* [hsa] Fix bitfield alignment ICEs
@ 2015-10-13 14:04 Martin Jambor
0 siblings, 0 replies; only message in thread
From: Martin Jambor @ 2015-10-13 14:04 UTC (permalink / raw)
To: GCC Patches; +Cc: Martin Liska
Hi,
yesterday I did not notice that I introduced an ICE on testcases with
bit-field memory accesses. The following fixes the issue. A better
solution would be to expand the bit-field parts of memory expressions
separately, which would often allow us to use better aligned accesses,
but that is not a priority at the moment.
Committed to the branch.
Thanks,
Martin
2015-10-13 Martin Jambor <mjambor@suse.cz>
* hsa-gen.c (hsa_bitref_alignment): New function
(gen_hsa_insns_for_load): Use it.
(gen_hsa_insns_for_store): Likewise.
diff --git a/gcc/hsa-gen.c b/gcc/hsa-gen.c
index 8f707b5..7f713f6 100644
--- a/gcc/hsa-gen.c
+++ b/gcc/hsa-gen.c
@@ -2028,6 +2028,38 @@ gen_hsa_insns_for_bitfield_load (hsa_op_reg *dest, hsa_op_address *addr,
gen_hsa_insns_for_bitfield (dest, value_reg, bitsize, bitpos, hbb);
}
+/* Return the alignment of base memory accesses we issue to perform bit-field
+ memory access REF. */
+
+static BrigAlignment8_t
+hsa_bitmemref_alignment (tree ref)
+{
+ unsigned HOST_WIDE_INT bit_offset = 0;
+
+ while (true)
+ {
+ if (TREE_CODE (ref) == BIT_FIELD_REF)
+ {
+ if (!tree_fits_uhwi_p (TREE_OPERAND (ref, 2)))
+ return BRIG_ALIGNMENT_1;
+ bit_offset += tree_to_uhwi (TREE_OPERAND (ref, 2));
+ }
+ else if (TREE_CODE (ref) == COMPONENT_REF
+ && DECL_BIT_FIELD (TREE_OPERAND (ref, 1)))
+ bit_offset += int_bit_position (TREE_OPERAND (ref, 1));
+ else
+ break;
+ ref = TREE_OPERAND (ref, 0);
+ }
+
+ unsigned HOST_WIDE_INT bits = bit_offset % BITS_PER_UNIT;
+ unsigned HOST_WIDE_INT byte_bits = bit_offset - bits;
+ BrigAlignment8_t base = hsa_alignment_encoding (get_object_alignment (ref));
+ if (byte_bits == 0)
+ return base;
+ return MIN (base, hsa_alignment_encoding (byte_bits & -byte_bits));
+}
+
/* Generate HSAIL instructions loading something into register DEST. RHS is
tree representation of the loaded data, which are loaded as type TYPE. Add
instructions to HBB, use SSA_MAP for HSA SSA lookup. */
@@ -2145,11 +2177,9 @@ gen_hsa_insns_for_load (hsa_op_reg *dest, tree rhs, tree type, hsa_bb *hbb,
return;
}
- BrigAlignment8_t req_align;
- req_align = hsa_alignment_encoding (get_object_alignment (rhs));
if (bitsize || bitpos)
gen_hsa_insns_for_bitfield_load (dest, addr, bitsize, bitpos, hbb,
- req_align);
+ hsa_bitmemref_alignment (rhs));
else
{
BrigType16_t mtype;
@@ -2158,7 +2188,7 @@ gen_hsa_insns_for_load (hsa_op_reg *dest, tree rhs, tree type, hsa_bb *hbb,
false));
hsa_insn_mem *mem = new hsa_insn_mem (BRIG_OPCODE_LD, mtype, dest,
addr);
- mem->set_align (req_align);
+ mem->set_align (hsa_alignment_encoding (get_object_alignment (rhs)));
hbb->append_insn (mem);
}
}
@@ -2194,6 +2224,7 @@ gen_hsa_insns_for_store (tree lhs, hsa_op_base *src, hsa_bb *hbb,
vec <hsa_op_reg_p> *ssa_map)
{
HOST_WIDE_INT bitsize = 0, bitpos = 0;
+ BrigAlignment8_t req_align;
BrigType16_t mtype;
mtype = mem_type_for_type (hsa_type_for_scalar_tree_type (TREE_TYPE (lhs),
false));
@@ -2227,10 +2258,11 @@ gen_hsa_insns_for_store (tree lhs, hsa_op_base *src, hsa_bb *hbb,
hsa_op_reg *value_reg = new hsa_op_reg (mem_type);
+ req_align = hsa_bitmemref_alignment (lhs);
/* Load value from memory. */
hsa_insn_mem *mem = new hsa_insn_mem (BRIG_OPCODE_LD, mem_type,
value_reg, addr);
- mem->set_align (hsa_alignment_encoding (get_object_alignment (lhs)));
+ mem->set_align (req_align);
hbb->append_insn (mem);
/* AND the loaded value with prepared mask. */
@@ -2271,9 +2303,11 @@ gen_hsa_insns_for_store (tree lhs, hsa_op_base *src, hsa_bb *hbb,
src = prepared_reg;
mtype = mem_type;
}
+ else
+ req_align = hsa_alignment_encoding (get_object_alignment (lhs));
hsa_insn_mem *mem = new hsa_insn_mem (BRIG_OPCODE_ST, mtype, src, addr);
- mem->set_align (hsa_alignment_encoding (get_object_alignment (lhs)));
+ mem->set_align (req_align);
/* XXX The HSAIL disasm has another constraint: if the source
is an immediate then it must match the destination type. If
^ permalink raw reply [flat|nested] only message in thread
only message in thread, other threads:[~2015-10-13 14:04 UTC | newest]
Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2015-10-13 14:04 [hsa] Fix bitfield alignment ICEs Martin Jambor
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).