[pypy-commit] pypy default: Only execute NEON instructions on CPUs supporting NEON

[stefanor]

Author: Stefano Rivera <stefano at rivera.za.net>
Branch: 
Changeset: r83380:fc95c9347679
Date: 2016-03-25 21:24 -0500
http://bitbucket.org/pypy/pypy/changeset/fc95c9347679/

Log:	Only execute NEON instructions on CPUs supporting NEON

	On Debian armhf, NEON is not mandated. In fact Debian's buildds
	don't support it.

diff --git a/rpython/jit/backend/arm/detect.py b/rpython/jit/backend/arm/detect.py
--- a/rpython/jit/backend/arm/detect.py
+++ b/rpython/jit/backend/arm/detect.py
@@ -1,6 +1,7 @@
 import os
 
 from rpython.translator.tool.cbuild import ExternalCompilationInfo
+from rpython.rtyper.lltypesystem import lltype, rffi
 from rpython.rtyper.tool import rffi_platform
 from rpython.rlib.clibffi import FFI_DEFAULT_ABI, FFI_SYSV, FFI_VFP
 from rpython.translator.platform import CompilationError
@@ -15,6 +16,7 @@
     asm volatile("VMOV s0, s1");
 }
     """])
+getauxval = rffi.llexternal("getauxval", [lltype.Unsigned], lltype.Unsigned)
 
 def detect_hardfloat():
     return FFI_DEFAULT_ABI == FFI_VFP
@@ -63,3 +65,10 @@
                     "falling back to", "ARMv%d" % n)
     debug_stop("jit-backend-arch")
     return n
+
+
+def detect_neon():
+    AT_HWCAP = 16
+    HWCAP_NEON = 1 << 12
+    hwcap = getauxval(AT_HWCAP)
+    return bool(hwcap & HWCAP_NEON)
diff --git a/rpython/jit/backend/arm/opassembler.py b/rpython/jit/backend/arm/opassembler.py
--- a/rpython/jit/backend/arm/opassembler.py
+++ b/rpython/jit/backend/arm/opassembler.py
@@ -1092,8 +1092,8 @@
         self.mc.VCVT_int_to_float(res.value, r.svfp_ip.value)
         return fcond
 
-    # the following five instructions are only ARMv7;
-    # regalloc.py won't call them at all on ARMv6
+    # the following five instructions are only ARMv7 with NEON;
+    # regalloc.py won't call them at all, in other cases
     emit_opx_llong_add = gen_emit_float_op('llong_add', 'VADD_i64')
     emit_opx_llong_sub = gen_emit_float_op('llong_sub', 'VSUB_i64')
     emit_opx_llong_and = gen_emit_float_op('llong_and', 'VAND_i64')
diff --git a/rpython/jit/backend/arm/regalloc.py b/rpython/jit/backend/arm/regalloc.py
--- a/rpython/jit/backend/arm/regalloc.py
+++ b/rpython/jit/backend/arm/regalloc.py
@@ -530,7 +530,7 @@
                             EffectInfo.OS_LLONG_AND,
                             EffectInfo.OS_LLONG_OR,
                             EffectInfo.OS_LLONG_XOR):
-                if self.cpu.cpuinfo.arch_version >= 7:
+                if self.cpu.cpuinfo.neon:
                     args = self._prepare_llong_binop_xx(op, fcond)
                     self.perform_extra(op, args, fcond)
                     return
diff --git a/rpython/jit/backend/arm/runner.py b/rpython/jit/backend/arm/runner.py
--- a/rpython/jit/backend/arm/runner.py
+++ b/rpython/jit/backend/arm/runner.py
@@ -7,13 +7,14 @@
 from rpython.rlib.jit_hooks import LOOP_RUN_CONTAINER
 from rpython.rtyper.lltypesystem import lltype, llmemory
 from rpython.jit.backend.arm.detect import detect_hardfloat
-from rpython.jit.backend.arm.detect import detect_arch_version
+from rpython.jit.backend.arm.detect import detect_arch_version, detect_neon
 
 jitframe.STATICSIZE = JITFRAME_FIXED_SIZE
 
 class CPUInfo(object):
     hf_abi = False
     arch_version = 6
+    neon = False
 
 class AbstractARMCPU(AbstractLLCPU):
 
@@ -48,6 +49,7 @@
     def setup_once(self):
         self.cpuinfo.arch_version = detect_arch_version()
         self.cpuinfo.hf_abi = detect_hardfloat()
+        self.cpuinfo.neon = detect_neon()
         #self.codemap.setup()
         self.assembler.setup_once()