[PATCH 15/18] x86: handle immediate operands for .insn

[Jan Beulich <jbeulich@suse.com>]

Since we have no insn suffix and it's also not realistic to infer
immediate size from the size of other (typically register) operands
(like optimize_imm() does), and since we also don't have a template
telling us permitted size(s), a new syntax construct is introduced to
allow size (and signedness) specification. In the absence of such, the
size is inferred from significant bits (which obviously may yield
inconsistent results at least for effectively negative values, depending
on whether BFD64 is enabled), and only if supplied expressions can be
evaluated at parsing time. Being explicit is generally recommended to
users.

Size specification is permitted at bit granularity, but of course the
eventually emitted immediate values will be padded up to 8-, 16-, 32-,
or 64-bit fields.
---
RFC: As to relocations emitted, there simply aren't enough to express
     the signed/unsigned distinction the new syntax allows. Similarly
     only significant bit counts 8, 16, 32, and 64 can be honored.
     Question is whether non-representable cases should be warned about
     or even be rejected.

--- a/gas/config/tc-i386.c
+++ b/gas/config/tc-i386.c
@@ -288,6 +288,7 @@ struct _i386_insn
     unsigned int flags[MAX_OPERANDS];
 #define Operand_PCrel 1
 #define Operand_Mem   2
+#define Operand_Signed 4 /* .insn only */
 
     /* Relocation type for operand */
     enum bfd_reloc_code_real reloc[MAX_OPERANDS];
@@ -310,6 +311,9 @@ struct _i386_insn
     /* .insn allows for reserved opcode spaces.  */
     unsigned char insn_opcode_space;
 
+    /* .insn also allows (requires) specifying immediate size.  */
+    unsigned char imm_bits[MAX_OPERANDS];
+
     /* Register is in low 3 bits of opcode.  */
     bool short_form;
 
@@ -5938,6 +5942,10 @@ swap_2_operands (unsigned int xchg1, uns
   i.reloc[xchg2] = i.reloc[xchg1];
   i.reloc[xchg1] = temp_reloc;
 
+  temp_flags = i.imm_bits[xchg2];
+  i.imm_bits[xchg2] = i.imm_bits[xchg1];
+  i.imm_bits[xchg1] = temp_flags;
+
   if (i.mask.reg)
     {
       if (i.mask.operand == xchg1)
@@ -10203,7 +10211,8 @@ output_imm (fragS *insn_start_frag, offs
 
 	      if (i.types[n].bitfield.imm32s
 		  && (i.suffix == QWORD_MNEM_SUFFIX
-		      || (!i.suffix && i.tm.opcode_modifier.no_lsuf)))
+		      || (!i.suffix && i.tm.opcode_modifier.no_lsuf)
+		      || dot_insn ()))
 		sign = 1;
 	      else
 		sign = 0;
@@ -11231,6 +11240,57 @@ s_insn (int dummy ATTRIBUTE_UNUSED)
       if (i.disp_operands && !optimize_disp (&i.tm))
 	goto done;
 
+      /* Establish size for immediate operands.  */
+      for (j = 0; j < i.imm_operands; ++j)
+	{
+	  expressionS *expP = i.op[j].imms;
+
+	  gas_assert (operand_type_check (i.types[j], imm));
+	  operand_type_set (&i.types[j], 0);
+
+	  if (i.imm_bits[j] > 32)
+	    i.types[j].bitfield.imm64 = 1;
+	  else if (i.imm_bits[j] > 16)
+	    {
+	      if (flag_code == CODE_64BIT && (i.flags[j] & Operand_Signed))
+		i.types[j].bitfield.imm32s = 1;
+	      else
+		i.types[j].bitfield.imm32 = 1;
+	    }
+	  else if (i.imm_bits[j] > 8)
+	    i.types[j].bitfield.imm16 = 1;
+	  else if (i.imm_bits[j] > 0)
+	    {
+	      if (i.flags[j] & Operand_Signed)
+		i.types[j].bitfield.imm8s = 1;
+	      else
+		i.types[j].bitfield.imm8 = 1;
+	    }
+	  else if (expP->X_op == O_constant)
+	    {
+	      i.types[j] = smallest_imm_type (expP->X_add_number);
+	      i.types[j].bitfield.imm1 = 0;
+	      /* Oddly enough imm_size() checks imm64 first, so the bit needs
+		 zapping since smallest_imm_type() sets it unconditionally.  */
+	      if (flag_code != CODE_64BIT)
+		{
+		  i.types[j].bitfield.imm64 = 0;
+		  i.types[j].bitfield.imm32s = 0;
+		  i.types[j].bitfield.imm32 = 1;
+		}
+	      else if (i.types[j].bitfield.imm32 || i.types[j].bitfield.imm32s)
+		i.types[j].bitfield.imm64 = 0;
+	    }
+	  else
+	    /* Non-constant expressions are sized heuristically.  */
+	    switch (flag_code)
+	      {
+	      case CODE_64BIT: i.types[j].bitfield.imm32s = 1; break;
+	      case CODE_32BIT: i.types[j].bitfield.imm32 = 1; break;
+	      case CODE_16BIT: i.types[j].bitfield.imm16 = 1; break;
+	      }
+	}
+
       for (j = 0; j < i.operands; ++j)
 	i.tm.operand_types[j] = i.types[j];
 
@@ -11411,10 +11471,11 @@ check_VecOperations (char *op_string)
 	  else if (dot_insn () && *op_string == ':')
 	    {
 	    dot_insn_modifier:
-	      if (op_string[1] == 'd')
+	      switch (op_string[1])
 		{
 		  unsigned long n;
 
+		case 'd':
 		  if (i.memshift < 32)
 		    goto duplicated_vec_op;
 
@@ -11424,6 +11485,27 @@ check_VecOperations (char *op_string)
 		      ++i.memshift;
 		  if (i.memshift < 32 && n == 1)
 		    op_string = end_op;
+		  break;
+
+		case 's': case 'u':
+		  /* This isn't really a "vector" operation, but a sign/size
+		     specifier for immediate operands of .insn.  Note that AT&T
+		     syntax handles the same in i386_immediate().  */
+		  if (!intel_syntax)
+		    break;
+
+		  if (i.imm_bits[this_operand])
+		    goto duplicated_vec_op;
+
+		  n = strtoul (op_string + 2, &end_op, 0);
+		  if (n && n <= (flag_code == CODE_64BIT ? 64 : 32))
+		    {
+		      i.imm_bits[this_operand] = n;
+		      if (op_string[1] == 's')
+			i.flags[this_operand] |= Operand_Signed;
+		      op_string = end_op;
+		    }
+		  break;
 		}
 	    }
 	  /* Check masking operation.  */
@@ -11562,6 +11644,22 @@ i386_immediate (char *imm_start)
 
   exp_seg = expression (exp);
 
+  /* For .insn immediates there may be a size specifier.  */
+  if (dot_insn () && *input_line_pointer == '{' && input_line_pointer[1] == ':'
+      && (input_line_pointer[2] == 's' || input_line_pointer[2] == 'u'))
+    {
+      char *e;
+      unsigned long n = strtoul (input_line_pointer + 3, &e, 0);
+
+      if (*e == '}' && n && n <= (flag_code == CODE_64BIT ? 64 : 32))
+	{
+	  i.imm_bits[this_operand] = n;
+	  if (input_line_pointer[2] == 's')
+	    i.flags[this_operand] |= Operand_Signed;
+	  input_line_pointer = e + 1;
+	}
+    }
+
   SKIP_WHITESPACE ();
   if (*input_line_pointer)
     as_bad (_("junk `%s' after expression"), input_line_pointer);
--- a/gas/config/tc-i386-intel.c
+++ b/gas/config/tc-i386-intel.c
@@ -965,7 +965,8 @@ i386_intel_operand (char *operand_string
       i386_operand_type temp;
 
       /* Register operand.  */
-      if (intel_state.base || intel_state.index || intel_state.seg)
+      if (intel_state.base || intel_state.index || intel_state.seg
+          || i.imm_bits[this_operand])
 	{
 	  as_bad (_("invalid operand"));
 	  return 0;
@@ -998,6 +999,12 @@ i386_intel_operand (char *operand_string
 	   || intel_state.is_mem)
     {
       /* Memory operand.  */
+      if (i.imm_bits[this_operand])
+	{
+	  as_bad (_("invalid operand"));
+	  return 0;
+	}
+
       if (i.mem_operands)
 	{
 	  /* Handle
--- a/gas/testsuite/gas/i386/insn-32.d
+++ b/gas/testsuite/gas/i386/insn-32.d
@@ -10,6 +10,7 @@ Disassembly of section .text:
 [ 	]*[a-f0-9]+:	f3 90[ 	]+pause
 [ 	]*[a-f0-9]+:	f3 90[ 	]+pause
 [ 	]*[a-f0-9]+:	d9 ee[ 	]+fldz
+[ 	]*[a-f0-9]+:	d9 ee[ 	]+fldz
 [ 	]*[a-f0-9]+:	f3 0f 01 e8[ 	]+setssbsy
 [ 	]*[a-f0-9]+:	8b c1[ 	]+mov    %ecx,%eax
 [ 	]*[a-f0-9]+:	66 8b c8[ 	]+mov    %ax,%cx
@@ -17,7 +18,10 @@ Disassembly of section .text:
 [ 	]*[a-f0-9]+:	8b 0c 05 44 44 00 00[ 	]+mov    0x4444\(,%eax,1\),%ecx
 [ 	]*[a-f0-9]+:	66 0f b6 cc[ 	]+movzbw %ah,%cx
 [ 	]*[a-f0-9]+:	0f b7 c8[ 	]+movzwl %ax,%ecx
+[ 	]*[a-f0-9]+:	64 f0 80 30 01[ 	]+lock xorb \$(0x)?1,%fs:\(%eax\)
 [ 	]*[a-f0-9]+:	0f ca[ 	]+bswap  %edx
+[ 	]*[a-f0-9]+:	c7 f8 02 00 00 00[ 	]+xbegin [0-9a-f]+ <insn\+.*>
+[ 	]*[a-f0-9]+:	e2 f8[ 	]+loop   [0-9a-f]+ <insn\+.*>
 [ 	]*[a-f0-9]+:	c5 fc 77[ 	]+vzeroall
 [ 	]*[a-f0-9]+:	c4 e1 7c 77[ 	]+vzeroall
 [ 	]*[a-f0-9]+:	c5 f1 58 d0[ 	]+vaddpd %xmm0,%xmm1,%xmm2
@@ -27,6 +31,9 @@ Disassembly of section .text:
 [ 	]*[a-f0-9]+:	c4 e3 69 68 19 00[ 	]+vfmaddps %xmm0,\(%ecx\),%xmm2,%xmm3
 [ 	]*[a-f0-9]+:	c4 e3 e9 68 19 00[ 	]+vfmaddps \(%ecx\),%xmm0,%xmm2,%xmm3
 [ 	]*[a-f0-9]+:	c4 e3 e9 68 18 10[ 	]+vfmaddps \(%eax\),%xmm1,%xmm2,%xmm3
+[ 	]*[a-f0-9]+:	c4 e3 69 48 19 00[ 	]+vpermil2ps \$(0x)?0,%xmm0,\(%ecx\),%xmm2,%xmm3
+[ 	]*[a-f0-9]+:	c4 e3 e9 48 19 02[ 	]+vpermil2ps \$(0x)?2,\(%ecx\),%xmm0,%xmm2,%xmm3
+[ 	]*[a-f0-9]+:	c4 e3 e9 48 18 13[ 	]+vpermil2ps \$(0x)?3,\(%eax\),%xmm1,%xmm2,%xmm3
 [ 	]*[a-f0-9]+:	c5 f8 92 c8[ 	]+kmovw  %eax,%k1
 [ 	]*[a-f0-9]+:	c5 f8 93 c1[ 	]+kmovw  %k1,%eax
 [ 	]*[a-f0-9]+:	62 f1 74 18 58 d0[ 	]+vaddps \{rn-sae\},%zmm0,%zmm1,%zmm2
@@ -41,4 +48,8 @@ Disassembly of section .text:
 [ 	]*[a-f0-9]+:	62 f5 fd 58 5a 40 01[ 	]+vcvtpd2ph (0x)?8\(%eax\)\{1to8\},%xmm0
 [ 	]*[a-f0-9]+:	62 f5 7c 48 5a 40 01[ 	]+vcvtph2pd 0x10\(%eax\),%zmm0
 [ 	]*[a-f0-9]+:	62 f5 7c 58 5a 40 01[ 	]+vcvtph2pd (0x)?2\(%eax\)\{1to8\},%zmm0
+[ 	]*[a-f0-9]+:	62 f3 7d 28 66 40 01 ff[ 	]+vfpclasspsy \$0xff,0x20\(%eax\),%k0
+[ 	]*[a-f0-9]+:	62 f3 7d 28 66 40 01 ff[ 	]+vfpclasspsy \$0xff,0x20\(%eax\),%k0
+[ 	]*[a-f0-9]+:	62 f3 7d 38 66 40 01 ff[ 	]+vfpclassps \$0xff,(0x)?4\(%eax\)\{1to8\},%k0
+[ 	]*[a-f0-9]+:	62 f3 7d 38 66 40 01 ff[ 	]+vfpclassps \$0xff,(0x)?4\(%eax\)\{1to8\},%k0
 #pass
--- a/gas/testsuite/gas/i386/insn-32.s
+++ b/gas/testsuite/gas/i386/insn-32.s
@@ -9,6 +9,7 @@ insn:
 
 	# fldz
 	.insn 0xd9ee
+	.insn 0xd9, $0xee
 
 	# setssbsy
 	.insn 0xf30f01e8
@@ -23,9 +24,20 @@ insn:
 	.insn 0x0fb6, %ah, %cx
 	.insn 0x0fb7, %eax, %ecx
 
+	# xorb
+	.insn lock 0x80/6, $1, %fs:(%eax)
+
 	# bswap
 	.insn 0x0fc8+r, %edx
 
+1:
+	# xbegin 3f
+	.insn 0xc7f8, $3f-2f{:s32}
+2:
+	# loop 1b
+	.insn 0xe2, $1b-3f{:s8}
+3:
+
 	# vzeroall
 	.insn VEX.256.0F.WIG 0x77
 	.insn {vex3} VEX.L1 0x0f77
@@ -43,6 +55,11 @@ insn:
 	.insn VEX.66.0F3A.W1 0x68, %xmm0, (%ecx), %xmm2, %xmm3
 	.insn VEX.66.0F3A.W1 0x68, (%eax), %xmm1, %xmm2, %xmm3
 
+	# vpermil2ps
+	.insn VEX.66.0F3A.W0 0x48, $0, %xmm0, (%ecx), %xmm2, %xmm3
+	.insn VEX.66.0F3A.W1 0x48, $2, %xmm0, (%ecx), %xmm2, %xmm3
+	.insn VEX.66.0F3A.W1 0x48, $3, (%eax), %xmm1, %xmm2, %xmm3
+
 	# kmovw
 	.insn VEX.L0.0F.W0 0x92, %eax, %k1
 	.insn VEX.L0.0F.W0 0x93, %k1, %eax
@@ -68,3 +85,10 @@ insn:
 	# vcvtph2pd
 	.insn EVEX.M5.W0 0x5a, 16(%eax){:d16}, %zmm0
 	.insn EVEX.M5.W0 0x5a, 2(%eax){1to8:d2}, %zmm0
+
+	.intel_syntax noprefix
+	# vfpclassps
+	.insn EVEX.256.66.0f3a.W0 0x66, k0, [eax+32], 0xff
+	.insn EVEX.66.0f3a.W0 0x66, k0, ymmword ptr [eax+32], 0xff
+	.insn EVEX.256.66.0f3a.W0 0x66, k0, [eax+4]{1to8}, 0xff
+	.insn EVEX.66.0f3a.W0 0x66, k0, dword ptr [eax+4]{1to8}, 0xff
--- a/gas/testsuite/gas/i386/insn-64.d
+++ b/gas/testsuite/gas/i386/insn-64.d
@@ -1,4 +1,4 @@
-#objdump: -dw
+#objdump: -dwr
 #name: .insn (64-bit code)
 
 .*: +file format .*
@@ -18,8 +18,14 @@ Disassembly of section .text:
 [ 	]*[a-f0-9]+:	66 0f be cc[ 	]+movsbw %ah,%cx
 [ 	]*[a-f0-9]+:	0f bf c8[ 	]+movswl %ax,%ecx
 [ 	]*[a-f0-9]+:	48 63 c8[ 	]+movslq %eax,%rcx
+[ 	]*[a-f0-9]+:	f0 80 35 00 00 00 00 01[ 	]+lock xorb \$(0x)?1,(0x)?0\(%rip\) *# .*: R_X86_64_PC32	lock-(0x)?5
 [ 	]*[a-f0-9]+:	48 0f ca[ 	]+bswap  %rdx
 [ 	]*[a-f0-9]+:	41 0f c8[ 	]+bswap  %r8d
+[ 	]*[a-f0-9]+:	c7 f8 02 00 00 00[ 	]+xbegin [0-9a-f]+ <insn\+.*>
+[ 	]*[a-f0-9]+:	e2 f8[ 	]+loop   [0-9a-f]+ <insn\+.*>
+[ 	]*[a-f0-9]+:	05 00 00 00 00[ 	]+add    \$(0x)?0,%eax	.*: R_X86_64_32	var
+[ 	]*[a-f0-9]+:	48 05 00 00 00 00[ 	]+add    \$(0x)?0,%rax	.*: R_X86_64_32S	var
+[ 	]*[a-f0-9]+:	81 3d 00 00 00 00 13 12 23 21[ 	]+cmpl   \$0x21231213,(0x)?0\(%rip\) *# .*: R_X86_64_PC32	var-(0x)?8
 [ 	]*[a-f0-9]+:	c5 fc 77[ 	]+vzeroall
 [ 	]*[a-f0-9]+:	c4 e1 7c 77[ 	]+vzeroall
 [ 	]*[a-f0-9]+:	c4 c1 71 58 d0[ 	]+vaddpd %xmm8,%xmm1,%xmm2
@@ -29,6 +35,9 @@ Disassembly of section .text:
 [ 	]*[a-f0-9]+:	c4 e3 69 68 19 80[ 	]+vfmaddps %xmm8,\(%rcx\),%xmm2,%xmm3
 [ 	]*[a-f0-9]+:	67 c4 e3 e9 68 19 00[ 	]+vfmaddps \(%ecx\),%xmm0,%xmm2,%xmm3
 [ 	]*[a-f0-9]+:	c4 c3 e9 68 18 10[ 	]+vfmaddps \(%r8\),%xmm1,%xmm2,%xmm3
+[ 	]*[a-f0-9]+:	c4 e3 69 48 19 80[ 	]+vpermil2ps \$(0x)0,%xmm8,\(%rcx\),%xmm2,%xmm3
+[ 	]*[a-f0-9]+:	67 c4 e3 e9 48 19 02[ 	]+vpermil2ps \$(0x)2,\(%ecx\),%xmm0,%xmm2,%xmm3
+[ 	]*[a-f0-9]+:	c4 c3 e9 48 18 13[ 	]+vpermil2ps \$(0x)3,\(%r8\),%xmm1,%xmm2,%xmm3
 [ 	]*[a-f0-9]+:	c4 c1 78 92 c8[ 	]+kmovw  %r8d,%k1
 [ 	]*[a-f0-9]+:	c5 78 93 c1[ 	]+kmovw  %k1,%r8d
 [ 	]*[a-f0-9]+:	62 b1 74 38 58 d0[ 	]+vaddps \{rd-sae\},%zmm16,%zmm1,%zmm2
--- a/gas/testsuite/gas/i386/insn-64.s
+++ b/gas/testsuite/gas/i386/insn-64.s
@@ -24,10 +24,30 @@ insn:
 	.insn 0x0fbf, %eax, %ecx
 	.insn 0x63, %rax, %rcx
 
+	# xorb
+	.insn lock 0x80/6, $1, lock(%rip)
+
 	# bswap
 	.insn 0x0fc8+r, %rdx
 	.insn 0x0fc8+r, %r8d
 
+1:
+	# xbegin 3f
+	.insn 0xc7f8, $3f-2f{:s32}
+2:
+	# loop 1b
+	.insn 0xe2, $1b-3f{:s8}
+3:
+
+	# add $var, %eax
+	.insn 0x05, $var{:u32}
+
+	# add $var, %rax
+	.insn rex.w 0x05, $var{:s32}
+
+	# cmpl (32-bit immediate split into two 16-bit halves)
+	.insn 0x81/7, $0x1213, $0x2123, var(%rip)
+
 	# vzeroall
 	.insn VEX.256.0F.WIG 0x77
 	.insn {vex3} VEX.L1 0x0f77
@@ -45,6 +65,11 @@ insn:
 	.insn VEX.66.0F3A.W1 0x68, %xmm0, (%ecx), %xmm2, %xmm3
 	.insn VEX.66.0F3A.W1 0x68, (%r8), %xmm1, %xmm2, %xmm3
 
+	# vpermil2ps
+	.insn VEX.66.0F3A.W0 0x48, $0, %xmm8, (%rcx), %xmm2, %xmm3
+	.insn VEX.66.0F3A.W1 0x48, $2, %xmm0, (%ecx), %xmm2, %xmm3
+	.insn VEX.66.0F3A.W1 0x48, $3, (%r8), %xmm1, %xmm2, %xmm3
+
 	# kmovw
 	.insn VEX.L0.0F.W0 0x92, %r8d, %k1
 	.insn VEX.L0.0F.W0 0x93, %k1, %r8d