + str('}')
 
diff --git a/lldb/packages/Python/lldbsuite/test/functionalities/register/intel_avx/TestZMMRegister.py b/lldb/packages/Python/lldbsuite/test/functionalities/register/intel_avx/TestZMMRegister.py
new file mode 100644
index 00000000000..823e40155ff
--- /dev/null
+++ b/lldb/packages/Python/lldbsuite/test/functionalities/register/intel_avx/TestZMMRegister.py
@@ -0,0 +1,126 @@
+"""
+Test that we correctly read the YMM registers.
+"""
+
+from __future__ import print_function
+
+
+import os
+import time
+import lldb
+from lldbsuite.test.decorators import *
+from lldbsuite.test.lldbtest import *
+from lldbsuite.test import lldbutil
+
+
+class TestYMMRegister(TestBase):
+    mydir = TestBase.compute_mydir(__file__)
+    NO_DEBUG_INFO_TESTCASE = True
+
+    @skipUnlessDarwin
+    @skipIfiOSSimulator
+    @skipIf(archs=no_match(['i386', 'x86_64']))
+    @debugserver_test
+    @skipUnlessFeature('hw.optional.avx512f')
+    def test(self):
+        self.build(dictionary={"CFLAGS_EXTRAS": "-mavx512f"})
+        self.setTearDownCleanup()
+
+        exe = self.getBuildArtifact("a.out")
+        target = self.dbg.CreateTarget(exe)
+
+        self.assertTrue(target, VALID_TARGET)
+
+        byte_pattern1 = 0x80
+        byte_pattern2 = 0xFF
+
+        # This test is working with assembly instructions; it'll make
+        # it easier to debug the console output if the assembly is
+        # displayed.
+        self.runCmd("settings set stop-disassembly-display always")
+
+        # Launch the process and stop.
+        self.expect("run", PROCESS_STOPPED, substrs=['stopped'])
+
+        # Check stop reason; Should be either signal SIGTRAP or EXC_BREAKPOINT
+        output = self.res.GetOutput()
+        matched = False
+        substrs = [
+            'stop reason = EXC_BREAKPOINT',
+            'stop reason = signal SIGTRAP']
+        for str1 in substrs:
+            matched = output.find(str1) != -1
+            with recording(self, False) as sbuf:
+                print("%s sub string: %s" % ('Expecting', str1), file=sbuf)
+                print("Matched" if matched else "Not Matched", file=sbuf)
+            if matched:
+                break
+        self.assertTrue(matched, STOPPED_DUE_TO_SIGNAL)
+
+        if self.getArchitecture() == 'x86_64':
+            register_range = 16
+        else:
+            register_range = 8
+        for i in range(register_range):
+            j = i - ((i / 8) * 8)
+            self.runCmd("thread step-inst")
+
+            register_byte = (byte_pattern1 | j)
+            pattern = "ymm" + str(i) + " = " + str('{') + (
+                str(hex(register_byte)) + ' ') * 31 + str(hex(register_byte)) + str('}')
+
+            self.expect(
+                "register read ymm" + str(i),
+                substrs=[pattern])
+
+            register_byte = (byte_pattern2 | j)
+            pattern = "ymm" + str(i) + " = " + str('{') + (
+                str(hex(register_byte)) + ' ') * 31 + str(hex(register_byte)) + str('}')
+
+            self.runCmd("thread step-inst")
+            self.expect(
+                "register read ymm" + str(i),
+                substrs=[pattern])
+        
+        self.expect("continue", PROCESS_STOPPED, substrs=['stopped'])
+
+        # Check stop reason; Should be either signal SIGTRAP or EXC_BREAKPOINT
+        output = self.res.GetOutput()
+        matched = False
+        substrs = [
+            'stop reason = EXC_BREAKPOINT',
+            'stop reason = signal SIGTRAP']
+        for str1 in substrs:
+            matched = output.find(str1) != -1
+            with recording(self, False) as sbuf:
+                print("%s sub string: %s" % ('Expecting', str1), file=sbuf)
+                print("Matched" if matched else "Not Matched", file=sbuf)
+            if matched:
+                break
+        self.assertTrue(matched, STOPPED_DUE_TO_SIGNAL)
+
+        if self.getArchitecture() == 'x86_64':
+            register_range = 32
+        else:
+            register_range = 8
+        for i in range(register_range):
+            j = i - ((i / 8) * 8)
+            self.runCmd("thread step-inst")
+            self.runCmd("thread step-inst")
+
+            register_byte = (byte_pattern2 | j)
+            pattern = "zmm" + str(i) + " = " + str('{') + (
+                str(hex(register_byte)) + ' ') * 63 + str(hex(register_byte)) + str('}')
+
+            self.expect(
+                "register read zmm" + str(i),
+                substrs=[pattern])
+
+            register_byte = (byte_pattern2 | j)
+            pattern = "zmm" + str(i) + " = " + str('{') + (
+                str(hex(register_byte)) + ' ') * 63 + str(hex(register_byte)) + str('}')
+
+            self.runCmd("thread step-inst")
+            self.expect(
+                "register read zmm" + str(i),
+                substrs=[pattern])
diff --git a/lldb/packages/Python/lldbsuite/test/functionalities/register/intel_avx/main.c b/lldb/packages/Python/lldbsuite/test/functionalities/register/intel_avx/main.c
index 4cfb2239c06..671331fe450 100644
--- a/lldb/packages/Python/lldbsuite/test/functionalities/register/intel_avx/main.c
+++ b/lldb/packages/Python/lldbsuite/test/functionalities/register/intel_avx/main.c
@@ -9,15 +9,14 @@
 
 void func() {
   unsigned int ymmvalues[16];
-  unsigned char val;
-  unsigned char i;
-  for (i = 0 ; i < 16 ; i++)
+  for (int i = 0 ; i < 16 ; i++)
   {
-    val = (0x80 | i);
+    unsigned char val = (0x80 | i);
     ymmvalues[i] = (val << 24) | (val << 16) | (val << 8) | val;
   }
 
   unsigned int ymmallones = 0xFFFFFFFF;
+#if defined(__AVX__)
   __asm__("int3;"
           "vbroadcastss %1, %%ymm0;"
           "vbroadcastss %0, %%ymm0;"
@@ -36,32 +35,109 @@ void func() {
           "vbroadcastss %8, %%ymm7;"
           "vbroadcastss %0, %%ymm7;"
 #if defined(__x86_64__)
-          "vbroadcastss %9, %%ymm8;"
+          "vbroadcastss %1, %%ymm8;"
           "vbroadcastss %0, %%ymm8;"
-          "vbroadcastss %10, %%ymm9;"
+          "vbroadcastss %2, %%ymm9;"
           "vbroadcastss %0, %%ymm9;"
-          "vbroadcastss %11, %%ymm10;"
+          "vbroadcastss %3, %%ymm10;"
           "vbroadcastss %0, %%ymm10;"
-          "vbroadcastss %12, %%ymm11;"
+          "vbroadcastss %4, %%ymm11;"
           "vbroadcastss %0, %%ymm11;"
-          "vbroadcastss %13, %%ymm12;"
+          "vbroadcastss %5, %%ymm12;"
           "vbroadcastss %0, %%ymm12;"
-          "vbroadcastss %14, %%ymm13;"
+          "vbroadcastss %6, %%ymm13;"
           "vbroadcastss %0, %%ymm13;"
-          "vbroadcastss %15, %%ymm14;"
+          "vbroadcastss %7, %%ymm14;"
           "vbroadcastss %0, %%ymm14;"
-          "vbroadcastss %16, %%ymm15;"
+          "vbroadcastss %8, %%ymm15;"
           "vbroadcastss %0, %%ymm15;"
 #endif
           ::"m"(ymmallones),
           "m"(ymmvalues[0]), "m"(ymmvalues[1]), "m"(ymmvalues[2]), "m"(ymmvalues[3]),
           "m"(ymmvalues[4]), "m"(ymmvalues[5]), "m"(ymmvalues[6]), "m"(ymmvalues[7])
+              );
+#endif
+
+#if defined(__AVX512F__)
+  unsigned int zmmvalues[32];
+  for (int i = 0 ; i < 32 ; i++)
+  {
+    unsigned char val = (0x80 | i);
+    zmmvalues[i] = (val << 24) | (val << 16) | (val << 8) | val;
+  }
+
+  __asm__("int3;"
+          "vbroadcastss %1, %%zmm0;"
+          "vbroadcastss %0, %%zmm0;"
+          "vbroadcastss %2, %%zmm1;"
+          "vbroadcastss %0, %%zmm1;"
+          "vbroadcastss %3, %%zmm2;"
+          "vbroadcastss %0, %%zmm2;"
+          "vbroadcastss %4, %%zmm3;"
+          "vbroadcastss %0, %%zmm3;"
+          "vbroadcastss %5, %%zmm4;"
+          "vbroadcastss %0, %%zmm4;"
+          "vbroadcastss %6, %%zmm5;"
+          "vbroadcastss %0, %%zmm5;"
+          "vbroadcastss %7, %%zmm6;"
+          "vbroadcastss %0, %%zmm6;"
+          "vbroadcastss %8, %%zmm7;"
+          "vbroadcastss %0, %%zmm7;"
 #if defined(__x86_64__)
-          ,
-          "m"(ymmvalues[8]), "m"(ymmvalues[9]), "m"(ymmvalues[10]), "m"(ymmvalues[11]),
-          "m"(ymmvalues[12]), "m"(ymmvalues[13]), "m"(ymmvalues[14]), "m"(ymmvalues[15])
+          "vbroadcastss %1, %%zmm8;"
+          "vbroadcastss %0, %%zmm8;"
+          "vbroadcastss %2, %%zmm9;"
+          "vbroadcastss %0, %%zmm9;"
+          "vbroadcastss %3, %%zmm10;"
+          "vbroadcastss %0, %%zmm10;"
+          "vbroadcastss %4, %%zmm11;"
+          "vbroadcastss %0, %%zmm11;"
+          "vbroadcastss %5, %%zmm12;"
+          "vbroadcastss %0, %%zmm12;"
+          "vbroadcastss %6, %%zmm13;"
+          "vbroadcastss %0, %%zmm13;"
+          "vbroadcastss %7, %%zmm14;"
+          "vbroadcastss %0, %%zmm14;"
+          "vbroadcastss %8, %%zmm15;"
+          "vbroadcastss %0, %%zmm15;"
+          "vbroadcastss %1, %%zmm16;"
+          "vbroadcastss %0, %%zmm16;"
+          "vbroadcastss %2, %%zmm17;"
+          "vbroadcastss %0, %%zmm17;"
+          "vbroadcastss %3, %%zmm18;"
+          "vbroadcastss %0, %%zmm18;"
+          "vbroadcastss %4, %%zmm19;"
+          "vbroadcastss %0, %%zmm19;"
+          "vbroadcastss %5, %%zmm20;"
+          "vbroadcastss %0, %%zmm20;"
+          "vbroadcastss %6, %%zmm21;"
+          "vbroadcastss %0, %%zmm21;"
+          "vbroadcastss %7, %%zmm22;"
+          "vbroadcastss %0, %%zmm22;"
+          "vbroadcastss %8, %%zmm23;"
+          "vbroadcastss %0, %%zmm23;"
+          "vbroadcastss %1, %%zmm24;"
+          "vbroadcastss %0, %%zmm24;"
+          "vbroadcastss %2, %%zmm25;"
+          "vbroadcastss %0, %%zmm25;"
+          "vbroadcastss %3, %%zmm26;"
+          "vbroadcastss %0, %%zmm26;"
+          "vbroadcastss %4, %%zmm27;"
+          "vbroadcastss %0, %%zmm27;"
+          "vbroadcastss %5, %%zmm28;"
+          "vbroadcastss %0, %%zmm28;"
+          "vbroadcastss %6, %%zmm29;"
+          "vbroadcastss %0, %%zmm29;"
+          "vbroadcastss %7, %%zmm30;"
+          "vbroadcastss %0, %%zmm30;"
+          "vbroadcastss %8, %%zmm31;"
+          "vbroadcastss %0, %%zmm31;"
+#endif
+          ::"m"(ymmallones),
+          "m"(zmmvalues[0]), "m"(zmmvalues[1]), "m"(zmmvalues[2]), "m"(zmmvalues[3]),
+          "m"(zmmvalues[4]), "m"(zmmvalues[5]), "m"(zmmvalues[6]), "m"(zmmvalues[7])
+  );
 #endif
-              );
 }
 
 int main(int argc, char const *argv[]) { func(); }