OpenE2K
/
binutils-gdb
12
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

[AArch64][PATCH 10/14] Rework code mapping vector types to operand qualifiers. 

ARMv8.2 adds 16-bit floating point operations as an optional extension
to the floating point and Adv.SIMD support. The FP16 additions to the
scalar pairwise group introduce a new vector type. This patch reworks
code in the assembler to allow the addition of the new type.

The new vector type requires the addtion of a new operand qualifier to
the enum aarch64_opnd_qualifier which is defined
include/opcodes/aarch64.h, in the group prefixed by AARCH64_OPN_QLF_V_.

The correctness of the GAS utility function
tc-aarch64.c:vectype_to_qualifier is heavily dependent on the number and
ordering of this group. In particular, it makes assumptions about the
positions of the members of the group that are not true if a qualifier
for type 2H is added before the qualifier for 4H.

This patch reworks the function to weaken its assumptions, making it
calculate positions in the group from the type (B, H, S, D, Q) and
register width.

gas/
2015-12-14  Matthew Wahab  <matthew.wahab@arm.com>

	* config/tc-aarch64.c (vectype_to_qualifier): Calculate operand
	qualifier from per-type base and offet.

Change-Id: I95535864e342a6dec46f69d2696b3900a008f0b1
This commit is contained in:
Matthew Wahab 2015-12-14 17:25:35 +00:00
parent 4b5fc357a1
commit 65f2205d60
2 changed files with 35 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
gas/ChangeLog
View File

@ -1,3 +1,8 @@
2015-12-14 Matthew Wahab <matthew.wahab@arm.com>
* config/tc-aarch64.c (vectype_to_qualifier): Calculate operand
qualifier from per-type base and offet.
2015-12-14 Jan Beulich <jbeulich@suse.com>
* dw2gencfi.c (dot_cfi_label): Free "name".

36
gas/config/tc-aarch64.c
View File

@ -4671,6 +4671,14 @@ vectype_to_qualifier (const struct neon_type_el *vectype)
/* Element size in bytes indexed by neon_el_type. */
const unsigned char ele_size[5]
= {1, 2, 4, 8, 16};
const unsigned int ele_base [5] =
{
AARCH64_OPND_QLF_V_8B,
AARCH64_OPND_QLF_V_4H,
AARCH64_OPND_QLF_V_2S,
AARCH64_OPND_QLF_V_1D,
AARCH64_OPND_QLF_V_1Q
};
if (!vectype->defined || vectype->type == NT_invtype)
goto vectype_conversion_fail;
@ -4685,14 +4693,30 @@ vectype_to_qualifier (const struct neon_type_el *vectype)
/* Vector register. */
int reg_size = ele_size[vectype->type] * vectype->width;
unsigned offset;
unsigned shift;
if (reg_size != 16 && reg_size != 8)
goto vectype_conversion_fail;
/* The conversion is calculated based on the relation of the order of
qualifiers to the vector element size and vector register size. */
offset = (vectype->type == NT_q)
? 8 : (vectype->type << 1) + (reg_size >> 4);
gas_assert (offset <= 8);
return AARCH64_OPND_QLF_V_8B + offset;
/* The conversion is by calculating the offset from the base operand
qualifier for the vector type. The operand qualifiers are regular
enough that the offset can established by shifting the vector width by
a vector-type dependent amount. */
shift = 0;
if (vectype->type == NT_b)
shift = 4;
else if (vectype->type == NT_h)
shift = 3;
else if (vectype->type == NT_s)
shift = 2;
else if (vectype->type >= NT_d)
shift = 1;
else
gas_assert (0);
offset = ele_base [vectype->type] + (vectype->width >> shift);
gas_assert (AARCH64_OPND_QLF_V_8B <= offset
&& offset <= AARCH64_OPND_QLF_V_1Q);
return offset;
}
vectype_conversion_fail: