x86/Intel: fix fallout from earlier template folding

While many templates allowing multiple suitably matching XMM/YMM/ZMM
operand sizes can be folded, a few need to be split in order to not
wrongly accept "xmmword ptr" operands when only XMM registers are
permitted (and memory operands are more narrow). Add a test case
validating this.

5 files changed, 244 insertions, 2 deletions

diff --git a/gas/ChangeLog b/gas/ChangeLog
index 1e0ebb2..00a4edc 100644
--- a/gas/ChangeLog
+++ b/gas/ChangeLog
@@ -1,5 +1,13 @@
 2018-03-22  Jan Beulich  <jbeulich@suse.com>
 
+	* config/tc-i386.c (match_mem_size): Extend sub-xmmword
+	exceptions.
+	* testsuite/gas/i386/xmmword.l, testsuite/gas/i386/xmmword.s:
+	New.
+	* testsuite/gas/i386/i386.exp: Run new test.
+
+2018-03-22  Jan Beulich  <jbeulich@suse.com>
+
 	* config/tc-i386.c (match_template): Also match 2nd and 4th
 	operand's register sizes.
 	* testsuite/gas/i386/unspec.l, testsuite/gas/i386/unspec.s: New.
diff --git a/gas/config/tc-i386.c b/gas/config/tc-i386.c
index 01365f3..8905de4 100644
--- a/gas/config/tc-i386.c
+++ b/gas/config/tc-i386.c
@@ -1937,10 +1937,13 @@ match_mem_size (const insn_template *t, unsigned int j)
 		   && !t->operand_types[j].bitfield.fword)
 	       /* For scalar opcode templates to allow register and memory
 		  operands at the same time, some special casing is needed
-		  here.  */
+		  here.  Also for v{,p}broadcast*, {,v}pmov{s,z}*, and
+		  down-conversion vpmov*.  */
 	       || ((t->operand_types[j].bitfield.regsimd
 		    && !t->opcode_modifier.broadcast
-		    && (t->operand_types[j].bitfield.dword
+		    && (t->operand_types[j].bitfield.byte
+			|| t->operand_types[j].bitfield.word
+			|| t->operand_types[j].bitfield.dword
 			|| t->operand_types[j].bitfield.qword))
 		   ? (i.types[j].bitfield.xmmword
 		      || i.types[j].bitfield.ymmword
diff --git a/gas/testsuite/gas/i386/i386.exp b/gas/testsuite/gas/i386/i386.exp
index 2fbb52b..035d510 100644
--- a/gas/testsuite/gas/i386/i386.exp
+++ b/gas/testsuite/gas/i386/i386.exp
@@ -553,6 +553,7 @@ if [expr [istarget "i*86-*-*"] || [istarget "x86_64-*-*"]] then {
     run_list_test "string-bad" ""
     run_list_test "reg-bad" ""
     run_list_test "space1" "-al"
+    run_list_test "xmmword" ""
     run_dump_test rept
     run_dump_test pr19498
     run_list_test "nop-bad-1" ""
diff --git a/gas/testsuite/gas/i386/xmmword.l b/gas/testsuite/gas/i386/xmmword.l
new file mode 100644
index 0000000..ce3af8d
--- /dev/null
+++ b/gas/testsuite/gas/i386/xmmword.l
@@ -0,0 +1,101 @@
+.*: Assembler messages:
+.*:4: Error: .* `addsd'
+.*:5: Error: .* `vaddsd'
+.*:6: Error: .* `vaddsd'
+.*:8: Error: .* `addss'
+.*:9: Error: .* `vaddss'
+.*:10: Error: .* `vaddss'
+.*:12: Error: .* `vbroadcastf32x2'
+.*:13: Error: .* `vbroadcastf32x2'
+.*:15: Error: .* `vbroadcasti32x2'
+.*:16: Error: .* `vbroadcasti32x2'
+.*:17: Error: .* `vbroadcasti32x2'
+.*:19: Error: .* `vbroadcastsd'
+.*:20: Error: .* `vbroadcastsd'
+.*:21: Error: .* `vbroadcastsd'
+.*:23: Error: .* `vbroadcastss'
+.*:24: Error: .* `vbroadcastss'
+.*:25: Error: .* `vbroadcastss'
+.*:26: Error: .* `vbroadcastss'
+.*:27: Error: .* `vbroadcastss'
+.*:29: Error: .* `cvtdq2pd'
+.*:30: Error: .* `vcvtdq2pd'
+.*:31: Error: .* `vcvtdq2pd'
+.*:33: Error: .* `vcvtph2ps'
+.*:34: Error: .* `vcvtph2ps'
+.*:36: Error: .* `cvtps2pd'
+.*:37: Error: .* `vcvtps2pd'
+.*:38: Error: .* `vcvtps2pd'
+.*:40: Error: .* `vcvtps2ph'
+.*:41: Error: .* `vcvtps2ph'
+.*:43: Error: .* `vcvtudq2pd'
+.*:45: Error: .* `insertps'
+.*:46: Error: .* `vinsertps'
+.*:47: Error: .* `vinsertps'
+.*:49: Error: .* `movddup'
+.*:50: Error: .* `vmovddup'
+.*:51: Error: .* `vmovddup'
+.*:53: Error: .* `vpbroadcastb'
+.*:54: Error: .* `vpbroadcastb'
+.*:55: Error: .* `vpbroadcastb'
+.*:56: Error: .* `vpbroadcastb'
+.*:57: Error: .* `vpbroadcastb'
+.*:59: Error: .* `vpbroadcastd'
+.*:60: Error: .* `vpbroadcastd'
+.*:61: Error: .* `vpbroadcastd'
+.*:62: Error: .* `vpbroadcastd'
+.*:63: Error: .* `vpbroadcastd'
+.*:65: Error: .* `vpbroadcastq'
+.*:66: Error: .* `vpbroadcastq'
+.*:67: Error: .* `vpbroadcastq'
+.*:68: Error: .* `vpbroadcastq'
+.*:69: Error: .* `vpbroadcastq'
+.*:71: Error: .* `vpbroadcastw'
+.*:72: Error: .* `vpbroadcastw'
+.*:73: Error: .* `vpbroadcastw'
+.*:74: Error: .* `vpbroadcastw'
+.*:75: Error: .* `vpbroadcastw'
+.*:77: Error: .* `pmovsxbd'
+.*:78: Error: .* `vpmovsxbd'
+.*:79: Error: .* `vpmovsxbd'
+.*:80: Error: .* `vpmovsxbd'
+.*:81: Error: .* `vpmovsxbd'
+.*:83: Error: .* `pmovsxbq'
+.*:84: Error: .* `vpmovsxbq'
+.*:85: Error: .* `vpmovsxbq'
+.*:86: Error: .* `vpmovsxbq'
+.*:87: Error: .* `vpmovsxbq'
+.*:88: Error: .* `vpmovsxbq'
+.*:90: Error: .* `pmovsxdq'
+.*:91: Error: .* `vpmovsxdq'
+.*:92: Error: .* `vpmovsxdq'
+.*:94: Error: .* `pmovsxwd'
+.*:95: Error: .* `vpmovsxwd'
+.*:96: Error: .* `vpmovsxwd'
+.*:98: Error: .* `pmovsxwq'
+.*:99: Error: .* `vpmovsxwq'
+.*:100: Error: .* `vpmovsxwq'
+.*:101: Error: .* `vpmovsxwq'
+.*:102: Error: .* `vpmovsxwq'
+.*:104: Error: .* `pmovzxbd'
+.*:105: Error: .* `vpmovzxbd'
+.*:106: Error: .* `vpmovzxbd'
+.*:107: Error: .* `vpmovzxbd'
+.*:108: Error: .* `vpmovzxbd'
+.*:110: Error: .* `pmovzxbq'
+.*:111: Error: .* `vpmovzxbq'
+.*:112: Error: .* `vpmovzxbq'
+.*:113: Error: .* `vpmovzxbq'
+.*:114: Error: .* `vpmovzxbq'
+.*:115: Error: .* `vpmovzxbq'
+.*:117: Error: .* `pmovzxdq'
+.*:118: Error: .* `vpmovzxdq'
+.*:119: Error: .* `vpmovzxdq'
+.*:121: Error: .* `pmovzxwd'
+.*:122: Error: .* `vpmovzxwd'
+.*:123: Error: .* `vpmovzxwd'
+.*:125: Error: .* `pmovzxwq'
+.*:126: Error: .* `vpmovzxwq'
+.*:127: Error: .* `vpmovzxwq'
+.*:128: Error: .* `vpmovzxwq'
+.*:129: Error: .* `vpmovzxwq'
diff --git a/gas/testsuite/gas/i386/xmmword.s b/gas/testsuite/gas/i386/xmmword.s
new file mode 100644
index 0000000..47d2d8d
--- /dev/null
+++ b/gas/testsuite/gas/i386/xmmword.s
@@ -0,0 +1,129 @@
+	.text
+	.intel_syntax noprefix
+xmmword:
+	addsd		xmm0, xmmword ptr [eax]
+	vaddsd		xmm0, xmm0, xmmword ptr [eax]
+	vaddsd		xmm0{k7}, xmm0, xmmword ptr [eax]
+
+	addss		xmm0, xmmword ptr [eax]
+	vaddss		xmm0, xmm0, xmmword ptr [eax]
+	vaddss		xmm0{k7}, xmm0, xmmword ptr [eax]
+
+	vbroadcastf32x2	ymm0, xmmword ptr [eax]
+	vbroadcastf32x2	zmm0, xmmword ptr [eax]
+
+	vbroadcasti32x2	xmm0, xmmword ptr [eax]
+	vbroadcasti32x2	ymm0, xmmword ptr [eax]
+	vbroadcasti32x2	zmm0, xmmword ptr [eax]
+
+	vbroadcastsd	ymm0, xmmword ptr [eax]
+	vbroadcastsd	ymm0{k7}, xmmword ptr [eax]
+	vbroadcastsd	zmm0{k7}, xmmword ptr [eax]
+
+	vbroadcastss	xmm0, xmmword ptr [eax]
+	vbroadcastss	xmm0{k7}, xmmword ptr [eax]
+	vbroadcastss	ymm0, xmmword ptr [eax]
+	vbroadcastss	ymm0{k7}, xmmword ptr [eax]
+	vbroadcastss	zmm0, xmmword ptr [eax]
+
+	cvtdq2pd	xmm0, xmmword ptr [eax]
+	vcvtdq2pd	xmm0, xmmword ptr [eax]
+	vcvtdq2pd	xmm0{k7}, xmmword ptr [eax]
+
+	vcvtph2ps	xmm0, xmmword ptr [eax]
+	vcvtph2ps	xmm0{k7}, xmmword ptr [eax]
+
+	cvtps2pd	xmm0, xmmword ptr [eax]
+	vcvtps2pd	xmm0, xmmword ptr [eax]
+	vcvtps2pd	xmm0{k7}, xmmword ptr [eax]
+
+	vcvtps2ph	xmmword ptr [eax], xmm0, 0
+	vcvtps2ph	xmmword ptr [eax]{k7}, xmm0, 0
+
+	vcvtudq2pd	xmm0, xmmword ptr [eax]
+
+	insertps	xmm0, xmmword ptr [eax], 0
+	vinsertps	xmm0, xmm0, xmmword ptr [eax], 0
+	{evex} vinsertps xmm0, xmm0, xmmword ptr [eax], 0
+
+	movddup		xmm0, xmmword ptr [eax]
+	vmovddup	xmm0, xmmword ptr [eax]
+	vmovddup	xmm0{k7}, xmmword ptr [eax]
+
+	vpbroadcastb	xmm0, xmmword ptr [eax]
+	vpbroadcastb	xmm0{k7}, xmmword ptr [eax]
+	vpbroadcastb	ymm0, xmmword ptr [eax]
+	vpbroadcastb	ymm0{k7}, xmmword ptr [eax]
+	vpbroadcastb	zmm0, xmmword ptr [eax]
+
+	vpbroadcastd	xmm0, xmmword ptr [eax]
+	vpbroadcastd	xmm0{k7}, xmmword ptr [eax]
+	vpbroadcastd	ymm0, xmmword ptr [eax]
+	vpbroadcastd	ymm0{k7}, xmmword ptr [eax]
+	vpbroadcastd	zmm0, xmmword ptr [eax]
+
+	vpbroadcastq	xmm0, xmmword ptr [eax]
+	vpbroadcastq	xmm0{k7}, xmmword ptr [eax]
+	vpbroadcastq	ymm0, xmmword ptr [eax]
+	vpbroadcastq	ymm0{k7}, xmmword ptr [eax]
+	vpbroadcastq	zmm0, xmmword ptr [eax]
+
+	vpbroadcastw	xmm0, xmmword ptr [eax]
+	vpbroadcastw	xmm0{k7}, xmmword ptr [eax]
+	vpbroadcastw	ymm0, xmmword ptr [eax]
+	vpbroadcastw	ymm0{k7}, xmmword ptr [eax]
+	vpbroadcastw	zmm0, xmmword ptr [eax]
+
+	pmovsxbd	xmm0, xmmword ptr [eax]
+	vpmovsxbd	xmm0, xmmword ptr [eax]
+	vpmovsxbd	xmm0{k7}, xmmword ptr [eax]
+	vpmovsxbd	ymm0, xmmword ptr [eax]
+	vpmovsxbd	ymm0{k7}, xmmword ptr [eax]
+
+	pmovsxbq	xmm0, xmmword ptr [eax]
+	vpmovsxbq	xmm0, xmmword ptr [eax]
+	vpmovsxbq	xmm0{k7}, xmmword ptr [eax]
+	vpmovsxbq	ymm0, xmmword ptr [eax]
+	vpmovsxbq	ymm0{k7}, xmmword ptr [eax]
+	vpmovsxbq	zmm0, xmmword ptr [eax]
+
+	pmovsxdq	xmm0, xmmword ptr [eax]
+	vpmovsxdq	xmm0, xmmword ptr [eax]
+	vpmovsxdq	xmm0{k7}, xmmword ptr [eax]
+
+	pmovsxwd	xmm0, xmmword ptr [eax]
+	vpmovsxwd	xmm0, xmmword ptr [eax]
+	vpmovsxwd	xmm0{k7}, xmmword ptr [eax]
+
+	pmovsxwq	xmm0, xmmword ptr [eax]
+	vpmovsxwq	xmm0, xmmword ptr [eax]
+	vpmovsxwq	xmm0{k7}, xmmword ptr [eax]
+	vpmovsxwq	ymm0, xmmword ptr [eax]
+	vpmovsxwq	ymm0{k7}, xmmword ptr [eax]
+
+	pmovzxbd	xmm0, xmmword ptr [eax]
+	vpmovzxbd	xmm0, xmmword ptr [eax]
+	vpmovzxbd	xmm0{k7}, xmmword ptr [eax]
+	vpmovzxbd	ymm0, xmmword ptr [eax]
+	vpmovzxbd	ymm0{k7}, xmmword ptr [eax]
+
+	pmovzxbq	xmm0, xmmword ptr [eax]
+	vpmovzxbq	xmm0, xmmword ptr [eax]
+	vpmovzxbq	xmm0{k7}, xmmword ptr [eax]
+	vpmovzxbq	ymm0, xmmword ptr [eax]
+	vpmovzxbq	ymm0{k7}, xmmword ptr [eax]
+	vpmovzxbq	zmm0, xmmword ptr [eax]
+
+	pmovzxdq	xmm0, xmmword ptr [eax]
+	vpmovzxdq	xmm0, xmmword ptr [eax]
+	vpmovzxdq	xmm0{k7}, xmmword ptr [eax]
+
+	pmovzxwd	xmm0, xmmword ptr [eax]
+	vpmovzxwd	xmm0, xmmword ptr [eax]
+	vpmovzxwd	xmm0{k7}, xmmword ptr [eax]
+
+	pmovzxwq	xmm0, xmmword ptr [eax]
+	vpmovzxwq	xmm0, xmmword ptr [eax]
+	vpmovzxwq	xmm0{k7}, xmmword ptr [eax]
+	vpmovzxwq	ymm0, xmmword ptr [eax]
+	vpmovzxwq	ymm0{k7}, xmmword ptr [eax]