OpenE2K
/
gcc
2
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

re PR target/54252 (Bad alignment code generated for Neon loads) 

Fix PR target/54252

2012-09-07  Ramana Radhakrishnan  <ramana.radhakrishnan@arm.com>

	Backport from mainline.
	2012-08-29  Ramana Radhakrishnan  <ramana.radhakrishnan@arm.com>
                   Richard Earnshaw  <richard.earnshaw@arm.com>

	PR target/54252
	* config/arm/arm.c (neon_dereference_pointer): Adjust nelems by
	element size. Use elem_type from the formal parameter. New parameter
	fcode.
	(neon_expand_args): Adjust call to neon_dereference_pointer.

From-SVN: r191059
This commit is contained in:
Ramana Radhakrishnan 2012-09-07 09:31:54 +00:00 committed by Ramana Radhakrishnan
parent 947305cf6e
commit ab1c58629a
2 changed files with 51 additions and 30 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
gcc/ChangeLog
View File

@ -1,3 +1,15 @@
2012-09-07 Ramana Radhakrishnan <ramana.radhakrishnan@arm.com>
Backport from mainline.
2012-08-29 Ramana Radhakrishnan <ramana.radhakrishnan@arm.com>
Richard Earnshaw <richard.earnshaw@arm.com>
PR target/54252
* config/arm/arm.c (neon_dereference_pointer): Adjust nelems by
element size. Use elem_type from the formal parameter. New parameter
fcode.
(neon_expand_args): Adjust call to neon_dereference_pointer.
2012-09-06 Uros Bizjak <ubizjak@gmail.com>
* config/i386/sse.md (<sse4_1_avx2>_pblendvb): Use

69
gcc/config/arm/arm.c
View File

@ -20568,12 +20568,13 @@ typedef enum {
and return an expression for the accessed memory.
The intrinsic function operates on a block of registers that has
mode REG_MODE. This block contains vectors of type TYPE_MODE.
The function references the memory at EXP in mode MEM_MODE;
this mode may be BLKmode if no more suitable mode is available. */
mode REG_MODE. This block contains vectors of type TYPE_MODE. The
function references the memory at EXP of type TYPE and in mode
MEM_MODE; this mode may be BLKmode if no more suitable mode is
available. */
static tree
neon_dereference_pointer (tree exp, enum machine_mode mem_mode,
neon_dereference_pointer (tree exp, tree type, enum machine_mode mem_mode,
enum machine_mode reg_mode,
neon_builtin_type_mode type_mode)
{
@ -20591,18 +20592,18 @@ neon_dereference_pointer (tree exp, enum machine_mode mem_mode,
gcc_assert (reg_size % vector_size == 0);
nvectors = reg_size / vector_size;
/* Work out the type of each element. */
gcc_assert (POINTER_TYPE_P (type));
elem_type = TREE_TYPE (type);
/* Work out how many elements are being loaded or stored.
MEM_MODE == REG_MODE implies a one-to-one mapping between register
and memory elements; anything else implies a lane load or store. */
if (mem_mode == reg_mode)
nelems = vector_size * nvectors;
nelems = vector_size * nvectors / int_size_in_bytes (elem_type);
else
nelems = nvectors;
/* Work out the type of each element. */
gcc_assert (POINTER_TYPE_P (TREE_TYPE (exp)));
elem_type = TREE_TYPE (TREE_TYPE (exp));
/* Create a type that describes the full access. */
upper_bound = build_int_cst (size_type_node, nelems - 1);
array_type = build_array_type (elem_type, build_index_type (upper_bound));
@ -20616,12 +20617,14 @@ neon_dereference_pointer (tree exp, enum machine_mode mem_mode,
static rtx
arm_expand_neon_args (rtx target, int icode, int have_retval,
neon_builtin_type_mode type_mode,
tree exp, ...)
tree exp, int fcode, ...)
{
va_list ap;
rtx pat;
tree arg[NEON_MAX_BUILTIN_ARGS];
rtx op[NEON_MAX_BUILTIN_ARGS];
tree arg_type;
tree formals;
enum machine_mode tmode = insn_data[icode].operand[0].mode;
enum machine_mode mode[NEON_MAX_BUILTIN_ARGS];
enum machine_mode other_mode;
@ -20634,7 +20637,9 @@ arm_expand_neon_args (rtx target, int icode, int have_retval,
|| !(*insn_data[icode].operand[0].predicate) (target, tmode)))
target = gen_reg_rtx (tmode);
va_start (ap, exp);
va_start (ap, fcode);
formals = TYPE_ARG_TYPES (TREE_TYPE (arm_builtin_decls[fcode]));
for (;;)
{
@ -20647,12 +20652,15 @@ arm_expand_neon_args (rtx target, int icode, int have_retval,
opno = argc + have_retval;
mode[argc] = insn_data[icode].operand[opno].mode;
arg[argc] = CALL_EXPR_ARG (exp, argc);
arg_type = TREE_VALUE (formals);
if (thisarg == NEON_ARG_MEMORY)
{
other_mode = insn_data[icode].operand[1 - opno].mode;
arg[argc] = neon_dereference_pointer (arg[argc], mode[argc],
other_mode, type_mode);
arg[argc] = neon_dereference_pointer (arg[argc], arg_type,
mode[argc], other_mode,
type_mode);
}
op[argc] = expand_normal (arg[argc]);
switch (thisarg)
@ -20689,6 +20697,7 @@ arm_expand_neon_args (rtx target, int icode, int have_retval,
}
argc++;
formals = TREE_CHAIN (formals);
}
}
@ -20771,7 +20780,7 @@ arm_expand_neon_builtin (int fcode, tree exp, rtx target)
case NEON_UNOP:
case NEON_CONVERT:
case NEON_DUPLANE:
return arm_expand_neon_args (target, icode, 1, type_mode, exp,
return arm_expand_neon_args (target, icode, 1, type_mode, exp, fcode,
NEON_ARG_COPY_TO_REG, NEON_ARG_CONSTANT, NEON_ARG_STOP);
case NEON_BINOP:
@ -20781,89 +20790,89 @@ arm_expand_neon_builtin (int fcode, tree exp, rtx target)
case NEON_SCALARMULH:
case NEON_SHIFTINSERT:
case NEON_LOGICBINOP:
return arm_expand_neon_args (target, icode, 1, type_mode, exp,
return arm_expand_neon_args (target, icode, 1, type_mode, exp, fcode,
NEON_ARG_COPY_TO_REG, NEON_ARG_COPY_TO_REG, NEON_ARG_CONSTANT,
NEON_ARG_STOP);
case NEON_TERNOP:
return arm_expand_neon_args (target, icode, 1, type_mode, exp,
return arm_expand_neon_args (target, icode, 1, type_mode, exp, fcode,
NEON_ARG_COPY_TO_REG, NEON_ARG_COPY_TO_REG, NEON_ARG_COPY_TO_REG,
NEON_ARG_CONSTANT, NEON_ARG_STOP);
case NEON_GETLANE:
case NEON_FIXCONV:
case NEON_SHIFTIMM:
return arm_expand_neon_args (target, icode, 1, type_mode, exp,
return arm_expand_neon_args (target, icode, 1, type_mode, exp, fcode,
NEON_ARG_COPY_TO_REG, NEON_ARG_CONSTANT, NEON_ARG_CONSTANT,
NEON_ARG_STOP);
case NEON_CREATE:
return arm_expand_neon_args (target, icode, 1, type_mode, exp,
return arm_expand_neon_args (target, icode, 1, type_mode, exp, fcode,
NEON_ARG_COPY_TO_REG, NEON_ARG_STOP);
case NEON_DUP:
case NEON_SPLIT:
case NEON_REINTERP:
return arm_expand_neon_args (target, icode, 1, type_mode, exp,
return arm_expand_neon_args (target, icode, 1, type_mode, exp, fcode,
NEON_ARG_COPY_TO_REG, NEON_ARG_STOP);
case NEON_COMBINE:
case NEON_VTBL:
return arm_expand_neon_args (target, icode, 1, type_mode, exp,
return arm_expand_neon_args (target, icode, 1, type_mode, exp, fcode,
NEON_ARG_COPY_TO_REG, NEON_ARG_COPY_TO_REG, NEON_ARG_STOP);
case NEON_RESULTPAIR:
return arm_expand_neon_args (target, icode, 0, type_mode, exp,
return arm_expand_neon_args (target, icode, 0, type_mode, exp, fcode,
NEON_ARG_COPY_TO_REG, NEON_ARG_COPY_TO_REG, NEON_ARG_COPY_TO_REG,
NEON_ARG_STOP);
case NEON_LANEMUL:
case NEON_LANEMULL:
case NEON_LANEMULH:
return arm_expand_neon_args (target, icode, 1, type_mode, exp,
return arm_expand_neon_args (target, icode, 1, type_mode, exp, fcode,
NEON_ARG_COPY_TO_REG, NEON_ARG_COPY_TO_REG, NEON_ARG_CONSTANT,
NEON_ARG_CONSTANT, NEON_ARG_STOP);
case NEON_LANEMAC:
return arm_expand_neon_args (target, icode, 1, type_mode, exp,
return arm_expand_neon_args (target, icode, 1, type_mode, exp, fcode,
NEON_ARG_COPY_TO_REG, NEON_ARG_COPY_TO_REG, NEON_ARG_COPY_TO_REG,
NEON_ARG_CONSTANT, NEON_ARG_CONSTANT, NEON_ARG_STOP);
case NEON_SHIFTACC:
return arm_expand_neon_args (target, icode, 1, type_mode, exp,
return arm_expand_neon_args (target, icode, 1, type_mode, exp, fcode,
NEON_ARG_COPY_TO_REG, NEON_ARG_COPY_TO_REG, NEON_ARG_CONSTANT,
NEON_ARG_CONSTANT, NEON_ARG_STOP);
case NEON_SCALARMAC:
return arm_expand_neon_args (target, icode, 1, type_mode, exp,
return arm_expand_neon_args (target, icode, 1, type_mode, exp, fcode,
NEON_ARG_COPY_TO_REG, NEON_ARG_COPY_TO_REG, NEON_ARG_COPY_TO_REG,
NEON_ARG_CONSTANT, NEON_ARG_STOP);
case NEON_SELECT:
case NEON_VTBX:
return arm_expand_neon_args (target, icode, 1, type_mode, exp,
return arm_expand_neon_args (target, icode, 1, type_mode, exp, fcode,
NEON_ARG_COPY_TO_REG, NEON_ARG_COPY_TO_REG, NEON_ARG_COPY_TO_REG,
NEON_ARG_STOP);
case NEON_LOAD1:
case NEON_LOADSTRUCT:
return arm_expand_neon_args (target, icode, 1, type_mode, exp,
return arm_expand_neon_args (target, icode, 1, type_mode, exp, fcode,
NEON_ARG_MEMORY, NEON_ARG_STOP);
case NEON_LOAD1LANE:
case NEON_LOADSTRUCTLANE:
return arm_expand_neon_args (target, icode, 1, type_mode, exp,
return arm_expand_neon_args (target, icode, 1, type_mode, exp, fcode,
NEON_ARG_MEMORY, NEON_ARG_COPY_TO_REG, NEON_ARG_CONSTANT,
NEON_ARG_STOP);
case NEON_STORE1:
case NEON_STORESTRUCT:
return arm_expand_neon_args (target, icode, 0, type_mode, exp,
return arm_expand_neon_args (target, icode, 0, type_mode, exp, fcode,
NEON_ARG_MEMORY, NEON_ARG_COPY_TO_REG, NEON_ARG_STOP);
case NEON_STORE1LANE:
case NEON_STORESTRUCTLANE:
return arm_expand_neon_args (target, icode, 0, type_mode, exp,
return arm_expand_neon_args (target, icode, 0, type_mode, exp, fcode,
NEON_ARG_MEMORY, NEON_ARG_COPY_TO_REG, NEON_ARG_CONSTANT,
NEON_ARG_STOP);
}