Allow LDPREINCREMENT to match non-word ADDI-type in-place address
updates followed by scalar integer or floating-point loads. Support
register-plus-offset and LO_SUM addresses, self moves, and either
load-destination relationship.
Put the common pre/post-index load/store constraints in a
parameterized helper for later fusion rules.
gcc/ChangeLog:
* config/riscv/riscv-fusion.cc (riscv_fuse_indexed_mem_p): New
function.
(riscv_fuse_ldpreincrement): Use it and accept scalar integer or
floating-point loads.
gcc/testsuite/ChangeLog:
* gcc.target/riscv/fusion-preindex-load.c: New test.
* gcc.target/riscv/fusion-preindex-load-rtl.c: Likewise.
Signed-off-by: Jin Ma <[email protected]>
---
gcc/config/riscv/riscv-fusion.cc | 77 +++++++++++++------
.../riscv/fusion-preindex-load-rtl.c | 66 ++++++++++++++++
.../gcc.target/riscv/fusion-preindex-load.c | 31 ++++++++
3 files changed, 149 insertions(+), 25 deletions(-)
create mode 100644 gcc/testsuite/gcc.target/riscv/fusion-preindex-load-rtl.c
create mode 100644 gcc/testsuite/gcc.target/riscv/fusion-preindex-load.c
diff --git a/gcc/config/riscv/riscv-fusion.cc b/gcc/config/riscv/riscv-fusion.cc
index c24fb29cbaf..c6e7073cb23 100644
--- a/gcc/config/riscv/riscv-fusion.cc
+++ b/gcc/config/riscv/riscv-fusion.cc
@@ -800,6 +800,45 @@ riscv_fuse_add_mem_p (rtx_insn *prev, rtx_insn *curr,
return true;
}
+/* Match an in-place ADDI-type address update and a scalar load or store using
+ the updated address. LOAD_P selects loads rather than stores, and
+ PREINDEX_P selects whether the update precedes the memory instruction. */
+
+static bool
+riscv_fuse_indexed_mem_p (rtx_insn *prev, rtx_insn *curr,
+ bool load_p, bool preindex_p)
+{
+ rtx prev_set, curr_set;
+ if (!riscv_fuse_sets_p (prev, curr, &prev_set, &curr_set))
+ return false;
+
+ rtx_insn *update_insn = preindex_p ? prev : curr;
+ rtx_insn *mem_insn = preindex_p ? curr : prev;
+ rtx update_set = preindex_p ? prev_set : curr_set;
+ rtx mem_set = preindex_p ? curr_set : prev_set;
+ rtx update_dest = SET_DEST (update_set);
+ rtx update_base = NULL_RTX;
+ struct riscv_fusion_mem_info mem;
+
+ if (!riscv_insn_is_addi_type_p (update_insn, false, &update_base)
+ || update_base == NULL_RTX
+ || !riscv_fuse_mem_p (mem_insn, &mem)
+ || (load_p
+ ? mem.type == SCHED_FUSION_ST
+ : mem.type != SCHED_FUSION_ST)
+ || !(mem.addr.type == ADDRESS_REG
+ || mem.addr.type == ADDRESS_LO_SUM)
+ || riscv_regno (update_dest) == INVALID_REGNUM
+ || !riscv_fuse_same_reg_p (update_base, update_dest)
+ || !riscv_fuse_same_reg_p (mem.addr.reg, update_dest))
+ return false;
+
+ if (load_p)
+ return true;
+
+ return !riscv_fuse_same_reg_p (SET_SRC (mem_set), update_dest);
+}
+
/* Check the common RTL for ZEXTW, ZEXTWS and ZEXTH fusion. */
static bool
@@ -1134,37 +1173,25 @@ riscv_fuse_expanded_ld (rtx_insn *prev, rtx_insn *curr)
}
/* Check for RISCV_FUSE_LDPREINCREMENT fusion.
- prev (addi) == (set (reg rd1)
- (plus (reg rd1) (const_int offset)))
- curr (load) == (set (reg rd2) (mem (reg rd1)))
+ prev (one of the following):
+ (addi) == (set (reg rd1) (plus (reg rd1) (const_int imm12)))
+ (self-mv) == (set (reg rd1) (reg rd1))
+ (addi) == (set (reg rd1) (lo_sum (reg rd1) symbol1))
+ curr (one of the following):
+ (load) == (set (reg rd2) (mem addr))
+ (load) == (set (reg rd2) (any_extend (mem addr)))
+ (fpload) == (set (reg frd) (mem addr))
+ addr (one of the following):
+ (rd1, offset)
+ (lo_sum (reg rd1) symbol2)
Constraints:
- rd1 == rd2. */
+ the ADDI-type instruction is not a word form. */
static bool
riscv_fuse_ldpreincrement (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_fuse_same_dest_p (prev_set, curr_set))
- return false;
-
- struct riscv_fusion_mem_info mem;
-
- if (riscv_fuse_mem_p (curr, &mem)
- && mem.type == SCHED_FUSION_LD
- && !mem.fp_p
- && mem.addr.type == ADDRESS_REG
- && INTVAL (mem.addr.offset) == 0
- && riscv_set_is_addi_p (prev_set)
- && riscv_fuse_same_reg_p (XEXP (SET_SRC (prev_set), 0),
- SET_DEST (prev_set))
- && riscv_fuse_same_reg_p (mem.addr.reg, SET_DEST (prev_set)))
- return true;
-
- return false;
+ return riscv_fuse_indexed_mem_p (prev, curr, true, true);
}
/* Check for RISCV_FUSE_LUI_ADDI fusion.
diff --git a/gcc/testsuite/gcc.target/riscv/fusion-preindex-load-rtl.c
b/gcc/testsuite/gcc.target/riscv/fusion-preindex-load-rtl.c
new file mode 100644
index 00000000000..c8cceb4740a
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/fusion-preindex-load-rtl.c
@@ -0,0 +1,66 @@
+/* { dg-do compile { target { rv64 } } } */
+/* { dg-skip-if "" { *-*-* } { "-O0" "-O1" "-O3" "-O[sgz]" "-flto" } } */
+/* { dg-options "-march=rv64gc -mabi=lp64d -mexplicit-relocs -O2
-mtune=xt-c9501fdvt -fdump-rtl-sched2-details" } */
+/* { dg-final { scan-rtl-dump-times "RISCV_FUSE_LDPREINCREMENT" 2 "sched2" {
xfail *-*-* } } } */
+
+extern double preindex_symbol;
+
+/* The load may write the address-update destination. */
+long __RTL (startwith ("sched2"))
+test_preindex_same_destination (void)
+{
+(function "test_preindex_same_destination"
+ (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 a0) (const_int 1))))
+ (cinsn 4 (set (reg:DI a0)
+ (zero_extend:DI
+ (mem:QI (reg:DI a0) [0 S1 A8]))))
+ (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))
+ ) ;; crtl
+) ;; function "test_preindex_same_destination"
+}
+
+/* LO_SUM updates and addresses are accepted for floating-point loads. */
+long __RTL (startwith ("sched2"))
+test_preindex_lo_sum_fload (void)
+{
+(function "test_preindex_lo_sum_fload"
+ (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 ("preindex_symbol")))))
+ (cinsn 4 (set (reg:DF fa0)
+ (mem:DF
+ (lo_sum:DI
+ (reg:DI a0)
+ (symbol_ref:DI ("preindex_symbol")))
+ [0 S8 A64])))
+ (cinsn 5 (use (reg:DI a0)))
+ (cinsn 6 (use (reg:DF fa0)))
+ (cjump_insn 7 (simple_return))
+ (edge-to exit)
+ ) ;; block 2
+ (cbarrier 8)
+ ) ;; insn-chain
+ (crtl
+ (return_rtx (reg/i:DI a0))
+ ) ;; crtl
+) ;; function "test_preindex_lo_sum_fload"
+}
diff --git a/gcc/testsuite/gcc.target/riscv/fusion-preindex-load.c
b/gcc/testsuite/gcc.target/riscv/fusion-preindex-load.c
new file mode 100644
index 00000000000..6fd83652d55
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/fusion-preindex-load.c
@@ -0,0 +1,31 @@
+/* { dg-do compile { target { rv64 } } } */
+/* { dg-skip-if "" { *-*-* } { "-O0" "-O1" "-O3" "-O[sgz]" "-funroll-loops"
"-flto" } } */
+/* { dg-options "-march=rv64gc -mabi=lp64d -mtune=xt-c9501fdvt -O2
-fdump-rtl-sched2-details" } */
+/* No tune enables this fusion pair yet. */
+/* { dg-final { scan-rtl-dump-times "RISCV_FUSE_LDPREINCREMENT" 2 "sched2" {
xfail *-*-* } } } */
+
+typedef signed char int8_t;
+
+int8_t
+test_preindex_lb (int8_t *p, int n)
+{
+ int8_t sum = 0;
+ for (int i = 0; i < n; ++i)
+ {
+ p += 2;
+ sum += *p;
+ }
+ return sum;
+}
+
+double
+test_preindex_fld (double *p, int n)
+{
+ double sum = 0;
+ for (int i = 0; i < n; ++i)
+ {
+ p += 2;
+ sum += *p;
+ }
+ return sum;
+}
--
2.52.0