Recognize ADD/ADDI-type instructions followed by ANDI when the result
register forms the required dependency. Parameterize the ANDI
classifier so the new recognizer and existing B_ALUI logic share RTL
parsing.
Leave the fusion disabled by default.
gcc/ChangeLog:
* config/riscv/riscv-fusion.cc (riscv_insn_is_add_addi_p): New
function.
(riscv_insn_is_andi_type_p): Likewise.
(riscv_fuse_b_alui): Reuse the ANDI matcher.
(riscv_fuse_add_andi): New function.
(riscv_fusion_table): Add RISCV_FUSE_ADD_ANDI.
* config/riscv/riscv-protos.h (enum riscv_fusion_pairs): Add
RISCV_FUSE_ADD_ANDI.
gcc/testsuite/ChangeLog:
* gcc.target/riscv/fusion-add-addi-andi.c: New test.
* gcc.target/riscv/fusion-mv-li-andi-rtl.c: Likewise.
* gcc.target/riscv/fusion-special-positive-rtl-add-andi.c: Likewise.
Signed-off-by: Jin Ma <[email protected]>
---
gcc/config/riscv/riscv-fusion.cc | 95 ++++++++++++++++++-
gcc/config/riscv/riscv-protos.h | 1 +
.../gcc.target/riscv/fusion-add-addi-andi.c | 45 +++++++++
.../gcc.target/riscv/fusion-mv-li-andi-rtl.c | 63 ++++++++++++
.../fusion-special-positive-rtl-add-andi.c | 66 +++++++++++++
5 files changed, 267 insertions(+), 3 deletions(-)
create mode 100644 gcc/testsuite/gcc.target/riscv/fusion-add-addi-andi.c
create mode 100644 gcc/testsuite/gcc.target/riscv/fusion-mv-li-andi-rtl.c
create mode 100644
gcc/testsuite/gcc.target/riscv/fusion-special-positive-rtl-add-andi.c
diff --git a/gcc/config/riscv/riscv-fusion.cc b/gcc/config/riscv/riscv-fusion.cc
index bb4076e9ffc..816c7dea452 100644
--- a/gcc/config/riscv/riscv-fusion.cc
+++ b/gcc/config/riscv/riscv-fusion.cc
@@ -702,6 +702,57 @@ riscv_insn_is_sub_type_p (rtx_insn *insn)
&& REG_P (XEXP (src, 1)));
}
+/* Matches an add- or addi-type instruction. */
+
+static bool
+riscv_insn_is_add_addi_p (rtx_insn *insn)
+{
+ return (riscv_insn_is_add_type_p (insn)
+ || riscv_insn_is_addi_type_p (insn));
+}
+
+/* Matches an andi:
+ (set (reg rd) (and (reg rs1) (const_int imm12)))
+ or an equivalent zero-extend RTL form. Store the normalized operands in
+ *SRC0 and *SRC1 when requested. */
+
+static bool
+riscv_insn_is_andi_type_p (rtx_insn *insn, rtx *src0 = NULL,
+ rtx *src1 = NULL)
+{
+ rtx set = single_set (insn);
+ if (!set)
+ return false;
+
+ rtx src = SET_SRC (set);
+ if (get_attr_type (insn) == TYPE_LOGICAL
+ && GET_CODE (src) == AND
+ && REG_P (XEXP (src, 0))
+ && CONST_INT_P (XEXP (src, 1))
+ && riscv_regno (SET_DEST (set)) != INVALID_REGNUM)
+ {
+ if (src0)
+ *src0 = XEXP (src, 0);
+ if (src1)
+ *src1 = XEXP (src, 1);
+ return true;
+ }
+
+ if (get_attr_move_type (insn) == MOVE_TYPE_ANDI
+ && GET_CODE (src) == ZERO_EXTEND
+ && riscv_regno (XEXP (src, 0)) != INVALID_REGNUM
+ && riscv_regno (SET_DEST (set)) != INVALID_REGNUM)
+ {
+ if (src0)
+ *src0 = XEXP (src, 0);
+ if (src1)
+ *src1 = GEN_INT (0xff);
+ return true;
+ }
+
+ return false;
+}
+
/* Fusion recognizers. */
/* Check for RISCV_FUSE_ZEXTW fusion.
@@ -1150,10 +1201,11 @@ riscv_fuse_b_alui (rtx_insn *prev, rtx_insn *curr)
return true;
/* ctz + andi. */
+ rtx andi_src1;
if (GET_CODE (SET_SRC (prev_set)) == CTZ
- && GET_CODE (SET_SRC (curr_set)) == AND
- && CONST_INT_P (XEXP (SET_SRC (curr_set), 1))
- && INTVAL (XEXP (SET_SRC (curr_set), 1)) == 63)
+ && riscv_insn_is_andi_type_p (curr, NULL, &andi_src1)
+ && CONST_INT_P (andi_src1)
+ && INTVAL (andi_src1) == 63)
return true;
/* sub + smax (abs pattern). */
@@ -1207,6 +1259,41 @@ riscv_fuse_sub_seqz (rtx_insn *prev, rtx_insn *curr)
return false;
}
+/* Check for RISCV_FUSE_ADD_ANDI fusion.
+ prev (one of the following):
+ (add) == (set (reg rd1) (plus (reg rs1) (reg rs2)))
+ (addi) == (set (reg rd1) (plus (reg rs1) (const_int imm12_1)))
+ (addw) == (set (reg rd1)
+ (sign_extend (plus (reg rs1) (reg rs2))))
+ (addiw) == (set (reg rd1)
+ (sign_extend (plus (reg rs1) (const_int imm12_1))))
+ (add.uw) == (set (reg rd1) (plus (zero_extend (reg rs1))
+ (reg rs2)))
+ (mv) == (set (reg rd1) (reg rs1))
+ (li) == (set (reg rd1) (const_int imm12_1))
+ (addi) == (set (reg rd1) (lo_sum (reg rs1) symbol))
+ curr (one of the following):
+ (andi) == (set (reg rd2) (and (reg rd1) (const_int imm12_2)))
+ (andi) == (set (reg rd2) (zero_extend (reg rd1)))
+
+ Constraints:
+ rd1 == rd2. */
+
+static bool
+riscv_fuse_add_andi (rtx_insn *prev, rtx_insn *curr)
+{
+ rtx prev_set, curr_set;
+ if (!riscv_fuse_sets_p (prev, curr, &prev_set, &curr_set))
+ return false;
+
+ if (riscv_insn_is_add_addi_p (prev)
+ && riscv_insn_is_andi_type_p (curr)
+ && riscv_fuse_same_dest_p (prev_set, curr_set, true))
+ return true;
+
+ return false;
+}
+
/* Type for a fusion checker function. Takes the two candidate insns
and returns true if they should be fused. */
@@ -1260,6 +1347,8 @@ static const struct riscv_fusion_entry
riscv_fusion_table[] =
riscv_fuse_b_alui, "RISCV_FUSE_B_ALUI" },
{ RISCV_FUSE_SUB_SEQZ,
riscv_fuse_sub_seqz, "RISCV_FUSE_SUB_SEQZ" },
+ { RISCV_FUSE_ADD_ANDI,
+ riscv_fuse_add_andi, "RISCV_FUSE_ADD_ANDI" },
};
/* Implement TARGET_SCHED_MACRO_FUSION_PAIR_P. Return true if PREV and CURR
diff --git a/gcc/config/riscv/riscv-protos.h b/gcc/config/riscv/riscv-protos.h
index 0bd26523a98..0848bc0ec4f 100644
--- a/gcc/config/riscv/riscv-protos.h
+++ b/gcc/config/riscv/riscv-protos.h
@@ -868,6 +868,7 @@ enum riscv_fusion_pairs
RISCV_FUSE_B_ALUI = HOST_WIDE_INT_1U << 12,
RISCV_FUSE_SUB_SEQZ = HOST_WIDE_INT_1U << 13,
RISCV_FUSE_ADD_ST = HOST_WIDE_INT_1U << 14,
+ RISCV_FUSE_ADD_ANDI = HOST_WIDE_INT_1U << 15,
};
extern bool riscv_macro_fusion_p (void);
diff --git a/gcc/testsuite/gcc.target/riscv/fusion-add-addi-andi.c
b/gcc/testsuite/gcc.target/riscv/fusion-add-addi-andi.c
new file mode 100644
index 00000000000..983d40c8f6b
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/fusion-add-addi-andi.c
@@ -0,0 +1,45 @@
+/* { dg-do compile { target { rv64 } } } */
+/* { dg-skip-if "" { *-*-* } { "-O0" "-O1" "-O3" "-O[sgz]" "-flto" } } */
+/* { dg-options "-march=rv64gc_zba -mabi=lp64d -mtune=xt-c9501fdvt -O2
-fdump-rtl-sched2-details" } */
+/* No tune enables these fusion pairs yet. */
+/* { dg-final { scan-rtl-dump-times "RISCV_FUSE_ADD_ANDI" 5 "sched2" { xfail
*-*-* } } } */
+
+typedef long int64_t;
+typedef unsigned long uint64_t;
+typedef int int32_t;
+typedef unsigned int uint32_t;
+
+/* addi + andi should fuse. */
+int64_t
+test_addi_andi (int64_t a)
+{
+ return (a + 16) & 0x55;
+}
+
+/* addiw + andi should fuse. */
+int64_t
+test_addiw_andi (int32_t a)
+{
+ return (int64_t) (a + 5) & 0x55;
+}
+
+/* add + andi should fuse. */
+int64_t
+test_add_andi (int64_t a, int64_t b)
+{
+ return (a + b) & 0x55;
+}
+
+/* addw + andi should fuse. */
+int64_t
+test_addw_andi (int32_t a, int32_t b)
+{
+ return (int64_t) (a + b) & 0x33;
+}
+
+/* add.uw + andi should fuse. */
+uint64_t
+test_adduw_andi (uint64_t a, uint64_t b)
+{
+ return ((uint64_t) (uint32_t) a + b) & -2UL;
+}
diff --git a/gcc/testsuite/gcc.target/riscv/fusion-mv-li-andi-rtl.c
b/gcc/testsuite/gcc.target/riscv/fusion-mv-li-andi-rtl.c
new file mode 100644
index 00000000000..507efd0c2c7
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/fusion-mv-li-andi-rtl.c
@@ -0,0 +1,63 @@
+/* { dg-do compile { target { rv64 } } } */
+/* { dg-skip-if "" { *-*-* } { "-O0" "-O1" "-O3" "-O[sgz]" "-flto" } } */
+/* { dg-options "-march=rv64gc -mabi=lp64d -mtune=xt-c9501fdvt -O2
-fdump-rtl-sched2-details" } */
+/* No tune enables these fusion pairs yet. */
+/* { dg-final { scan-rtl-dump-times "RISCV_FUSE_ADD_ANDI" 2 "sched2" { xfail
*-*-* } } } */
+
+/* mv followed by andi should fuse. */
+long __RTL (startwith ("sched2"))
+test_mv_andi (void)
+{
+(function "test_mv_andi"
+ (insn-chain
+ (block 2
+ (edge-from entry (flags "FALLTHRU"))
+ (cnote 1 [bb 2] NOTE_INSN_BASIC_BLOCK)
+ (cnote 2 NOTE_INSN_FUNCTION_BEG)
+ (cinsn 3 (set (reg:DI a0)
+ (reg:DI a1)))
+ (cinsn 4 (set (reg:DI a0)
+ (and:DI (reg:DI a0)
+ (const_int 85))))
+ (cinsn 5 (use (reg/i:DI a0)))
+ (cjump_insn 6 (simple_return))
+ (edge-to exit)
+ ) ;; block 2
+ (cbarrier 7)
+ ) ;; insn-chain
+ (crtl
+ (return_rtx
+ (reg/i:DI a0)
+ ) ;; return_rtx
+ ) ;; crtl
+) ;; function "test_mv_andi"
+}
+
+/* li followed by andi should fuse. */
+long __RTL (startwith ("sched2"))
+test_li_andi (void)
+{
+(function "test_li_andi"
+ (insn-chain
+ (block 2
+ (edge-from entry (flags "FALLTHRU"))
+ (cnote 1 [bb 2] NOTE_INSN_BASIC_BLOCK)
+ (cnote 2 NOTE_INSN_FUNCTION_BEG)
+ (cinsn 3 (set (reg:DI a0)
+ (const_int 31)))
+ (cinsn 4 (set (reg:DI a0)
+ (and:DI (reg:DI a0)
+ (const_int 85))))
+ (cinsn 5 (use (reg/i:DI a0)))
+ (cjump_insn 6 (simple_return))
+ (edge-to exit)
+ ) ;; block 2
+ (cbarrier 7)
+ ) ;; insn-chain
+ (crtl
+ (return_rtx
+ (reg/i:DI a0)
+ ) ;; return_rtx
+ ) ;; crtl
+) ;; function "test_li_andi"
+}
diff --git
a/gcc/testsuite/gcc.target/riscv/fusion-special-positive-rtl-add-andi.c
b/gcc/testsuite/gcc.target/riscv/fusion-special-positive-rtl-add-andi.c
new file mode 100644
index 00000000000..40a554e1cbd
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/fusion-special-positive-rtl-add-andi.c
@@ -0,0 +1,66 @@
+/* { dg-do compile { target { rv64 } } } */
+/* { dg-skip-if "" { *-*-* } { "-O0" "-O1" "-O3" "-O[sgz]" "-flto" } } */
+/* { dg-options "-march=rv64gc_zbb -mabi=lp64d -mtune=xt-c9501fdvt -O2
-fdump-rtl-sched2-details" } */
+/* { dg-final { scan-rtl-dump-times "RISCV_FUSE_ADD_ANDI" 2 "sched2" { xfail
*-*-* } } } */
+
+extern long fusion_low_symbol;
+
+/* add followed by andi expressed as zero_extend should fuse. */
+long __RTL (startwith ("sched2"))
+test_add_zero_extend_andi (void)
+{
+(function "test_add_zero_extend_andi"
+ (insn-chain
+ (block 2
+ (edge-from entry (flags "FALLTHRU"))
+ (cnote 1 [bb 2] NOTE_INSN_BASIC_BLOCK)
+ (cnote 2 NOTE_INSN_FUNCTION_BEG)
+ (cinsn 3 (set (reg:DI a0)
+ (plus:DI (reg:DI a1)
+ (reg:DI a2))))
+ (cinsn 4 (set (reg:DI a0)
+ (zero_extend:DI
+ (subreg:QI (reg:DI a0) 0))))
+ (cinsn 5 (use (reg/i:DI a0)))
+ (cjump_insn 6 (simple_return))
+ (edge-to exit)
+ ) ;; block 2
+ (cbarrier 7)
+ ) ;; insn-chain
+ (crtl
+ (return_rtx
+ (reg/i:DI a0)
+ ) ;; return_rtx
+ ) ;; crtl
+) ;; function "test_add_zero_extend_andi"
+}
+/* lo_sum update followed by andi should fuse. */
+long __RTL (startwith ("sched2"))
+test_lo_sum_andi (void)
+{
+(function "test_lo_sum_andi"
+ (insn-chain
+ (block 2
+ (edge-from entry (flags "FALLTHRU"))
+ (cnote 1 [bb 2] NOTE_INSN_BASIC_BLOCK)
+ (cnote 2 NOTE_INSN_FUNCTION_BEG)
+ (cinsn 3 (set (reg:DI a0)
+ (lo_sum:DI
+ (reg:DI a0)
+ (symbol_ref:DI ("fusion_low_symbol")))))
+ (cinsn 4 (set (reg:DI a0)
+ (and:DI (reg:DI a0)
+ (const_int 85))))
+ (cinsn 5 (use (reg/i:DI a0)))
+ (cjump_insn 6 (simple_return))
+ (edge-to exit)
+ ) ;; block 2
+ (cbarrier 7)
+ ) ;; insn-chain
+ (crtl
+ (return_rtx
+ (reg/i:DI a0)
+ ) ;; return_rtx
+ ) ;; crtl
+) ;; function "test_lo_sum_andi"
+}
--
2.52.0