From fff5a47e65c26bf8974bcd8fcc61b7b82430aa78 Mon Sep 17 00:00:00 2001
From: "H. Peter Anvin" <hpa@zytor.com>
Date: Tue, 20 May 2008 09:46:24 -0700
Subject: [PATCH] Same some space by introducing shorthand byte codes for SSE
 prefixes

Properly done, all SSE instructions which has the 66/F2/F3 opcode
multiplex need two prefixes: one to control the use of OSP and one to
control the use of REP.  However, it's a four-way select: np/66/F2/F3;
so introduce shorthand bytecodes for that purpose.
---
 assemble.c | 27 +++++++++++++++++++++++++++
 disasm.c   | 20 ++++++++++++++++++++
 insns.pl   | 10 ++++++----
 3 files changed, 53 insertions(+), 4 deletions(-)

diff --git a/assemble.c b/assemble.c
index 67dfcac..d10e5b0 100644
--- a/assemble.c
+++ b/assemble.c
@@ -91,6 +91,10 @@
  * \335          - disassemble a rep (0xF3 byte) prefix as repe not rep.
  * \340          - reserve <operand 0> bytes of uninitialized storage.
  *                 Operand 0 had better be a segmentless constant.
+ * \360		 - no SSE prefix (== \364\331)
+ * \361          - 66 SSE prefix (== \366\331)
+ * \362          - F2 SSE prefix (== \364\332)
+ * \363          - F3 SSE prefix (== \364\333)
  * \364          - operand-size prefix (0x66) not permitted
  * \365          - address-size prefix (0x67) not permitted
  * \366          - operand-size prefix (0x66) used as opcode extension
@@ -1077,6 +1081,13 @@ static int64_t calcsize(int32_t segment, int64_t offset, int bits,
             else
                 length += ins->oprs[0].offset;
             break;
+	case 0360:
+	    break;
+	case 0361:
+	case 0362:
+	case 0363:
+	    length++;
+	    break;
 	case 0364:
 	case 0365:
 	    break;
@@ -1732,6 +1743,22 @@ static void gencode(int32_t segment, int64_t offset, int bits,
             }
             break;
 
+	case 0360:
+	    break;
+
+	case 0361:
+	    bytes[0] = 0x66;
+            out(offset, segment, bytes, OUT_RAWDATA, 1, NO_SEG, NO_SEG);
+            offset += 1;
+	    break;
+
+	case 0362:
+	case 0363:
+	    bytes[0] = c - 0362 + 0xf2;
+            out(offset, segment, bytes, OUT_RAWDATA, 1, NO_SEG, NO_SEG);
+            offset += 1;
+	    break;
+
 	case 0364:
 	case 0365:
 	    break;
diff --git a/disasm.c b/disasm.c
index 6afc88e..711304c 100644
--- a/disasm.c
+++ b/disasm.c
@@ -874,6 +874,26 @@ static int matches(const struct itemplate *t, uint8_t *data,
 	case 0340:
 	    return false;
 
+	case 0360:
+	    if (prefix->osp || prefix->rep)
+		return false;
+	    break;
+
+	case 0361:
+	    if (!prefix->osp || prefix->rep)
+		return false;
+	    break;
+
+	case 0362:
+	    if (prefix->osp || prefix->rep != 0xf2)
+		return false;
+	    break;
+
+	case 0363:
+	    if (prefix->osp || prefix->rep != 0xf3)
+		return false;
+	    break;
+
 	case 0364:
 	    if (prefix->osp)
 		return false;
diff --git a/insns.pl b/insns.pl
index ac63909..e9dd72f 100644
--- a/insns.pl
+++ b/insns.pl
@@ -539,13 +539,13 @@ sub byte_code_compile($) {
 	} elsif ($prefix_ok && $op =~ /^(66|f2|f3|np)$/) {
 	    # 66/F2/F3 prefix used as an opcode extension, or np = no prefix
 	    if ($op eq '66') {
-		push(@codes, 0366, 0331);
+		push(@codes, 0361);
 	    } elsif ($op eq 'f2') {
-		push(@codes, 0332, 0364);
+		push(@codes, 0362);
 	    } elsif ($op eq 'f3') {
-		push(@codes, 0333, 0364);
+		push(@codes, 0363);
 	    } else {
-		push(@codes, 0331, 0364);
+		push(@codes, 0360);
 	    }
 	} elsif ($op =~ /^[0-9a-f]{2}$/) {
 	    if (defined($litix) && $litix+$codes[$litix]+1 == scalar @codes) {
@@ -678,6 +678,7 @@ sub byte_code_compile($) {
 		die "$0: $line: $op without 'i' and 's' operands\n";
 	    }
 	    push(@codes, 0172, ($oppos{'s'} << 3)+$oppos{'i'});
+	    $prefix_ok = 0;
 	} elsif ($op =~ /^(is4|imz2)\=([0-9]+)$/) {
 	    my $imm = $2;
 	    if (!defined($oppos{'s'})) {
@@ -687,6 +688,7 @@ sub byte_code_compile($) {
 		die "$0: $line: invalid imm4 value for $op: $imm\n";
 	    }
 	    push(@codes, 0173, ($oppos{'s'} << 4) + $imm);
+	    $prefix_ok = 0;
 	} elsif ($op =~ /^([0-9a-f]{2})\+s$/) {
 	    if (!defined($oppos{'i'})) {
 		die "$0: $op without 'i' operand\n";
-- 
2.7.4