[aarch64-port-dev ] frintz instruction in simulator

[D.Sturm]

Ah does the mailing list swallow the attachments or did you overlook it?
(not usual to not inline patches here I see, sorry!)

Here's the patch:

# HG changeset patch
# User Daniel Sturm <d.sturm42 at gmail.com>
# Date 1398462569 -7200
#      Fri Apr 25 23:49:29 2014 +0200
# Node ID 8866f73c6637b0ff0107eb9f14c1fd319e1f50bd
# Parent  d70a2050eb423046283bc8f1280fd9d9e1fabc21
implemented frintz instruction.

diff -r d70a2050eb42 -r 8866f73c6637 simulator.cpp
--- a/simulator.cpp Fri Jan 03 13:42:36 2014 +0000
+++ b/simulator.cpp Fri Apr 25 23:49:29 2014 +0200
@@ -1373,10 +1373,19 @@
   case 0b000111: // FCVT double/single to half precision
     errorCode = ERROR_NYI;
     return error();
+  case 0b001011:                // FRINTZ
+    if (type == 0b01) {
+      frintzd();
+    } else if (type == 0b00) {
+      frintzs();
+    } else {
+      errorCode = ERROR_UNALLOC;
+      return error();
+    }
+    return STATUS_READY;
   case 0b001000: // FRINTN etc
   case 0b001001:
   case 0b001010:
-  case 0b001011:
   case 0b001100:
   case 0b001110:
   case 0b001111:
@@ -7988,6 +7997,32 @@
   dreg(0) = (double)sreg(5);
 }

+template <class T>
+static inline T frintz(T val)
+{
+  if (val == -0.0 || isinf(val) || isnan(val)) {
+    return val;
+  }
+  return (T)(int64_t)val;
+}
+
+// round to integral towards zero
+void AArch64Simulator::frintzs()
+{
+  // instr[9,5] = Sn
+  // instr[4,0] = Sd
+  sreg(0) = frintz(sreg(5));
+}
+
+// round to integral towards zero
+void AArch64Simulator::frintzd()
+{
+  // instr[9,5] = Dn
+  // instr[4,0] = Dd
+  dreg(0) = frintz(dreg(5));
+}
+
+
 // 2 sources

 // float add
diff -r d70a2050eb42 -r 8866f73c6637 simulator.hpp
--- a/simulator.hpp Fri Jan 03 13:42:36 2014 +0000
+++ b/simulator.hpp Fri Apr 25 23:49:29 2014 +0200
@@ -1287,6 +1287,10 @@
   // etc

   // TODO FP round to integral/ nearest integral floating
+  // float round to integral, towards zero.
+  void frintzs();
+  // double round to integral, towards zero.
+  void frintzd();

   // TODO FP arithmetic
   //



On 26 April 2014 15:47, Andrew Haley <aph at redhat.com> wrote:

> On 04/25/2014 11:16 PM, D.Sturm wrote:
> > Hi,
> > Since I need a frintz instruction when generating the floating point
> > remainder operation* I had to implement it in the simulator.
> >
> > The implementation should be correct except in case of a signaling NaN
> > since that should throw an exception (could make a difference if someone
> > generates a signaling NaN manually from the integer pattern, but that
> seems
> > like an extreme edge case).
> >
> > Maybe useful for the general branch too.
>
> Sure, er, please post it.  :-)
>
> Andrew.
>
>
>