[risu PATCH 0/2] Add support for instructions with length > 32 bit

[risu PATCH 0/2] Add support for instructions with length > 32 bit

[Thomas Huth]

RISU currently only supports instructions with a length of
16 bit or 32 bit, however classical CISC systems like s390x
also have instructions that are longer than 32 bit. The first
patch improves the generator to support instructions with
more than 32 bits, and the second patch adds some 48-bit
instructions to the s390x.risu file.

Thomas Huth (2):
  risugen: allow instructions with length > 32 bit
  s390x.risu: Add some instructions with a length of 48 bits

 risugen                | 48 +++++++++++++++++++++++++++++-------------
 risugen_arm.pm         |  6 +++---
 risugen_common.pm      |  2 +-
 risugen_loongarch64.pm |  4 ++--
 risugen_m68k.pm        |  6 +++---
 risugen_ppc64.pm       |  4 ++--
 risugen_s390x.pm       | 17 +++++++++++----
 s390x.risu             | 20 ++++++++++++++++++
 8 files changed, 77 insertions(+), 30 deletions(-)

-- 
2.41.0

[risu PATCH 1/2] risugen: allow instructions with length > 32 bit

[Thomas Huth]

RISU currently only supports instructions with a length of
16 bit or 32 bit, however classical CISC systems like s390x
also have instructions that are longer than 32 bit. Thus let's
change the generator to support longer instructions, too.

This adds support for 48-bit instructions on s390x, while
the other architectures are just minimally changed to preserve
the current state.

Signed-off-by: Thomas Huth <thuth@redhat.com>
---
 risugen                | 48 +++++++++++++++++++++++++++++-------------
 risugen_arm.pm         |  6 +++---
 risugen_common.pm      |  2 +-
 risugen_loongarch64.pm |  4 ++--
 risugen_m68k.pm        |  6 +++---
 risugen_ppc64.pm       |  4 ++--
 risugen_s390x.pm       | 17 +++++++++++----
 7 files changed, 57 insertions(+), 30 deletions(-)

diff --git a/risugen b/risugen
index fa94a39..833b459 100755
--- a/risugen
+++ b/risugen
@@ -105,6 +105,16 @@ sub read_tokenised_line(*)
     return @tokens;
 }
 
+sub check_bitmask($$)
+{
+    my ($fixedbits, $fixedbitmask) = @_;
+
+    if ((($fixedbits & $fixedbitmask) != $fixedbits)
+        || (($fixedbits & ~$fixedbitmask) != 0)) {
+        die "internal error: fixed bits not lined up with mask";
+    }
+}
+
 sub parse_config_file($)
 {
     # Read in the config file defining the instructions we can generate
@@ -160,10 +170,11 @@ sub parse_config_file($)
             exit(1);
         }
 
-        my $fixedbits = 0;
-        my $fixedbitmask = 0;
+        my @fixedbits = (0, 0, 0, 0);
+        my @fixedbitmask = (0, 0, 0, 0);
         my $bitpos = 32;
-        my $insnwidth = 32;
+        my $wordpos = 0;
+        my $insnwidth = 0;
         my $seenblock = 0;
 
         while (@bits) {
@@ -217,36 +228,43 @@ sub parse_config_file($)
 
             my $bitmask = oct("0b". '1' x $bitlen);
             $bitpos -= $bitlen;
+            $insnwidth += $bitlen;
             if ($bitpos < 0) {
                 print STDERR "$file:$.: ($insn $enc) too many bits specified\n";
                 exit(1);
             }
 
             if (defined $bitval) {
-                $fixedbits |= ($bitval << $bitpos);
-                $fixedbitmask |= ($bitmask << $bitpos);
+                $fixedbits[$wordpos] |= ($bitval << $bitpos);
+                $fixedbitmask[$wordpos] |= ($bitmask << $bitpos);
             } else {
-                push @fields, [ $var, $bitpos, $bitmask ];
+                push @fields, [ $var, $bitpos, $bitmask, $wordpos ];
+            }
+
+            if ($bitpos == 0) {
+                check_bitmask($fixedbits[$wordpos], $fixedbitmask[$wordpos]);
+
+                $wordpos += 1;
+                if (@bits) {
+                    $bitpos = 32;
+                }
             }
         }
         if ($bitpos == 16) {
             # assume this is a half-width thumb instruction
             # Note that we don't fiddle with the bitmasks or positions,
             # which means the generated insn will be in the high halfword!
-            $insnwidth = 16;
-        } elsif ($bitpos != 0) {
-            print STDERR "$file:$.: ($insn $enc) not enough bits specified\n";
+            check_bitmask($fixedbits[$wordpos], $fixedbitmask[$wordpos]);
+        } elsif ($bitpos != 0 && $bitpos != 32) {
+            print STDERR "$file:$.: ($insn $enc) not enough bits specified ($bitpos)\n";
             exit(1);
         }
-        if ((($fixedbits & $fixedbitmask) != $fixedbits)
-            || (($fixedbits & ~$fixedbitmask) != 0)) {
-            die "internal error: fixed bits not lined up with mask";
-        }
         #  Stick the fixedbit info on the front of the array now we know it
         $insnrec->{name} = $insnname;
         $insnrec->{width} = $insnwidth;
-        $insnrec->{fixedbits} = $fixedbits;
-        $insnrec->{fixedbitmask} = $fixedbitmask;
+        $insnrec->{fixedbits} = [ @fixedbits ];
+        $insnrec->{fixedbitmask} = [ @fixedbitmask ];
+        $insnrec->{words} = $wordpos;
         $insnrec->{fields} = [ @fields ];
         if (@insn_groups) {
             $insnrec->{groups} = [ @insn_groups ];
diff --git a/risugen_arm.pm b/risugen_arm.pm
index 8d423b1..9dd6139 100644
--- a/risugen_arm.pm
+++ b/risugen_arm.pm
@@ -964,15 +964,15 @@ sub gen_one_insn($$)
         my $insn = int(rand(0xffffffff));
         my $insnname = $rec->{name};
         my $insnwidth = $rec->{width};
-        my $fixedbits = $rec->{fixedbits};
-        my $fixedbitmask = $rec->{fixedbitmask};
+        my $fixedbits = (@{ $rec->{fixedbits} })[0];
+        my $fixedbitmask = (@{  $rec->{fixedbitmask} })[0];
         my $constraint = $rec->{blocks}{"constraints"};
         my $memblock = $rec->{blocks}{"memory"};
 
         $insn &= ~$fixedbitmask;
         $insn |= $fixedbits;
         for my $tuple (@{ $rec->{fields} }) {
-            my ($var, $pos, $mask) = @$tuple;
+            my ($var, $pos, $mask, $wordpos) = @$tuple;
             my $val = ($insn >> $pos) & $mask;
             # XXX (claudio) ARM-specific - maybe move to arm.risu?
             # Check constraints here:
diff --git a/risugen_common.pm b/risugen_common.pm
index 71ee996..c816895 100644
--- a/risugen_common.pm
+++ b/risugen_common.pm
@@ -116,7 +116,7 @@ sub eval_with_fields($$$$$) {
     my $calling_package = caller;
     my $evalstr = "{ package $calling_package; ";
     for my $tuple (@{ $rec->{fields} }) {
-        my ($var, $pos, $mask) = @$tuple;
+        my ($var, $pos, $mask, $wordpos) = @$tuple;
         my $val = ($insn >> $pos) & $mask;
         $evalstr .= "my (\$$var) = $val; ";
     }
diff --git a/risugen_loongarch64.pm b/risugen_loongarch64.pm
index 5394fdc..8c74b6e 100644
--- a/risugen_loongarch64.pm
+++ b/risugen_loongarch64.pm
@@ -390,8 +390,8 @@ sub gen_one_insn($$)
         my $insn = int(rand(0xffffffff));
         my $insnname = $rec->{name};
         my $insnwidth = $rec->{width};
-        my $fixedbits = $rec->{fixedbits};
-        my $fixedbitmask = $rec->{fixedbitmask};
+        my $fixedbits = (@{ $rec->{fixedbits} })[0];
+        my $fixedbitmask = (@{  $rec->{fixedbitmask} })[0];
         my $constraint = $rec->{blocks}{"constraints"};
         my $memblock = $rec->{blocks}{"memory"};
         my $safefloat = $rec->{blocks}{"safefloat"};
diff --git a/risugen_m68k.pm b/risugen_m68k.pm
index 7d62b13..49f3100 100644
--- a/risugen_m68k.pm
+++ b/risugen_m68k.pm
@@ -111,8 +111,8 @@ sub gen_one_insn($$)
         my $insn = int(rand(0xffffffff));
         my $insnname = $rec->{name};
         my $insnwidth = $rec->{width};
-        my $fixedbits = $rec->{fixedbits};
-        my $fixedbitmask = $rec->{fixedbitmask};
+        my $fixedbits = (@{ $rec->{fixedbits} })[0];
+        my $fixedbitmask = (@{  $rec->{fixedbitmask} })[0];
         my $constraint = $rec->{blocks}{"constraints"};
         my $memblock = $rec->{blocks}{"memory"};
 
@@ -120,7 +120,7 @@ sub gen_one_insn($$)
         $insn |= $fixedbits;
 
         for my $tuple (@{ $rec->{fields} }) {
-            my ($var, $pos, $mask) = @$tuple;
+            my ($var, $pos, $mask, $wordpos) = @$tuple;
             my $val = ($insn >> $pos) & $mask;
             # Check constraints here:
             # not allowed to use or modify sp (A7) or fp (A6)
diff --git a/risugen_ppc64.pm b/risugen_ppc64.pm
index b241172..d1d3fc9 100644
--- a/risugen_ppc64.pm
+++ b/risugen_ppc64.pm
@@ -300,8 +300,8 @@ sub gen_one_insn($$)
         my $insn = int(rand(0xffffffff));
         my $insnname = $rec->{name};
         my $insnwidth = $rec->{width};
-        my $fixedbits = $rec->{fixedbits};
-        my $fixedbitmask = $rec->{fixedbitmask};
+        my $fixedbits = (@{ $rec->{fixedbits} })[0];
+        my $fixedbitmask = (@{  $rec->{fixedbitmask} })[0];
         my $constraint = $rec->{blocks}{"constraints"};
         my $memblock = $rec->{blocks}{"memory"};
 
diff --git a/risugen_s390x.pm b/risugen_s390x.pm
index 260e2dd..04ea6fb 100644
--- a/risugen_s390x.pm
+++ b/risugen_s390x.pm
@@ -84,15 +84,21 @@ sub gen_one_insn($$)
     INSN: while(1) {
         my ($forcecond, $rec) = @_;
         my $insn = int(rand(0xffffffff));
+        my $insn48 = int(rand(0xffff0000));
         my $insnname = $rec->{name};
         my $insnwidth = $rec->{width};
-        my $fixedbits = $rec->{fixedbits};
-        my $fixedbitmask = $rec->{fixedbitmask};
+        my @fixedbits = (@{ $rec->{fixedbits} });
+        my @fixedbitmask = (@{  $rec->{fixedbitmask} });
         my $constraint = $rec->{blocks}{"constraints"};
         my $memblock = $rec->{blocks}{"memory"};
 
-        $insn &= ~$fixedbitmask;
-        $insn |= $fixedbits;
+        $insn &= ~$fixedbitmask[0];
+        $insn |= $fixedbits[0];
+
+        if ($insnwidth == 48) {
+            $insn48 &= ~$fixedbitmask[1];
+            $insn48 |= $fixedbits[1];
+        }
 
         if (defined $constraint) {
             # user-specified constraint: evaluate in an environment
@@ -121,6 +127,9 @@ sub gen_one_insn($$)
             insn16(($insn >> 16) & 0xffff);
         } else {
             insn32($insn);
+            if ($insnwidth == 48) {
+                insn16(($insn48 >> 16) & 0xffff);
+            }
         }
 
         return;
-- 
2.41.0

[risu PATCH 2/2] s390x.risu: Add some instructions with a length of 48 bits

[Thomas Huth]

Now that risugen can handle instructions with a length > 32 bit,
we can test some instructions with the length of 48-bit, too.

Signed-off-by: Thomas Huth <thuth@redhat.com>
---
 s390x.risu | 20 ++++++++++++++++++++
 1 file changed, 20 insertions(+)

diff --git a/s390x.risu b/s390x.risu
index 1661be6..b70678a 100644
--- a/s390x.risu
+++ b/s390x.risu
@@ -27,6 +27,19 @@ ARK STFLE45 10111001 11111000 r3:4 0000 r1:4 r2:4
 AGRK STFLE45 10111001 11101000 r3:4 0000 r1:4 r2:4
 
 
+# format:RIL-a Add immediate (32 bit)
+AFI EI 11000010 r1:4 1001 i2a:16 i2b:16
+
+# format:RIL-a Add immediate (64 bit)
+AGFI EI 11000010 r1:4 1000 i2a:16 i2b:16
+
+# format:RIE-d Add Immediate (32 bit, distinct operand)
+AHIK STFLE35 11101100 r1:4 r3:4 i2:16 00000000 11011000
+
+# format:RIE-d Add Immediate (64 bit, distinct operand)
+AGHIK STFLE35 11101100 r1:4 r3:4 i2:16 00000000 11011001
+
+
 # format:RRE Add Halfword Immediate (32 bit)
 AHI Z 10100111 r1:4 1010 i2:16
 
@@ -44,6 +57,13 @@ ALGR Z 10111001 00001010 00000000 r1:4 r2:4
 ALGFR Z 10111001 00011010 00000000 r1:4 r2:4
 
 
+# format:RIL-a Insert Immediate (32 bit to upper word)
+IIHF EI 11000000 r1:4 1000 i2a:16 i2b:16
+
+# format:RIL-a Insert Immediate (32 bit to upper word)
+IILF EI 11000000 r1:4 1001 i2a:16 i2b:16
+
+
 # format:RRF-c Population Count
 POPCNT STFLE45 10111001 11100001 m3:4 0000 r1:4 r2:4
 
-- 
2.41.0

Re: [risu PATCH 1/2] risugen: allow instructions with length > 32 bit

[Peter Maydell]

On Fri, 27 Oct 2023 at 11:05, Thomas Huth <thuth@redhat.com> wrote:
>
> RISU currently only supports instructions with a length of
> 16 bit or 32 bit, however classical CISC systems like s390x
> also have instructions that are longer than 32 bit. Thus let's
> change the generator to support longer instructions, too.
>
> This adds support for 48-bit instructions on s390x, while
> the other architectures are just minimally changed to preserve
> the current state.
>
> Signed-off-by: Thomas Huth <thuth@redhat.com>
> ---
>  risugen                | 48 +++++++++++++++++++++++++++++-------------
>  risugen_arm.pm         |  6 +++---
>  risugen_common.pm      |  2 +-
>  risugen_loongarch64.pm |  4 ++--
>  risugen_m68k.pm        |  6 +++---
>  risugen_ppc64.pm       |  4 ++--
>  risugen_s390x.pm       | 17 +++++++++++----
>  7 files changed, 57 insertions(+), 30 deletions(-)
>
> diff --git a/risugen b/risugen
> index fa94a39..833b459 100755
> --- a/risugen
> +++ b/risugen
> @@ -105,6 +105,16 @@ sub read_tokenised_line(*)
>      return @tokens;
>  }
>
> +sub check_bitmask($$)
> +{
> +    my ($fixedbits, $fixedbitmask) = @_;
> +
> +    if ((($fixedbits & $fixedbitmask) != $fixedbits)
> +        || (($fixedbits & ~$fixedbitmask) != 0)) {
> +        die "internal error: fixed bits not lined up with mask";
> +    }
> +}
> +
>  sub parse_config_file($)
>  {
>      # Read in the config file defining the instructions we can generate
> @@ -160,10 +170,11 @@ sub parse_config_file($)
>              exit(1);
>          }
>
> -        my $fixedbits = 0;
> -        my $fixedbitmask = 0;
> +        my @fixedbits = (0, 0, 0, 0);
> +        my @fixedbitmask = (0, 0, 0, 0);

I wonder if rather than turning these into arrays of integers, we
should use Perl bit vectors (see 'perldoc -f vec'). You can use those
to create arbitrary length bit-strings, and they directly support
the usual bitwise logical operators.

It's probably a bit of a bigger conversion job, though (and I
haven't needed to use them before, so there might be awkwardnesses
I haven't anticipated).

>          my $bitpos = 32;
> -        my $insnwidth = 32;
> +        my $wordpos = 0;
> +        my $insnwidth = 0;
>          my $seenblock = 0;
>
>          while (@bits) {
> @@ -217,36 +228,43 @@ sub parse_config_file($)
>
>              my $bitmask = oct("0b". '1' x $bitlen);
>              $bitpos -= $bitlen;
> +            $insnwidth += $bitlen;
>              if ($bitpos < 0) {
>                  print STDERR "$file:$.: ($insn $enc) too many bits specified\n";
>                  exit(1);
>              }
>
>              if (defined $bitval) {
> -                $fixedbits |= ($bitval << $bitpos);
> -                $fixedbitmask |= ($bitmask << $bitpos);
> +                $fixedbits[$wordpos] |= ($bitval << $bitpos);
> +                $fixedbitmask[$wordpos] |= ($bitmask << $bitpos);
>              } else {
> -                push @fields, [ $var, $bitpos, $bitmask ];
> +                push @fields, [ $var, $bitpos, $bitmask, $wordpos ];
> +            }
> +
> +            if ($bitpos == 0) {
> +                check_bitmask($fixedbits[$wordpos], $fixedbitmask[$wordpos]);
> +
> +                $wordpos += 1;
> +                if (@bits) {
> +                    $bitpos = 32;
> +                }
>              }
>          }
>          if ($bitpos == 16) {
>              # assume this is a half-width thumb instruction
>              # Note that we don't fiddle with the bitmasks or positions,
>              # which means the generated insn will be in the high halfword!

I suspect the process of conversion to bit-vectors will probably
imply fixing this bit of ugliness en route, incidentally.

> -            $insnwidth = 16;
> -        } elsif ($bitpos != 0) {
> -            print STDERR "$file:$.: ($insn $enc) not enough bits specified\n";
> +            check_bitmask($fixedbits[$wordpos], $fixedbitmask[$wordpos]);
> +        } elsif ($bitpos != 0 && $bitpos != 32) {
> +            print STDERR "$file:$.: ($insn $enc) not enough bits specified ($bitpos)\n";
>              exit(1);
>          }

thanks
-- PMM