[x86] Recognize vpcmov in combine with -mxop.
Commit Message
By way of an apology for causing PR target/105791, where I'd overlooked
the need to support V1TImode in TARGET_XOP's vpcmov instruction, this
patch further improves support for TARGET_XOP's vpcmov instruction, by
recognizing it in combine.
Currently, the test case:
typedef int v4si __attribute__ ((vector_size (16)));
v4si foo(v4si c, v4si t, v4si f)
{
return (c&t)|(~c&f);
}
on x86_64 with -O2 -mxop generates:
vpxor %xmm2, %xmm1, %xmm1
vpand %xmm0, %xmm1, %xmm1
vpxor %xmm2, %xmm1, %xmm0
ret
but with this patch now generates:
vpcmov %xmm0, %xmm2, %xmm1, %xmm0
ret
On its own, the new combine splitter works fine on TARGET_64BIT, but
alas with -m32 combine incorrectly thinks the replacement instruction
is more expensive, as IF_THEN_ELSE isn't currently/correctly handled
in ix86_rtx_costs. So to avoid the need for a target selector in the
new testcase, I've updated ix86_rtx_costs to report that AMD's vpcmov
has a latency of two cycles [it's now an obsolete instruction set
extension and there's unlikely to ever be a processor where this
instruction has a different timing], and while there I also added
rtx_costs for x86_64's integer conditional move instructions (which
have single cycle latency).
This patch has been tested on x86_64-pc-linux-gnu with make bootstrap
and make -k check, both with and without --target_board=unix{-m32},
with no new failures. Ok for mainline?
2022-06-04 Roger Sayle <roger@nextmovesoftware.com>
gcc/ChangeLog
* config/i386/i386.cc (ix86_rtx_costs): Add a new case for
IF_THEN_ELSE, and provide costs for TARGET_XOP's vpcmov and
TARGET_CMOVE's (scalar integer) conditional moves.
* config/i386/sse.md (define_split): Recognize XOP's vpcmov
from its equivalent (canonical) pxor;pand;pxor sequence.
gcc/testsuite/ChangeLog
* gcc.target/i386/xop-pcmov3.c: New test case.
Thanks in advance,
Roger
--
Comments
On Sat, Jun 4, 2022 at 1:03 PM Roger Sayle <roger@nextmovesoftware.com> wrote:
>
>
> By way of an apology for causing PR target/105791, where I'd overlooked
> the need to support V1TImode in TARGET_XOP's vpcmov instruction, this
> patch further improves support for TARGET_XOP's vpcmov instruction, by
> recognizing it in combine.
>
> Currently, the test case:
>
> typedef int v4si __attribute__ ((vector_size (16)));
> v4si foo(v4si c, v4si t, v4si f)
> {
> return (c&t)|(~c&f);
> }
>
> on x86_64 with -O2 -mxop generates:
> vpxor %xmm2, %xmm1, %xmm1
> vpand %xmm0, %xmm1, %xmm1
> vpxor %xmm2, %xmm1, %xmm0
> ret
>
> but with this patch now generates:
> vpcmov %xmm0, %xmm2, %xmm1, %xmm0
> ret
>
> On its own, the new combine splitter works fine on TARGET_64BIT, but
> alas with -m32 combine incorrectly thinks the replacement instruction
> is more expensive, as IF_THEN_ELSE isn't currently/correctly handled
> in ix86_rtx_costs. So to avoid the need for a target selector in the
> new testcase, I've updated ix86_rtx_costs to report that AMD's vpcmov
> has a latency of two cycles [it's now an obsolete instruction set
> extension and there's unlikely to ever be a processor where this
> instruction has a different timing], and while there I also added
> rtx_costs for x86_64's integer conditional move instructions (which
> have single cycle latency).
>
> This patch has been tested on x86_64-pc-linux-gnu with make bootstrap
> and make -k check, both with and without --target_board=unix{-m32},
> with no new failures. Ok for mainline?
>
>
> 2022-06-04 Roger Sayle <roger@nextmovesoftware.com>
>
> gcc/ChangeLog
> * config/i386/i386.cc (ix86_rtx_costs): Add a new case for
> IF_THEN_ELSE, and provide costs for TARGET_XOP's vpcmov and
> TARGET_CMOVE's (scalar integer) conditional moves.
> * config/i386/sse.md (define_split): Recognize XOP's vpcmov
> from its equivalent (canonical) pxor;pand;pxor sequence.
>
> gcc/testsuite/ChangeLog
> * gcc.target/i386/xop-pcmov3.c: New test case.
OK with a nit below.
Thanks,
Uros.
+{
+ operands[5] = REGNO (operands[4]) == REGNO (operands[1]) ? operands[2]
+ : operands[1];
+})
+
Please expand this to enhance readability, it is a bit too cryptic for me ...
@@ -21009,6 +21009,37 @@ ix86_rtx_costs (rtx x, machine_mode mode, int outer_code_i, int opno,
}
return false;
+ case IF_THEN_ELSE:
+ if (TARGET_XOP
+ && VECTOR_MODE_P (mode)
+ && (GET_MODE_SIZE (mode) == 16 || GET_MODE_SIZE (mode) == 32))
+ {
+ /* vpcmov. */
+ *total = speed ? COSTS_N_INSNS (2) : COSTS_N_BYTES (6);
+ if (!REG_P (XEXP (x, 0)))
+ *total += rtx_cost (XEXP (x, 0), mode, code, 0, speed);
+ if (!REG_P (XEXP (x, 1)))
+ *total += rtx_cost (XEXP (x, 1), mode, code, 1, speed);
+ if (!REG_P (XEXP (x, 2)))
+ *total += rtx_cost (XEXP (x, 2), mode, code, 2, speed);
+ return true;
+ }
+ else if (TARGET_CMOVE
+ && SCALAR_INT_MODE_P (mode)
+ && GET_MODE_SIZE (mode) <= UNITS_PER_WORD)
+ {
+ /* cmov. */
+ *total = COSTS_N_INSNS (1);
+ if (!REG_P (XEXP (x, 0)))
+ *total += rtx_cost (XEXP (x, 0), mode, code, 0, speed);
+ if (!REG_P (XEXP (x, 1)))
+ *total += rtx_cost (XEXP (x, 1), mode, code, 1, speed);
+ if (!REG_P (XEXP (x, 2)))
+ *total += rtx_cost (XEXP (x, 2), mode, code, 2, speed);
+ return true;
+ }
+ return false;
+
default:
return false;
}
@@ -23858,6 +23858,26 @@
"vpcmov\t{%3, %2, %1, %0|%0, %1, %2, %3}"
[(set_attr "type" "sse4arg")])
+;; Recognize XOP's vpcmov from canonical (xor (and (xor t f) c) f)
+(define_split
+ [(set (match_operand:V_128_256 0 "register_operand")
+ (xor:V_128_256
+ (and:V_128_256
+ (xor:V_128_256 (match_operand:V_128_256 1 "register_operand")
+ (match_operand:V_128_256 2 "register_operand"))
+ (match_operand:V_128_256 3 "nonimmediate_operand"))
+ (match_operand:V_128_256 4 "register_operand")))]
+ "TARGET_XOP
+ && (REGNO (operands[4]) == REGNO (operands[1])
+ || REGNO (operands[4]) == REGNO (operands[2]))"
+ [(set (match_dup 0) (if_then_else:V_128_256 (match_dup 3)
+ (match_dup 5)
+ (match_dup 4)))]
+{
+ operands[5] = REGNO (operands[4]) == REGNO (operands[1]) ? operands[2]
+ : operands[1];
+})
+
;; XOP horizontal add/subtract instructions
(define_insn "xop_phadd<u>bw"
[(set (match_operand:V8HI 0 "register_operand" "=x")
new file mode 100644
@@ -0,0 +1,10 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -mxop" } */
+typedef int v4si __attribute__ ((vector_size (16)));
+
+v4si foo(v4si c, v4si t, v4si f)
+{
+ return (c&t)|(~c&f);
+}
+
+/* { dg-final { scan-assembler "vpcmov" } } */