So pre-commit CI flagged an issue with the initial version of this
patch. In particular the cmp-mem-const-{1,2} tests are failing.
I didn't see that in my internal testing, but that well could be an
artifact of having multiple patches touching in the same broad space
that the tester is evaluating. If I apply just this patch I can trigger
the cmp-mem-const{1,2} failures.
The code we're getting now is actually better than we were getting
before, but the new patterns avoid the path through combine that emits
the message about narrowing the load down to a byte load, hence the failure.
Given we're getting better code now than before, I'm just skipping this
test on risc-v. That's the only non-whitespace change since the
original version of this patch.
--
This addresses the first level issues seen in generating better
performing code for testcases derived from pr121136. It likely
regresses code size in some cases as in many cases it selects code
sequences that should be better performing, though larger to encode.
Improving -Os code generation should remain the primary focus of
pr121136. Any improvements in code size with this change are a nice
side effect, but not the primary goal.
--
Let's take this test (derived from the PR):
_Bool func1_0x1U (unsigned int x) { return x <= 0x1U; }
_Bool func2_0x1U (unsigned int x) { return ((x >> __builtin_ctz (0x1U +
1U)) == 0); }
_Bool func3_0x1U (unsigned int x) { return ((x / (0x1U + 1U)) == 0); }
Those should produce the same output. We currently get these fragments
for the 3 cases. In particular note how the second variant is a two
instruction sequence.
sltiu a0,a0,2
srliw a0,a0,1
seqz a0,a0
sltiu a0,a0,2
This patch will adjust that second sequence to match the first and third
and is optimal.
Let's take another case. This is interesting as it's right at the
simm12 border:
_Bool func1_0x7ffU (unsigned long x) { return x <= 0x7ffU; }
_Bool func2_0x7ffU (unsigned long x) { return ((x >> __builtin_ctzl
(0x7ffU + 1UL)) == 0); }
_Bool func3_0x7ffU (unsigned long x) { return ((x / (0x7ffU + 1UL)) == 0); }
We get:
li a5,2047
sltu a0,a5,a0
seqz a0,a0
srli a0,a0,11
seqz a0,a0
li a5,2047
sltu a0,a5,a0
seqz a0,a0
In this case the second sequence is pretty good. Not perfect, but
clearly better than the other two. This patch will fix the code for
case #1 and case #3 to match case #2.
So anyway, that's the basic motivation here. So to be 100% clear, while
the bug is focused on code size, I'm focused on the performance of the
resulting code.
This has been tested on riscv32-elf and riscv64-elf. It's also
bootstrapped and regression tested on the Pioneer. The BPI won't have
results for this patch until late tomorrow.
--
PR rtl-optimization/121136
gcc/
* config/riscv/riscv.md: Add define_insn to test the
upper bits of a register against zero using sltiu when
the bits are extracted via zero_extract or logial right shift.
Add 3->2 define_splits for gtu/leu cases testing upper bits
against zero.
gcc/testsuite
* gcc.target/riscv/pr121136.c: New test.
* gcc.dg/cmp-mem-const-1.c: Skip for risc-v.
* gcc.dg/cmp-mem-const-2.c: Likewise.
diff --git a/gcc/config/riscv/riscv.md b/gcc/config/riscv/riscv.md
index 640ca5f9b0ea..1ec15da2e77e 100644
--- a/gcc/config/riscv/riscv.md
+++ b/gcc/config/riscv/riscv.md
@@ -3752,6 +3752,57 @@ (define_insn "*sle<u>_<X:mode><GPR:mode>"
[(set_attr "type" "slt")
(set_attr "mode" "<X:MODE>")])
+;; We can sometimes do better for unsigned comparisons against
+;; values where there's a run of 1s in the LSBs.
+;;
+(define_split
+ [(set (match_operand:X 0 "register_operand")
+ (gtu:X (match_operand:X 1 "register_operand")
+ (match_operand 2 "const_int_operand")))
+ (clobber (match_operand:X 3 "register_operand"))]
+ "exact_log2 (INTVAL (operands[2]) + 1) >= 0"
+ [(set (match_dup 3) (lshiftrt:X (match_dup 1) (match_dup 2)))
+ (set (match_dup 0) (ne:X (match_dup 3) (const_int 0)))]
+{ operands[2] = GEN_INT (exact_log2 (INTVAL (operands[2]) + 1)); })
+
+(define_split
+ [(set (match_operand:X 0 "register_operand")
+ (leu:X (match_operand:X 1 "register_operand")
+ (match_operand 2 "const_int_operand")))
+ (clobber (match_operand:X 3 "register_operand"))]
+ "exact_log2 (INTVAL (operands[2]) + 1) >= 0"
+ [(set (match_dup 3) (lshiftrt:X (match_dup 1) (match_dup 2)))
+ (set (match_dup 0) (eq:X (match_dup 3) (const_int 0)))]
+{ operands[2] = GEN_INT (exact_log2 (INTVAL (operands[2]) + 1)); })
+
+;; Alternate forms that are ultimately just sltiu
+(define_insn ""
+ [(set (match_operand:X 0 "register_operand" "=r")
+ (eq:X (zero_extract:X (match_operand:X 1 "register_operand" "r")
+ (match_operand 2 "const_int_operand")
+ (match_operand 3 "const_int_operand"))
+ (const_int 0)))]
+ "(INTVAL (operands[3]) < 11
+ && INTVAL (operands[2]) + INTVAL (operands[3]) == BITS_PER_WORD)"
+{
+ operands[2] = GEN_INT (HOST_WIDE_INT_1U << INTVAL (operands[3]));
+ return "sltiu\t%0,%1,%2";
+}
+ [(set_attr "type" "slt")
+ (set_attr "mode" "<X:MODE>")])
+
+(define_insn ""
+ [(set (match_operand:X 0 "register_operand" "=r")
+ (eq:X (lshiftrt:X (match_operand:X 1 "register_operand" "r")
+ (match_operand 2 "const_int_operand"))
+ (const_int 0)))]
+ "INTVAL (operands[2]) < 11"
+{
+ operands[2] = GEN_INT (HOST_WIDE_INT_1U << INTVAL (operands[2]));
+ return "sltiu\t%0,%1,%2";
+}
+ [(set_attr "type" "slt")
+ (set_attr "mode" "<X:MODE>")])
;;
;; ....................
;;
diff --git a/gcc/testsuite/gcc.dg/cmp-mem-const-1.c
b/gcc/testsuite/gcc.dg/cmp-mem-const-1.c
index 0b0e7331354a..4f94902a9287 100644
--- a/gcc/testsuite/gcc.dg/cmp-mem-const-1.c
+++ b/gcc/testsuite/gcc.dg/cmp-mem-const-1.c
@@ -1,6 +1,7 @@
/* { dg-do compile { target { lp64 } } } */
/* { dg-options "-O2 -fdump-rtl-combine-details" } */
/* { dg-final { scan-rtl-dump "narrow comparison from mode .I to QI" "combine"
} } */
+/* { dg-skip-if "" { riscv*-*-* } } */
typedef __UINT64_TYPE__ uint64_t;
diff --git a/gcc/testsuite/gcc.dg/cmp-mem-const-2.c
b/gcc/testsuite/gcc.dg/cmp-mem-const-2.c
index 8022137a8eca..12d962aa3584 100644
--- a/gcc/testsuite/gcc.dg/cmp-mem-const-2.c
+++ b/gcc/testsuite/gcc.dg/cmp-mem-const-2.c
@@ -1,6 +1,7 @@
/* { dg-do compile { target { lp64 } } } */
/* { dg-options "-O2 -fdump-rtl-combine-details" } */
/* { dg-final { scan-rtl-dump "narrow comparison from mode .I to QI" "combine"
} } */
+/* { dg-skip-if "" { riscv*-*-* } } */
typedef __UINT64_TYPE__ uint64_t;
diff --git a/gcc/testsuite/gcc.target/riscv/pr121136.c
b/gcc/testsuite/gcc.target/riscv/pr121136.c
new file mode 100644
index 000000000000..77585f42cbf4
--- /dev/null
+++ b/gcc/testsuite/gcc.target/riscv/pr121136.c
@@ -0,0 +1,103 @@
+/* { dg-do compile } */
+/* { dg-options "-std=gnu23 -O2 -march=rv64gc -mabi=lp64d" { target rv64} } */
+/* { dg-options "-std=gnu23 -O2 -march=rv32gc -mabi=ilp32" { target rv32} } */
+
+/* We need to adjust the constant so this works for rv32 and rv64. */
+#if __riscv_xlen == 32
+#define ONE 1U
+#define TYPE unsigned int
+#define CTZ __builtin_ctz
+#else
+#define ONE 1UL
+#define TYPE unsigned long
+#define CTZ __builtin_ctzl
+#endif
+
+#define F1(C) _Bool func1_##C(TYPE x) { return x <= C; }
+#define F2(C) _Bool func2_##C(TYPE x) { return ((x >> CTZ (C+ONE)) == 0); }
+#define F3(C) _Bool func3_##C(TYPE x) { return ((x / (C+ONE)) == 0); }
+
+#define F(C) F1(C) F2(C) F3(C)
+
+F (0x1U)
+F (0x3U)
+F (0x7U)
+F (0xfU)
+F (0x1fU)
+F (0x3fU)
+F (0x7fU)
+F (0xffU)
+F (0x1ffU)
+F (0x3ffU)
+F (0x7ffU)
+F (0xfffU)
+F (0x1fffU)
+F (0x3fffU)
+F (0x7fffU)
+F (0xffffU)
+F (0x1ffffU)
+F (0x3ffffU)
+F (0x7ffffU)
+F (0xfffffU)
+F (0x1fffffU)
+F (0x3fffffU)
+F (0x7fffffU)
+F (0xffffffU)
+F (0x1ffffffU)
+F (0x3ffffffU)
+F (0x7ffffffU)
+F (0xfffffffU)
+F (0x1fffffffU)
+F (0x3fffffffU)
+F (0x7fffffffU)
+#if __riscv_xlen == 64
+F (0xffffffffUL)
+F (0x1ffffffffUL)
+F (0x3ffffffffUL)
+F (0x7ffffffffUL)
+F (0xfffffffffUL)
+F (0x1fffffffffUL)
+F (0x3fffffffffUL)
+F (0x7fffffffffUL)
+F (0xffffffffffUL)
+F (0x1ffffffffffUL)
+F (0x3ffffffffffUL)
+F (0x7ffffffffffUL)
+F (0xfffffffffffUL)
+F (0x1fffffffffffUL)
+F (0x3fffffffffffUL)
+F (0x7fffffffffffUL)
+F (0xffffffffffffUL)
+F (0x1ffffffffffffUL)
+F (0x3ffffffffffffUL)
+F (0x7ffffffffffffUL)
+F (0xfffffffffffffUL)
+F (0x1fffffffffffffUL)
+F (0x3fffffffffffffUL)
+F (0x7fffffffffffffUL)
+F (0xffffffffffffffUL)
+F (0x1ffffffffffffffUL)
+F (0x3ffffffffffffffUL)
+F (0x7ffffffffffffffUL)
+F (0xfffffffffffffffUL)
+F (0x1fffffffffffffffUL)
+F (0x3fffffffffffffffUL)
+F (0x7fffffffffffffffUL)
+#endif
+
+/* These represent current state. They are not optimal as some of the cases
+ where we shift are better implemented by loading 2^n constant and using
+ sltu as the lui has no incoming data dependencies. */
+/* { dg-final { scan-assembler-times "\tsltiu" 30 { target { rv64 } } } } */
+/* { dg-final { scan-assembler-times "\tnot" 3 { target { rv64 } } } } */
+/* { dg-final { scan-assembler-times "\tsrli" 121 { target { rv64 } } } } */
+/* { dg-final { scan-assembler-times "\tsltu" 38 { target { rv64 } } } } */
+/* { dg-final { scan-assembler-times "\tseqz" 118 { target { rv64 } } } } */
+/* { dg-final { scan-assembler-times "\tli" 38 { target { rv64 } } } } */
+
+/* { dg-final { scan-assembler-times "\tsltiu" 30 { target { rv32 } } } } */
+/* { dg-final { scan-assembler-times "\tnot" 3 { target { rv32 } } } } */
+/* { dg-final { scan-assembler-times "\tsrli" 25 { target { rv32 } } } } */
+/* { dg-final { scan-assembler-times "\tsltu" 38 { target { rv32 } } } } */
+/* { dg-final { scan-assembler-times "\tseqz" 22 { target { rv32 } } } } */
+/* { dg-final { scan-assembler-times "\tli" 38 { target { rv32 } } } } */
