[PATCH] RISC-V: zicond: remove bogus opt2 pattern

[Vineet Gupta]

This was tripping up gcc.c-torture/execute/pr60003.c at -O1 since the
pattern semantics can't be expressed by zicond instructions.

This involves test code snippet:

      if (a == 0)
	return 0;
      else
	return x;
    }

which is equivalent to:  "x = (a != 0) ? x : a"

and matches define_insn "*czero.nez.<GPR:mode><X:mode>.opt2"

| (insn 41 20 38 3 (set (reg/v:DI 136 [ x ])
|        (if_then_else:DI (ne (reg/v:DI 134 [ a ])
|                (const_int 0 [0]))
|            (reg/v:DI 136 [ x ])
|            (reg/v:DI 134 [ a ]))) {*czero.nez.didi.opt2}

The corresponding asm pattern generates
    czero.nez x, x, a   ; %0, %2, %1
implying
    "x = (a != 0) ? 0 : a"

which is not what the pattern semantics are.

Essentially "(a != 0) ? x : a" cannot be expressed with CZERO.nez

As a side note, while correctness prevails, this test still gets a
czero in the end, albeit a different one.

if-convert generates two if_then_else

| (insn 43 20 44 3 (set (reg:DI 143)
|        (reg/v:DI 136 [ x ])) "pr60003.c":36:9 179 {*movdi_64bit}
| (insn 44 43 46 3 (set (reg:DI 142)
|        (reg/v:DI 134 [ a ])) "pr60003.c":36:9 179 {*movdi_64bit}
|
| (insn 46 44 47 3 (set (reg:DI 145)
|        (if_then_else:DI (ne:DI (reg/v:DI 134 [ a ])
|                (const_int 0 [0]))
|            (const_int 0 [0])
|            (reg:DI 142))) "pr60003.c":36:9 14532 {*czero.nez.didi}
|
| (insn 47 46 48 3 (set (reg:DI 144)
|        (if_then_else:DI (eq:DI (reg/v:DI 134 [ a ])
|                (const_int 0 [0]))
|            (const_int 0 [0])
|            (reg:DI 143))) "pr60003.c":36:9 14531 {*czero.eqz.didi}

and combine is able to fuse them together

| (insn 38 48 39 3 (set (reg/i:DI 10 a0)
|        (if_then_else:DI (eq:DI (reg/v:DI 134 [ a ])
|                (const_int 0 [0]))
|            (const_int 0 [0])
|            (reg:DI 143))) "pr60003.c":40:1 14531 {*czero.eqz.didi}

before fix		after fix
-----------------	-----------------
li        a5,1		li        a0,1
ld        a4,8(sp)	ld        a5,8(sp)
czero.nez a0,a4,a5	czero.eqz a0,a5,a0

The issue only happens at -O1 as at higher optimization levels, the
whole conditional move gets optimized away.

gcc/ChangeLog:
	* config/riscv/zicond.md: Remove incorrect op2 pattern.

Fixes: 1d5bc3285e8a ("[committed][RISC-V] Fix 20010221-1.c with zicond")
Signed-off-by: Vineet Gupta <vineetg@rivosinc.com>
---
 gcc/config/riscv/zicond.md | 10 ----------
 1 file changed, 10 deletions(-)

diff --git a/gcc/config/riscv/zicond.md b/gcc/config/riscv/zicond.md
index 25f21d33487e..aa5607a9efd8 100644
--- a/gcc/config/riscv/zicond.md
+++ b/gcc/config/riscv/zicond.md
@@ -52,13 +52,3 @@
   "TARGET_ZICOND && rtx_equal_p (operands[1], operands[2])"
   "czero.eqz\t%0,%3,%1"
 )
-
-(define_insn "*czero.nez.<GPR:mode><X:mode>.opt2"
-  [(set (match_operand:GPR 0 "register_operand"                   "=r")
-        (if_then_else:GPR (ne (match_operand:X 1 "register_operand" "r")
-                              (const_int 0))
-                          (match_operand:GPR 2 "register_operand" "r")
-                          (match_operand:GPR 3 "register_operand" "1")))]
-  "TARGET_ZICOND && rtx_equal_p (operands[1], operands[3])"
-  "czero.nez\t%0,%2,%1"
-)
-- 
2.34.1