aboutsummaryrefslogtreecommitdiff
diff options
context:
space:
mode:
Diffstat
-rw-r--r--gcc/config/riscv/autovec-opt.md41
-rw-r--r--gcc/gimple-iterator.cc14
-rw-r--r--gcc/match.pd44
-rw-r--r--gcc/testsuite/g++.dg/torture/pr122589.C22
-rw-r--r--gcc/testsuite/gcc.dg/vect/vect-bool-3.c13
-rw-r--r--gcc/testsuite/gcc.target/riscv/rvv/autovec/pr121959-1.c9
-rw-r--r--gcc/testsuite/gcc.target/riscv/rvv/autovec/pr121959-2.c9
-rw-r--r--gcc/testsuite/gcc.target/riscv/rvv/autovec/pr121959-3.c9
-rw-r--r--gcc/testsuite/gcc.target/riscv/rvv/autovec/pr121959-4.c9
-rw-r--r--gcc/testsuite/gcc.target/riscv/rvv/autovec/pr121959-5.c9
-rw-r--r--gcc/testsuite/gcc.target/riscv/rvv/autovec/pr121959-run-1.c65
-rw-r--r--gcc/testsuite/gcc.target/riscv/rvv/autovec/pr121959.h24
-rw-r--r--gcc/tree-vect-stmts.cc9
13 files changed, 228 insertions, 49 deletions
diff --git a/gcc/config/riscv/autovec-opt.md b/gcc/config/riscv/autovec-opt.md
index 063c9a0..52ab79c 100644
--- a/gcc/config/riscv/autovec-opt.md
+++ b/gcc/config/riscv/autovec-opt.md
@@ -2424,3 +2424,44 @@
}
[(set_attr "type" "vfalu")]
)
+
+;; Combine vsext.vf + vsll.vi to vwsll.vi depends on ZVBB.
+;; The vwsll.vi is zero extend, thus only the ashift bits
+;; is equal or greater than double truncated bits is valid.
+;; Appears in the satd function of x264.
+(define_insn_and_split "*vwsll_sign_extend_<mode>"
+ [(set (match_operand:VWEXTI 0 "register_operand")
+ (ashift:VWEXTI
+ (sign_extend:VWEXTI
+ (match_operand:<V_DOUBLE_TRUNC> 1 "register_operand"))
+ (match_operand 2 "const_int_operand")))]
+ "TARGET_VECTOR && TARGET_ZVBB && can_create_pseudo_p ()"
+ "#"
+ "&& 1"
+ [(const_int 0)]
+ {
+ int imm = INTVAL (operands[2]);
+ int trunc_prec = GET_MODE_PRECISION (GET_MODE_INNER (<V_DOUBLE_TRUNC>mode));
+
+ if (imm >= trunc_prec)
+ {
+ insn_code icode = code_for_pred_vwsll_scalar (<MODE>mode);
+ emit_vlmax_insn (icode, riscv_vector::BINARY_OP, operands);
+ }
+ else
+ {
+ insn_code icode = code_for_pred_vf2 (SIGN_EXTEND, <MODE>mode);
+ rtx extend = gen_reg_rtx (<MODE>mode);
+ rtx unary_ops[] = {extend, operands[1]};
+ riscv_vector::emit_vlmax_insn (icode, riscv_vector::UNARY_OP,
+ unary_ops);
+
+ icode = code_for_pred_scalar (ASHIFT, <MODE>mode);
+ rtx binary_ops[] = {operands[0], extend, operands[2]};
+ riscv_vector::emit_vlmax_insn (icode, riscv_vector::BINARY_OP,
+ binary_ops);
+ }
+
+ DONE;
+ }
+)
diff --git a/gcc/gimple-iterator.cc b/gcc/gimple-iterator.cc
index 3af672b..741e5a5 100644
--- a/gcc/gimple-iterator.cc
+++ b/gcc/gimple-iterator.cc
@@ -389,8 +389,8 @@ gsi_set_stmt (gimple_stmt_iterator *gsi, gimple *stmt)
}
-/* Move all statements in the sequence before I to a new sequence.
- Return this new sequence. I is set to the head of the new list. */
+/* Move all statements in the sequence starting at I to a new sequence.
+ Set *PNEW_SEQ to this sequence. I is set to the head of the new list. */
void
gsi_split_seq_before (gimple_stmt_iterator *i, gimple_seq *pnew_seq)
@@ -471,18 +471,16 @@ void
gsi_replace_with_seq (gimple_stmt_iterator *gsi, gimple_seq seq,
bool update_eh_info)
{
- gimple_stmt_iterator seqi;
- gimple *last;
if (gimple_seq_empty_p (seq))
{
gsi_remove (gsi, true);
return;
}
- seqi = gsi_last (seq);
- last = gsi_stmt (seqi);
- gsi_remove (&seqi, false);
+ gimple_seq tail;
+ gimple_stmt_iterator lasti = gsi_last (seq);
+ gsi_split_seq_before (&lasti, &tail);
gsi_insert_seq_before (gsi, seq, GSI_SAME_STMT);
- gsi_replace (gsi, last, update_eh_info);
+ gsi_replace (gsi, gsi_stmt (lasti), update_eh_info);
}
diff --git a/gcc/match.pd b/gcc/match.pd
index 0ea86d9..3cd9ab1 100644
--- a/gcc/match.pd
+++ b/gcc/match.pd
@@ -3726,53 +3726,27 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
bool c2_is_type_precision_p = c2 == prec;
}
(if (widen_prec > prec && c2_is_type_precision_p && c4_is_max_p)))))
- (for mult_op (mult widen_mult)
- (match (unsigned_integer_sat_mul @0 @1)
- /* SAT_U_MUL (X, Y) = {
- WT x = (WT)a * (WT)b;
- NT hi = x >> (sizeof(NT) * 8);
- NT lo = (NT)x;
- return lo | -!!hi;
- } while WT is uint128_t, uint64_t, uint32_t, uint16_t,
- and T is uint64_t, uint32_t, uint16_t, uint8_t. */
- (convert1?
- (bit_ior
- (convert?
- (negate
- (convert (ne (convert2? (rshift @3 INTEGER_CST@2)) integer_zerop))))
- (convert (mult_op:c@3 (convert@4 @0) (convert@5 @1)))))
- (if (types_match (type, @0, @1))
- (with
- {
- unsigned prec = TYPE_PRECISION (type);
- unsigned widen_prec = TYPE_PRECISION (TREE_TYPE (@3));
- unsigned cvt4_prec = TYPE_PRECISION (TREE_TYPE (@4));
- unsigned cvt5_prec = TYPE_PRECISION (TREE_TYPE (@5));
-
- bool widen_mult_p = mult_op == WIDEN_MULT_EXPR && cvt4_prec == cvt5_prec
- && widen_prec == cvt5_prec * 2;
- bool mult_p = mult_op == MULT_EXPR && cvt4_prec == cvt5_prec
- && cvt4_prec == widen_prec && widen_prec > prec;
- bool c2_is_type_precision_p = tree_to_uhwi (@2) == prec;
- }
- (if (c2_is_type_precision_p && (mult_p || widen_mult_p)))))))
(match (unsigned_integer_sat_mul @0 @1)
+ /* SAT_U_MUL (X, Y) = {
+ WT x = (WT)a * (WT)b;
+ NT hi = x >> (sizeof(NT) * 8);
+ NT lo = (NT)x;
+ return lo | -!!hi;
+ } while WT is uint128_t, uint64_t, uint32_t, uint16_t,
+ and T is uint64_t, uint32_t, uint16_t, uint8_t. */
(convert1?
(bit_ior
(convert?
(negate
(convert (ne (convert2? (rshift @3 INTEGER_CST@2)) integer_zerop))))
- (convert (widen_mult:c@3 @0 @1))))
+ (convert (usmul_widen_mult@3 @0 @1))))
(if (types_match (type, @0, @1))
(with
{
unsigned prec = TYPE_PRECISION (type);
- unsigned widen_prec = TYPE_PRECISION (TREE_TYPE (@3));
-
bool c2_is_type_precision_p = tree_to_uhwi (@2) == prec;
- bool widen_mult_p = prec * 2 == widen_prec;
}
- (if (c2_is_type_precision_p && widen_mult_p)))))
+ (if (c2_is_type_precision_p)))))
)
/* The boundary condition for case 10: IMM = 1:
diff --git a/gcc/testsuite/g++.dg/torture/pr122589.C b/gcc/testsuite/g++.dg/torture/pr122589.C
new file mode 100644
index 0000000..ac441a0
--- /dev/null
+++ b/gcc/testsuite/g++.dg/torture/pr122589.C
@@ -0,0 +1,22 @@
+// { dg-do compile }
+// { dg-additional-options "-g" }
+
+struct QPointF {
+ QPointF(double xpos, double ypos) : xp(xpos), yp(ypos) {}
+ double xp;
+ double yp;
+};
+double xp, yp, w, h;
+struct QRectF {
+ QRectF(QPointF, int);
+ QPointF center() { return QPointF(xp + w / 2, yp + h / 2); }
+};
+void clientArea(QPointF &);
+int workspace_size;
+void workspace() {
+ QRectF geom(QPointF(0, 0), workspace_size);
+ xp = 0 - w / 2;
+ yp = -h;
+ QPointF __trans_tmp_2 = geom.center();
+ clientArea(__trans_tmp_2);
+}
diff --git a/gcc/testsuite/gcc.dg/vect/vect-bool-3.c b/gcc/testsuite/gcc.dg/vect/vect-bool-3.c
new file mode 100644
index 0000000..671f602
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/vect/vect-bool-3.c
@@ -0,0 +1,13 @@
+/* { dg-do compile } */
+/* { dg-require-effective-target vect_int } */
+/* { dg-require-effective-target vect_unpack } */
+
+int count_true(const bool *values, int len)
+{
+ int count = 0;
+ for (int i = 0; i < len; i++)
+ count += values[i];
+ return count;
+}
+
+/* { dg-final { scan-tree-dump "optimized: loop vectorized" "vect" } } */
diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr121959-1.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr121959-1.c
new file mode 100644
index 0000000..a42d7c4
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr121959-1.c
@@ -0,0 +1,9 @@
+/* { dg-do compile } */
+/* { dg-options "-march=rv64gcv_zvbb -mabi=lp64d -O3" } */
+
+#include "pr121959.h"
+
+DEF_VWSLL_FUNC_0(int32_t, uint8_t, 16)
+
+/* { dg-final { scan-assembler-times {vwsll.vi} 1 } } */
+/* { dg-final { scan-assembler-not {vsll.vi} } } */
diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr121959-2.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr121959-2.c
new file mode 100644
index 0000000..2a3ef8d
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr121959-2.c
@@ -0,0 +1,9 @@
+/* { dg-do compile } */
+/* { dg-options "-march=rv64gcv -mabi=lp64d -O3" } */
+
+#include "pr121959.h"
+
+DEF_VWSLL_FUNC_0(int32_t, uint8_t, 16)
+
+/* { dg-final { scan-assembler-times {vsll.vi} 1 } } */
+/* { dg-final { scan-assembler-not {vwsll.vi} } } */
diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr121959-3.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr121959-3.c
new file mode 100644
index 0000000..59a930a
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr121959-3.c
@@ -0,0 +1,9 @@
+/* { dg-do compile } */
+/* { dg-options "-march=rv64gcv_zvbb -mabi=lp64d -O3" } */
+
+#include "pr121959.h"
+
+DEF_VWSLL_FUNC_0(int32_t, uint8_t, 17)
+
+/* { dg-final { scan-assembler-times {vwsll.vi} 1 } } */
+/* { dg-final { scan-assembler-not {vsll.vi} } } */
diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr121959-4.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr121959-4.c
new file mode 100644
index 0000000..59a6d36
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr121959-4.c
@@ -0,0 +1,9 @@
+/* { dg-do compile } */
+/* { dg-options "-march=rv64gcv -mabi=lp64d -O3" } */
+
+#include "pr121959.h"
+
+DEF_VWSLL_FUNC_0(int32_t, uint8_t, 17)
+
+/* { dg-final { scan-assembler-times {vsll.vi} 1 } } */
+/* { dg-final { scan-assembler-not {vwsll.vi} } } */
diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr121959-5.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr121959-5.c
new file mode 100644
index 0000000..a9319a3
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr121959-5.c
@@ -0,0 +1,9 @@
+/* { dg-do compile } */
+/* { dg-options "-march=rv64gcv_zvbb -mabi=lp64d -O3" } */
+
+#include "pr121959.h"
+
+DEF_VWSLL_FUNC_0(int32_t, uint8_t, 15)
+
+/* { dg-final { scan-assembler-times {vsll.vi} 1 } } */
+/* { dg-final { scan-assembler-not {vwsll.vi} } } */
diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr121959-run-1.c b/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr121959-run-1.c
new file mode 100644
index 0000000..77fd95b
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr121959-run-1.c
@@ -0,0 +1,65 @@
+/* { dg-do run } */
+/* { dg-require-effective-target "riscv_zvbb_ok" } */
+/* { dg-add-options "riscv_v" } */
+/* { dg-add-options "riscv_zvbb" } */
+/* { dg-additional-options "-std=c99" } */
+
+#include "pr121959.h"
+
+#define WT int32_t
+#define NT uint8_t
+#define IMM 16
+#define N 16
+
+DEF_VWSLL_FUNC_0_WRAP(WT, NT, IMM)
+
+NT g_data[][2][N] = {
+ {
+ /* a */
+ {
+ 2, 2, 2, 1,
+ 255, 255, 255, 255,
+ 128, 128, 128, 128,
+ 127, 127, 127, 127,
+ },
+ /* b */
+ {
+ 1, 1, 1, 1,
+ 0, 0, 0, 0,
+ 2, 2, 2, 2,
+ 7, 7, 7, 7,
+ },
+ },
+};
+
+WT g_expect[][N] = {
+ /* 0 */
+ {
+ 65536, 65536, 65536, 65536,
+ 16711680, 16711680, 16711680, 16711680,
+ 8257536, 8257536, 8257536, 8257536,
+ 7864320, 7864320, 7864320, 7864320,
+ },
+};
+
+int
+main ()
+{
+ unsigned i, k;
+ WT out[N];
+
+ for (i = 0; i < sizeof (g_data) / sizeof (g_data[0]); i++)
+ {
+ NT *a = g_data[i][0];
+ NT *b = g_data[i][1];
+ WT *expect = g_expect[i];
+
+ RUN_VWSLL_FUNC_0_WRAP (WT, NT, IMM, out, a, b, N);
+
+ for (k = 0; k < N; k++)
+ if (out[k] != expect[k])
+ __builtin_abort ();
+ }
+
+ return 0;
+}
diff --git a/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr121959.h b/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr121959.h
new file mode 100644
index 0000000..10b1b62
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/rvv/autovec/pr121959.h
@@ -0,0 +1,24 @@
+#ifndef HAVE_DEFINED_PR121959_H
+#define HAVE_DEFINED_PR121959_H
+
+#include <stdint.h>
+
+#define DEF_VWSLL_FUNC_0(WT, NT, IMM) \
+void \
+test_from_##NT##_to_##WT##_##IMM##_0(WT * restrict res, \
+ NT * restrict a, \
+ NT * restrict b, \
+ int n) \
+{ \
+ for (int i = 0; i < n; i++) \
+ { \
+ res[i] = (a[i] - b[i]) << IMM; \
+ } \
+}
+#define DEF_VWSLL_FUNC_0_WRAP(WT, NT, IMM) DEF_VWSLL_FUNC_0(WT, NT, IMM)
+#define RUN_VWSLL_FUNC_0(WT, NT, IMM, res, a, b, n) \
+ test_from_##NT##_to_##WT##_##IMM##_0(res, a, b, n)
+#define RUN_VWSLL_FUNC_0_WRAP(WT, NT, IMM, res, a, b, n) \
+ RUN_VWSLL_FUNC_0(WT, NT, IMM, res, a, b, n)
+
+#endif
diff --git a/gcc/tree-vect-stmts.cc b/gcc/tree-vect-stmts.cc
index 83acbb3..8692d44 100644
--- a/gcc/tree-vect-stmts.cc
+++ b/gcc/tree-vect-stmts.cc
@@ -5282,15 +5282,12 @@ vectorizable_conversion (vec_info *vinfo,
return false;
if (!VECTOR_BOOLEAN_TYPE_P (vectype_out)
- && ((INTEGRAL_TYPE_P (lhs_type)
- && !type_has_mode_precision_p (lhs_type))
- || (INTEGRAL_TYPE_P (rhs_type)
- && !type_has_mode_precision_p (rhs_type))))
+ && INTEGRAL_TYPE_P (lhs_type)
+ && !type_has_mode_precision_p (lhs_type))
{
if (dump_enabled_p ())
dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location,
- "type conversion to/from bit-precision unsupported."
- "\n");
+ "type conversion to bit-precision unsupported\n");
return false;
}
generated by cgit v1.1 at 2025-11-07 22:37:41 +0000