x86: improve matching diagnostics when "accumulator" registers are involved

In templates, the expectation of an "accumulator" register to be used is
expressed solely by operand size; there's no "class" specifier there.
Hence operand_size_match() is too eager in invoking
match_{operand,simd}_size(), resulting in "operand size mismatch" errors
when it's the type (of register), not the size that's wrong.
Interestingly adjustments there alone lead to no error at all then: To
"compensate", operand_type_match() needs to disambiguate register types
when register instances are specified in the template (matching the
actual operand), by checking a match (overlap) in operand sizes.
diff --git a/gas/config/tc-i386.c b/gas/config/tc-i386.c
index 7c9b94a..69b5677 100644
--- a/gas/config/tc-i386.c
+++ b/gas/config/tc-i386.c
@@ -2611,7 +2611,12 @@
 
       if (i.types[j].bitfield.class == Reg
 	  && (t->operand_types[j].bitfield.class == Reg
-	      || t->operand_types[j].bitfield.instance == Accum)
+	      || (t->operand_types[j].bitfield.instance == Accum
+		  && (t->operand_types[j].bitfield.byte
+		      || t->operand_types[j].bitfield.word
+		      || t->operand_types[j].bitfield.dword
+		      || t->operand_types[j].bitfield.qword
+		      || t->operand_types[j].bitfield.tbyte)))
 	  && !match_operand_size (t, j, j))
 	{
 	  match = 0;
@@ -2620,7 +2625,9 @@
 
       if (i.types[j].bitfield.class == RegSIMD
 	  && (t->operand_types[j].bitfield.class == RegSIMD
-	      || t->operand_types[j].bitfield.instance == Accum)
+	      || (t->operand_types[j].bitfield.instance == Accum
+		  /* Note: %ymm0, %zmm0, and %tmm0 aren't marked Accum.  */
+		  && t->operand_types[j].bitfield.xmmword))
 	  && !match_simd_size (t, j, j))
 	{
 	  match = 0;
@@ -2655,7 +2662,12 @@
 
       if (i.types[given].bitfield.class == Reg
 	  && (t->operand_types[j].bitfield.class == Reg
-	      || t->operand_types[j].bitfield.instance == Accum)
+	      || (t->operand_types[j].bitfield.instance == Accum
+		  && (t->operand_types[j].bitfield.byte
+		      || t->operand_types[j].bitfield.word
+		      || t->operand_types[j].bitfield.dword
+		      || t->operand_types[j].bitfield.qword
+		      || t->operand_types[j].bitfield.tbyte)))
 	  && !match_operand_size (t, j, given))
 	return match;
 
@@ -2695,6 +2707,23 @@
   if (operand_type_all_zero (&temp))
     goto mismatch;
 
+  /* When a (register) instance is expected, operand size needs checking
+     to disambiguate.  */
+  if (overlap.bitfield.instance != InstanceNone
+      && !overlap.bitfield.byte
+      && !overlap.bitfield.word
+      && !overlap.bitfield.dword
+      && !overlap.bitfield.qword
+      && !overlap.bitfield.tbyte
+      && !overlap.bitfield.xmmword
+      && !overlap.bitfield.ymmword
+      && !overlap.bitfield.zmmword
+      && !overlap.bitfield.tmmword)
+    {
+      gas_assert (overlap.bitfield.class == ClassNone);
+      goto mismatch;
+    }
+
   if (given.bitfield.baseindex == overlap.bitfield.baseindex)
     return 1;
 
diff --git a/gas/testsuite/gas/i386/inval-type.l b/gas/testsuite/gas/i386/inval-type.l
index 5784949e..7188cd5 100644
--- a/gas/testsuite/gas/i386/inval-type.l
+++ b/gas/testsuite/gas/i386/inval-type.l
@@ -1,6 +1,10 @@
 .*: Assembler messages:
+.*: Error: operand type mismatch for .blendvps.
+.*: Error: operand type mismatch for .blendvps.
 .*: Error: operand type mismatch for .bsf.
 .*: Error: operand type mismatch for .bswap.
 .*: Error: operand type mismatch for .bswap.
+.*: Error: operand type mismatch for .fcomi.
+.*: Error: operand type mismatch for .in.
 .*: Error: operand type mismatch for .movntdqa.
 .*: Error: operand type mismatch for .movntdq.
diff --git a/gas/testsuite/gas/i386/inval-type.s b/gas/testsuite/gas/i386/inval-type.s
index 493d12e..78cda94 100644
--- a/gas/testsuite/gas/i386/inval-type.s
+++ b/gas/testsuite/gas/i386/inval-type.s
@@ -3,8 +3,12 @@
 # All the following should yield "operand type mismatch" (or something yet more
 # specific), but _not_ "operand size mismatch".
 
+	blendvps %eax, %xmm1, %xmm2
+	blendvps %st, %xmm1, %xmm2
 	bsf	%eax, (%eax)
 	bswap	%mm0
 	bswapl	%xmm0
+	fcomi	%st(1), %xmm0
+	inl	$0, %xmm0
 	movntdqa %xmm0, (%eax)
 	movntdq	(%eax), %xmm0