* [RFC/RFA][PATCH v2 01/12] Implement internal functions for efficient CRC computation
@ 2024-07-26 18:05 Mariam Arutunian
2024-08-25 17:51 ` Jeff Law
0 siblings, 1 reply; 2+ messages in thread
From: Mariam Arutunian @ 2024-07-26 18:05 UTC (permalink / raw)
To: GCC Patches
[-- Attachment #1.1: Type: text/plain, Size: 1796 bytes --]
Add two new internal functions (IFN_CRC, IFN_CRC_REV), to provide faster
CRC generation.
One performs bit-forward and the other bit-reversed CRC computation.
If CRC optabs are supported, they are used for the CRC computation.
Otherwise, table-based CRC is generated.
The supported data and CRC sizes are 8, 16, 32, and 64 bits.
The polynomial is without the leading 1.
A table with 256 elements is used to store precomputed CRCs.
For the reflection of inputs and the output, a simple algorithm involving
SHIFT, AND, and OR operations is used.
Co-authored-by: Joern Rennecke <joern.rennecke@embecosm.com>
gcc/
* doc/md.texi (crc@var{m}@var{n}4,
crc_rev@var{m}@var{n}4): Document.
* expr.cc (calculate_crc): New function.
(assemble_crc_table): Likewise.
(generate_crc_table): Likewise.
(calculate_table_based_CRC): Likewise.
(emit_crc): Likewise.
(expand_crc_table_based): Likewise.
(gen_common_operation_to_reflect): Likewise.
(reflect_64_bit_value): Likewise.
(reflect_32_bit_value): Likewise.
(reflect_16_bit_value): Likewise.
(reflect_8_bit_value): Likewise.
(generate_reflecting_code_standard): Likewise.
(expand_reversed_crc_table_based): Likewise.
* expr.h (generate_reflecting_code_standard): New function
declaration.
(expand_crc_table_based): Likewise.
(expand_reversed_crc_table_based): Likewise.
* internal-fn.cc: (crc_direct): Define.
(direct_crc_optab_supported_p): Likewise.
(expand_crc_optab_fn): New function
* internal-fn.def (CRC, CRC_REV): New internal functions.
* optabs.def (crc_optab, crc_rev_optab): New optabs.
Signed-off-by: Mariam Arutunian <mariamarutunian@gmail.com>
[-- Attachment #2: 0001-Implement-internal-functions-for-efficient-CRC-compu.patch --]
[-- Type: text/x-patch, Size: 19069 bytes --]
diff --git a/gcc/doc/md.texi b/gcc/doc/md.texi
index 5730bda80dc..be68ef860f9 100644
--- a/gcc/doc/md.texi
+++ b/gcc/doc/md.texi
@@ -8557,6 +8557,20 @@ operand 2, greater than operand 2 or is unordered with operand 2.
This pattern is not allowed to @code{FAIL}.
+@cindex @code{crc@var{m}@var{n}4} instruction pattern
+@item @samp{crc@var{m}@var{n}4}
+Calculate a bit-forward CRC using operands 1, 2 and 3,
+then store the result in operand 0.
+Operands 1 is the initial CRC, operands 2 is the data and operands 3 is the
+polynomial without leading 1.
+Operands 0, 1 and 3 have mode @var{n} and operand 2 has mode @var{m}, where
+both modes are integers. The size of CRC to be calculated is determined by the
+mode; for example, if @var{n} is 'hi', a CRC16 is calculated.
+
+@cindex @code{crc_rev@var{m}@var{n}4} instruction pattern
+@item @samp{crc_rev@var{m}@var{n}4}
+Similar to @samp{crc@var{m}@var{n}4}, but calculates a bit-reversed CRC.
+
@end table
@end ifset
diff --git a/gcc/expr.cc b/gcc/expr.cc
index 1baa39b98eb..c9a049aeecc 100644
--- a/gcc/expr.cc
+++ b/gcc/expr.cc
@@ -14091,3 +14091,374 @@ int_expr_size (const_tree exp)
return tree_to_shwi (size);
}
+
+/* Calculate CRC for the initial CRC and given POLYNOMIAL.
+ CRC_BITS is CRC size. */
+
+static unsigned HOST_WIDE_INT
+calculate_crc (unsigned HOST_WIDE_INT crc,
+ unsigned HOST_WIDE_INT polynomial,
+ unsigned short crc_bits)
+{
+ unsigned HOST_WIDE_INT msb = HOST_WIDE_INT_1U << (crc_bits - 1);
+ crc = crc << (crc_bits - 8);
+ for (short i = 8; i > 0; --i)
+ {
+ if (crc & msb)
+ crc = (crc << 1) ^ polynomial;
+ else
+ crc <<= 1;
+ }
+ /* Zero out bits in crc beyond the specified number of crc_bits. */
+ if (crc_bits < sizeof (crc) * CHAR_BIT)
+ crc &= (HOST_WIDE_INT_1U << crc_bits) - 1;
+ return crc;
+}
+
+/* Assemble CRC table with 256 elements for the given POLYNOM and CRC_BITS with
+ given ID.
+ ID is the identifier of the table, the name of the table is unique,
+ contains CRC size and the polynomial.
+ POLYNOM is the polynomial used to calculate the CRC table's elements.
+ CRC_BITS is the size of CRC, may be 8, 16, ... . */
+
+rtx
+assemble_crc_table (tree id, unsigned HOST_WIDE_INT polynom,
+ unsigned short crc_bits)
+{
+ unsigned table_el_n = 0x100;
+ tree ar = build_array_type (make_unsigned_type (crc_bits),
+ build_index_type (size_int (table_el_n - 1)));
+ tree decl = build_decl (UNKNOWN_LOCATION, VAR_DECL, id, ar);
+ SET_DECL_ASSEMBLER_NAME (decl, id);
+ DECL_ARTIFICIAL (decl) = 1;
+ rtx tab = gen_rtx_SYMBOL_REF (Pmode, IDENTIFIER_POINTER (id));
+ TREE_ASM_WRITTEN (decl) = 0;
+
+ /* Initialize the table. */
+ vec<tree, va_gc> *initial_values;
+ vec_alloc (initial_values, table_el_n);
+ for (size_t i = 0; i < table_el_n; ++i)
+ {
+ unsigned HOST_WIDE_INT crc = calculate_crc (i, polynom, crc_bits);
+ tree element = build_int_cstu (make_unsigned_type (crc_bits), crc);
+ vec_safe_push (initial_values, element);
+ }
+ DECL_INITIAL (decl) = build_constructor_from_vec (ar, initial_values);
+
+ TREE_READONLY (decl) = 1;
+ TREE_STATIC (decl) = 1;
+
+ if (TREE_PUBLIC (id))
+ {
+ TREE_PUBLIC (decl) = 1;
+ make_decl_one_only (decl, DECL_ASSEMBLER_NAME (decl));
+ }
+
+ mark_decl_referenced (decl);
+ varpool_node::finalize_decl (decl);
+
+ return tab;
+}
+
+/* Generate CRC lookup table by calculating CRC for all possible
+ 8-bit data values. The table is stored with a specific name in the read-only
+ static data section.
+ POLYNOM is the polynomial used to calculate the CRC table's elements.
+ CRC_BITS is the size of CRC, may be 8, 16, ... . */
+
+rtx
+generate_crc_table (unsigned HOST_WIDE_INT polynom, unsigned short crc_bits)
+{
+ gcc_assert (crc_bits <= 64);
+
+ /* Buf size - 24 letters + 6 '_'
+ + 20 numbers (2 for crc bit size + 2 for 0x + 16 for 64-bit polynomial)
+ + 1 for \0. */
+ char buf[51];
+ sprintf (buf, "crc_table_for_crc_%u_polynomial_" HOST_WIDE_INT_PRINT_HEX,
+ crc_bits, polynom);
+
+ tree id = maybe_get_identifier (buf);
+ if (id)
+ return gen_rtx_SYMBOL_REF (Pmode, IDENTIFIER_POINTER (id));
+
+ id = get_identifier (buf);
+ return assemble_crc_table (id, polynom, crc_bits);
+}
+
+/* Generate table-based CRC code for the given CRC, INPUT_DATA and the
+ POLYNOMIAL (without leading 1).
+
+ First, using POLYNOMIAL's value generates CRC table of 256 elements,
+ then generates the assembly for the following code,
+ where crc_bit_size and data_bit_size may be 8, 16, 32, 64, depending on CRC:
+
+ for (int i = 0; i < data_bit_size / 8; i++)
+ crc = (crc << 8) ^ crc_table[(crc >> (crc_bit_size - 8))
+ ^ (data >> (data_bit_size - (i + 1) * 8)
+ & 0xFF))];
+
+ So to take values from the table, we need 8-bit data.
+ If input data size is not 8, then first we extract upper 8 bits,
+ then the other 8 bits, and so on. */
+
+void
+calculate_table_based_CRC (rtx *crc, const rtx &input_data,
+ const rtx &polynomial,
+ machine_mode crc_mode, machine_mode data_mode)
+{
+ unsigned short crc_bit_size = GET_MODE_BITSIZE (crc_mode).to_constant ();
+ unsigned short data_size = GET_MODE_SIZE (data_mode).to_constant ();
+
+ rtx tab = generate_crc_table (UINTVAL (polynomial), crc_bit_size);
+
+ for (unsigned short i = 0; i < data_size; i++)
+ {
+ /* crc >> (crc_bit_size - 8). */
+ rtx op1 = expand_shift (RSHIFT_EXPR, word_mode, *crc, crc_bit_size - 8,
+ NULL_RTX, 1);
+
+ /* data >> (8 * (GET_MODE_SIZE (data_mode).to_constant () - i - 1)). */
+ unsigned range_8 = 8 * (data_size - i - 1);
+ rtx data = expand_shift (RSHIFT_EXPR, word_mode, input_data, range_8,
+ NULL_RTX, 1);
+
+ /* data >> (8 * (GET_MODE_SIZE (data_mode)
+ .to_constant () - i - 1)) & 0xFF. */
+ rtx data_final = expand_and (word_mode, data,
+ gen_int_mode (255, data_mode), NULL_RTX);
+
+ /* (crc >> (crc_bit_size - 8)) ^ data_8bit. */
+ rtx in = expand_binop (Pmode, xor_optab, op1, data_final,
+ NULL_RTX, 1, OPTAB_WIDEN);
+
+ /* ((crc >> (crc_bit_size - 8)) ^ data_8bit) & 0xFF. */
+ rtx index = expand_and (Pmode, in, gen_int_mode (255, word_mode),
+ NULL_RTX);
+ int log_crc_size = exact_log2 (GET_MODE_SIZE (crc_mode).to_constant ());
+ index = expand_shift (LSHIFT_EXPR, Pmode, index,
+ log_crc_size, NULL_RTX, 0);
+
+ index = expand_binop (Pmode, add_optab, index, tab, NULL_RTX,
+ 0, OPTAB_DIRECT);
+
+ /* crc_table[(crc >> (crc_bit_size - 8)) ^ data_8bit] */
+ rtx tab_el = validize_mem (gen_rtx_MEM (crc_mode, index));
+
+ /* (crc << 8) if CRC is larger than 8, otherwise crc = 0. */
+ rtx high = NULL_RTX;
+ if (crc_bit_size != 8)
+ {
+ high = expand_shift (LSHIFT_EXPR, word_mode, *crc, 8, NULL_RTX, 0);
+ if (crc_mode != word_mode)
+ {
+ rtx crc_mode_mask = gen_int_mode (GET_MODE_MASK (crc_mode),
+ word_mode);
+ high = expand_and (word_mode, high, crc_mode_mask, NULL_RTX);
+ }
+ }
+ else
+ high = gen_int_mode (0, word_mode);
+
+ /* crc = (crc << 8)
+ ^ crc_table[(crc >> (crc_bit_size - 8)) ^ data_8bit]; */
+ *crc = expand_binop (word_mode, xor_optab, tab_el, high, NULL_RTX, 1,
+ OPTAB_WIDEN);
+ }
+}
+
+/* Converts and moves a CRC value to a target register.
+
+ CRC_MODE is the mode (data type) of the CRC value.
+ CRC is the initial CRC value.
+ OP0 is the target register. */
+
+void
+emit_crc (machine_mode crc_mode, rtx* crc, rtx* op0)
+{
+ if (GET_MODE_BITSIZE (crc_mode).to_constant () == 32
+ && GET_MODE_BITSIZE (word_mode) == 64)
+ {
+ rtx a_low = gen_lowpart_SUBREG (crc_mode, *crc);
+ *crc = gen_rtx_SIGN_EXTEND (word_mode, a_low);
+ }
+ rtx tgt = *op0;
+ if (word_mode != crc_mode)
+ tgt = simplify_gen_subreg (word_mode, *op0, crc_mode, 0);
+ emit_move_insn (tgt, *crc);
+}
+
+/* Generate table-based CRC code for the given CRC, INPUT_DATA and the
+ POLYNOMIAL (without leading 1).
+
+ CRC is OP1, data is OP2 and the polynomial is OP3.
+ This must generate a CRC table and an assembly for the following code,
+ where crc_bit_size and data_bit_size may be 8, 16, 32, 64:
+ uint_crc_bit_size_t
+ crc_crc_bit_size (uint_crc_bit_size_t crc_init,
+ uint_data_bit_size_t data, size_t size)
+ {
+ uint_crc_bit_size_t crc = crc_init;
+ for (int i = 0; i < data_bit_size / 8; i++)
+ crc = (crc << 8) ^ crc_table[(crc >> (crc_bit_size - 8))
+ ^ (data >> (data_bit_size - (i + 1) * 8)
+ & 0xFF))];
+ return crc;
+ } */
+
+void
+expand_crc_table_based (rtx op0, rtx op1, rtx op2, rtx op3,
+ machine_mode data_mode)
+{
+ gcc_assert (!CONST_INT_P (op0));
+ gcc_assert (CONST_INT_P (op3));
+ machine_mode crc_mode = GET_MODE (op0);
+ rtx crc = gen_reg_rtx (word_mode);
+ convert_move (crc, op1, 0);
+ calculate_table_based_CRC (&crc, op2, op3, crc_mode, data_mode);
+ emit_crc (crc_mode, &crc, &op0);
+}
+
+/* Generate the common operation for reflecting values:
+ *OP = (*OP & AND1_VALUE) << SHIFT_VAL | (*OP & AND2_VALUE) >> SHIFT_VAL; */
+
+void
+gen_common_operation_to_reflect (rtx *op,
+ unsigned HOST_WIDE_INT and1_value,
+ unsigned HOST_WIDE_INT and2_value,
+ unsigned shift_val)
+{
+ rtx op1 = expand_and (word_mode, *op, gen_int_mode (and1_value, word_mode),
+ NULL_RTX);
+ op1 = expand_shift (LSHIFT_EXPR, word_mode, op1, shift_val, op1, 0);
+ rtx op2 = expand_and (word_mode, *op, gen_int_mode (and2_value, word_mode),
+ NULL_RTX);
+ op2 = expand_shift (RSHIFT_EXPR, word_mode, op2, shift_val, op2, 1);
+ *op = expand_binop (word_mode, ior_optab, op1, op2, *op, 0, OPTAB_DIRECT);
+}
+
+/* Reflect 64-bit value for the 64-bit target. */
+
+void
+reflect_64_bit_value (rtx *op)
+{
+ gen_common_operation_to_reflect (op, 0x00000000FFFFFFFF,
+ 0xFFFFFFFF00000000, 32);
+ gen_common_operation_to_reflect (op, 0x0000FFFF0000FFFF,
+ 0xFFFF0000FFFF0000, 16);
+ gen_common_operation_to_reflect (op, 0x00FF00FF00FF00FF,
+ 0xFF00FF00FF00FF00, 8);
+ gen_common_operation_to_reflect (op, 0x0F0F0F0F0F0F0F0F,
+ 0xF0F0F0F0F0F0F0F0, 4);
+ gen_common_operation_to_reflect (op, 0x3333333333333333,
+ 0xCCCCCCCCCCCCCCCC, 2);
+ gen_common_operation_to_reflect (op, 0x5555555555555555,
+ 0xAAAAAAAAAAAAAAAA, 1);
+}
+
+/* Reflect 32-bit value for the 32-bit target. */
+
+void
+reflect_32_bit_value (rtx *op)
+{
+ gen_common_operation_to_reflect (op, 0x0000FFFF, 0xFFFF0000, 16);
+ gen_common_operation_to_reflect (op, 0x00FF00FF, 0xFF00FF00, 8);
+ gen_common_operation_to_reflect (op, 0x0F0F0F0F, 0xF0F0F0F0, 4);
+ gen_common_operation_to_reflect (op, 0x33333333, 0xCCCCCCCC, 2);
+ gen_common_operation_to_reflect (op, 0x55555555, 0xAAAAAAAA, 1);
+}
+
+/* Reflect 16-bit value for the 16-bit target. */
+
+void
+reflect_16_bit_value (rtx *op)
+{
+ gen_common_operation_to_reflect (op, 0x00FF, 0xFF00, 8);
+ gen_common_operation_to_reflect (op, 0x0F0F, 0xF0F0, 4);
+ gen_common_operation_to_reflect (op, 0x3333, 0xCCCC, 2);
+ gen_common_operation_to_reflect (op, 0x5555, 0xAAAA, 1);
+}
+
+/* Reflect 8-bit value for the 8-bit target. */
+
+void
+reflect_8_bit_value (rtx *op)
+{
+ gen_common_operation_to_reflect (op, 0x0F, 0xF0, 4);
+ gen_common_operation_to_reflect (op, 0x33, 0xCC, 2);
+ gen_common_operation_to_reflect (op, 0x55, 0xAA, 1);
+}
+
+/* Generate instruction sequence
+ which reflects the value of the OP using shift, and, or operations.
+ OP's mode may be less than word_mode. To get the correct number,
+ after reflecting we shift right the value by SHIFT_VAL.
+ E.g. we have 1111 0001, after reflection (target 32-bit) we will get
+ 1000 1111 0000 0000, if we shift-out 16 bits,
+ we will get the desired one: 1000 1111. */
+
+void
+generate_reflecting_code_standard (rtx *op, int shift_val)
+{
+ gcc_assert (BITS_PER_WORD >= 8 && BITS_PER_WORD <= 64);
+
+ if (BITS_PER_WORD == 64)
+ reflect_64_bit_value (op);
+ else if (BITS_PER_WORD == 32)
+ reflect_32_bit_value (op);
+ else if (BITS_PER_WORD == 16)
+ reflect_16_bit_value (op);
+ else
+ reflect_8_bit_value (op);
+
+ *op = expand_shift (RSHIFT_EXPR, word_mode, *op, shift_val, *op, 1);
+}
+
+/* Generate table-based reversed CRC code for the given CRC, INPUT_DATA and
+ the POLYNOMIAL (without leading 1).
+
+ CRC is OP1, data is OP2 and the polynomial is OP3.
+ This must generate CRC table and assembly for the following code,
+ where crc_bit_size and data_bit_size may be 8, 16, 32, 64:
+ uint_crc_bit_size_t
+ crc_crc_bit_size (uint_crc_bit_size_t crc_init,
+ uint_data_bit_size_t data, size_t size)
+ {
+ reflect (crc_init)
+ uint_crc_bit_size_t crc = crc_init;
+ reflect (data);
+ for (int i = 0; i < data_bit_size / 8; i++)
+ crc = (crc << 8) ^ crc_table[(crc >> (crc_bit_size - 8))
+ ^ (data >> (data_bit_size - (i + 1) * 8) & 0xFF))];
+ reflect (crc);
+ return crc;
+ } */
+
+void
+expand_reversed_crc_table_based (rtx op0, rtx op1, rtx op2, rtx op3,
+ machine_mode data_mode,
+ void (*gen_reflecting_code) (rtx *op,
+ int shift_val))
+{
+ gcc_assert (!CONST_INT_P (op0));
+ gcc_assert (CONST_INT_P (op3));
+ machine_mode crc_mode = GET_MODE (op0);
+
+ unsigned short crc_bit_size = GET_MODE_BITSIZE (crc_mode).to_constant ();
+ unsigned short data_bit_size = GET_MODE_BITSIZE (data_mode).to_constant ();
+ unsigned short word_size = GET_MODE_BITSIZE (word_mode);
+
+ rtx crc = gen_reg_rtx (word_mode);
+ convert_move (crc, op1, 0);
+ gen_reflecting_code (&crc, word_size - crc_bit_size);
+
+ rtx data = gen_reg_rtx (word_mode);
+ convert_move (data, op2, 0);
+ gen_reflecting_code (&data, word_size - data_bit_size);
+
+ calculate_table_based_CRC (&crc, data, op3, crc_mode, data_mode);
+
+ gen_reflecting_code (&crc, word_size - crc_bit_size);
+ emit_crc (crc_mode, &crc, &op0);
+}
diff --git a/gcc/expr.h b/gcc/expr.h
index 75181584108..74634d22777 100644
--- a/gcc/expr.h
+++ b/gcc/expr.h
@@ -374,4 +374,10 @@ extern rtx expr_size (tree);
extern bool mem_ref_refers_to_non_mem_p (tree);
extern bool non_mem_decl_p (tree);
+/* Generate table-based CRC. */
+extern void generate_reflecting_code_standard (rtx *, int);
+extern void expand_crc_table_based (rtx, rtx, rtx, rtx, machine_mode);
+extern void expand_reversed_crc_table_based (rtx, rtx, rtx, rtx, machine_mode,
+ void (*) (rtx *, int));
+
#endif /* GCC_EXPR_H */
diff --git a/gcc/internal-fn.cc b/gcc/internal-fn.cc
index 9c09026793f..598b6fd1816 100644
--- a/gcc/internal-fn.cc
+++ b/gcc/internal-fn.cc
@@ -189,6 +189,7 @@ init_internal_fns ()
#define mask_fold_left_direct { 1, 1, false }
#define mask_len_fold_left_direct { 1, 1, false }
#define check_ptrs_direct { 0, 0, false }
+#define crc_direct { 1, -1, true }
const direct_internal_fn_info direct_internal_fn_array[IFN_LAST + 1] = {
#define DEF_INTERNAL_FN(CODE, FLAGS, FNSPEC) not_direct,
@@ -3918,6 +3919,58 @@ expand_convert_optab_fn (internal_fn fn, gcall *stmt, convert_optab optab,
expand_fn_using_insn (stmt, icode, 1, nargs);
}
+/* Expand CRC call STMT. */
+
+static void
+expand_crc_optab_fn (internal_fn fn, gcall *stmt, convert_optab optab)
+{
+ tree lhs = gimple_call_lhs (stmt);
+ tree rhs1 = gimple_call_arg (stmt, 0); // crc
+ tree rhs2 = gimple_call_arg (stmt, 1); // data
+ tree rhs3 = gimple_call_arg (stmt, 2); // polynomial
+
+ tree result_type = TREE_TYPE (lhs);
+ tree data_type = TREE_TYPE (rhs2);
+
+ gcc_assert (TYPE_MODE (result_type) >= TYPE_MODE (data_type));
+ gcc_assert (word_mode >= TYPE_MODE (result_type));
+
+ rtx dest = expand_expr (lhs, NULL_RTX, VOIDmode, EXPAND_WRITE);
+ rtx crc = expand_normal (rhs1);
+ rtx data = expand_normal (rhs2);
+ gcc_assert (TREE_CODE (rhs3) == INTEGER_CST);
+ rtx polynomial = gen_rtx_CONST_INT (TYPE_MODE (result_type),
+ TREE_INT_CST_LOW (rhs3));
+
+ /* Use target specific expansion if it exists.
+ Otherwise, generate table-based CRC. */
+ if (direct_internal_fn_supported_p (fn, tree_pair (data_type, result_type),
+ OPTIMIZE_FOR_SPEED))
+ {
+ class expand_operand ops[4];
+ create_call_lhs_operand (&ops[0], dest, TYPE_MODE (result_type));
+ create_input_operand (&ops[1], crc, TYPE_MODE (result_type));
+ create_input_operand (&ops[2], data, TYPE_MODE (data_type));
+ create_input_operand (&ops[3], polynomial, TYPE_MODE (result_type));
+ insn_code icode = convert_optab_handler (optab, TYPE_MODE (data_type),
+ TYPE_MODE (result_type));
+ expand_insn (icode, 4, ops);
+ assign_call_lhs (lhs, dest, &ops[0]);
+ }
+ else
+ {
+ /* If it's IFN_CRC generate bit-forward CRC. */
+ if (fn == IFN_CRC)
+ expand_crc_table_based (dest, crc, data, polynomial,
+ TYPE_MODE (data_type));
+ else
+ /* If it's IFN_CRC_REV generate bit-reversed CRC. */
+ expand_reversed_crc_table_based (dest, crc, data, polynomial,
+ TYPE_MODE (data_type),
+ generate_reflecting_code_standard);
+ }
+}
+
/* Expanders for optabs that can use expand_direct_optab_fn. */
#define expand_unary_optab_fn(FN, STMT, OPTAB) \
@@ -4054,6 +4107,7 @@ multi_vector_optab_supported_p (convert_optab optab, tree_pair types,
#define direct_cond_len_unary_optab_supported_p direct_optab_supported_p
#define direct_cond_len_binary_optab_supported_p direct_optab_supported_p
#define direct_cond_len_ternary_optab_supported_p direct_optab_supported_p
+#define direct_crc_optab_supported_p convert_optab_supported_p
#define direct_mask_load_optab_supported_p convert_optab_supported_p
#define direct_load_lanes_optab_supported_p multi_vector_optab_supported_p
#define direct_mask_load_lanes_optab_supported_p multi_vector_optab_supported_p
diff --git a/gcc/internal-fn.def b/gcc/internal-fn.def
index 25badbb86e5..76585100a63 100644
--- a/gcc/internal-fn.def
+++ b/gcc/internal-fn.def
@@ -200,6 +200,8 @@ along with GCC; see the file COPYING3. If not see
cond_len_##UNSIGNED_OPTAB, cond_len_##TYPE)
#endif
+DEF_INTERNAL_OPTAB_FN (CRC, ECF_CONST | ECF_NOTHROW, crc, crc)
+DEF_INTERNAL_OPTAB_FN (CRC_REV, ECF_CONST | ECF_NOTHROW, crc_rev, crc)
DEF_INTERNAL_OPTAB_FN (MASK_LOAD, ECF_PURE, maskload, mask_load)
DEF_INTERNAL_OPTAB_FN (LOAD_LANES, ECF_CONST, vec_load_lanes, load_lanes)
DEF_INTERNAL_OPTAB_FN (MASK_LOAD_LANES, ECF_PURE,
diff --git a/gcc/optabs.def b/gcc/optabs.def
index 3f2cb46aff8..bf1aaac90c6 100644
--- a/gcc/optabs.def
+++ b/gcc/optabs.def
@@ -78,6 +78,8 @@ OPTAB_CD(smsub_widen_optab, "msub$b$a4")
OPTAB_CD(umsub_widen_optab, "umsub$b$a4")
OPTAB_CD(ssmsub_widen_optab, "ssmsub$b$a4")
OPTAB_CD(usmsub_widen_optab, "usmsub$a$b4")
+OPTAB_CD(crc_optab, "crc$a$b4")
+OPTAB_CD(crc_rev_optab, "crc_rev$a$b4")
OPTAB_CD(vec_load_lanes_optab, "vec_load_lanes$a$b")
OPTAB_CD(vec_store_lanes_optab, "vec_store_lanes$a$b")
OPTAB_CD(vec_mask_load_lanes_optab, "vec_mask_load_lanes$a$b")
--
2.25.1
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: [RFC/RFA][PATCH v2 01/12] Implement internal functions for efficient CRC computation
2024-07-26 18:05 [RFC/RFA][PATCH v2 01/12] Implement internal functions for efficient CRC computation Mariam Arutunian
@ 2024-08-25 17:51 ` Jeff Law
0 siblings, 0 replies; 2+ messages in thread
From: Jeff Law @ 2024-08-25 17:51 UTC (permalink / raw)
To: Mariam Arutunian, GCC Patches
On 7/26/24 12:05 PM, Mariam Arutunian wrote:
> Add two new internal functions (IFN_CRC, IFN_CRC_REV), to provide
> faster CRC generation.
> One performs bit-forward and the other bit-reversed CRC computation.
> If CRC optabs are supported, they are used for the CRC computation.
> Otherwise, table-based CRC is generated.
> The supported data and CRC sizes are 8, 16, 32, and 64 bits.
> The polynomial is without the leading 1.
> A table with 256 elements is used to store precomputed CRCs.
> For the reflection of inputs and the output, a simple algorithm
> involving
> SHIFT, AND, and OR operations is used.
>
> Co-authored-by: Joern Rennecke <joern.rennecke@embecosm.com
> <mailto:joern.rennecke@embecosm.com>>
>
> gcc/
>
> * doc/md.texi (crc@var{m}@var{n}4,
> crc_rev@var{m}@var{n}4): Document.
> * expr.cc (calculate_crc): New function.
> (assemble_crc_table): Likewise.
> (generate_crc_table): Likewise.
> (calculate_table_based_CRC): Likewise.
> (emit_crc): Likewise.
> (expand_crc_table_based): Likewise.
> (gen_common_operation_to_reflect): Likewise.
> (reflect_64_bit_value): Likewise.
> (reflect_32_bit_value): Likewise.
> (reflect_16_bit_value): Likewise.
> (reflect_8_bit_value): Likewise.
> (generate_reflecting_code_standard): Likewise.
> (expand_reversed_crc_table_based): Likewise.
> * expr.h (generate_reflecting_code_standard): New function
> declaration.
> (expand_crc_table_based): Likewise.
> (expand_reversed_crc_table_based): Likewise.
> * internal-fn.cc: (crc_direct): Define.
> (direct_crc_optab_supported_p): Likewise.
> (expand_crc_optab_fn): New function
> * internal-fn.def (CRC, CRC_REV): New internal functions.
> * optabs.def (crc_optab, crc_rev_optab): New optabs.
>
> Signed-off-by: Mariam Arutunian <mariamarutunian@gmail.com
> <mailto:mariamarutunian@gmail.com>>
>
> 0001-Implement-internal-functions-for-efficient-CRC-compu.patch
>
> diff --git a/gcc/expr.cc b/gcc/expr.cc
> index 1baa39b98eb..c9a049aeecc 100644
> --- a/gcc/expr.cc
> +++ b/gcc/expr.cc
> +
> +/* Converts and moves a CRC value to a target register.
> +
> + CRC_MODE is the mode (data type) of the CRC value.
> + CRC is the initial CRC value.
> + OP0 is the target register. */
> +
> +void
> +emit_crc (machine_mode crc_mode, rtx* crc, rtx* op0)
> +{
> + if (GET_MODE_BITSIZE (crc_mode).to_constant () == 32
> + && GET_MODE_BITSIZE (word_mode) == 64)
> + {
> + rtx a_low = gen_lowpart_SUBREG (crc_mode, *crc);
I may have asked this before, but is there a reason we're not just using
gen_lowpart?
Otherwise this looks pretty reasonable.
Just a note. IIRC, you were having some problems with formatting
issues. Rather than forcing you to fight with those, I'm comfortable if
you send me the final patchkit and I'll do a once-over for formatting
and commit the final result.
Jeff
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2024-08-25 17:51 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2024-07-26 18:05 [RFC/RFA][PATCH v2 01/12] Implement internal functions for efficient CRC computation Mariam Arutunian
2024-08-25 17:51 ` Jeff Law
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).