[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[PATCH 20/20] tcg/i386: Implement bitsel for avx512
From: |
Richard Henderson |
Subject: |
[PATCH 20/20] tcg/i386: Implement bitsel for avx512 |
Date: |
Sat, 18 Dec 2021 11:42:50 -0800 |
The general ternary logic operation can implement BITSEL.
Funnel the 4-operand operation into three variants of the
3-operand instruction, depending on input operand overlap.
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
tcg/i386/tcg-target.h | 2 +-
tcg/i386/tcg-target.c.inc | 20 +++++++++++++++++++-
2 files changed, 20 insertions(+), 2 deletions(-)
diff --git a/tcg/i386/tcg-target.h b/tcg/i386/tcg-target.h
index 433a632d6a..ae3612d745 100644
--- a/tcg/i386/tcg-target.h
+++ b/tcg/i386/tcg-target.h
@@ -204,7 +204,7 @@ extern bool have_movbe;
#define TCG_TARGET_HAS_mul_vec 1
#define TCG_TARGET_HAS_sat_vec 1
#define TCG_TARGET_HAS_minmax_vec 1
-#define TCG_TARGET_HAS_bitsel_vec 0
+#define TCG_TARGET_HAS_bitsel_vec have_avx512vl
#define TCG_TARGET_HAS_cmpsel_vec -1
#define TCG_TARGET_deposit_i32_valid(ofs, len) \
diff --git a/tcg/i386/tcg-target.c.inc b/tcg/i386/tcg-target.c.inc
index 86c66f9522..be88ccae4d 100644
--- a/tcg/i386/tcg-target.c.inc
+++ b/tcg/i386/tcg-target.c.inc
@@ -2805,7 +2805,7 @@ static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
TCGType type = vecl + TCG_TYPE_V64;
int insn, sub;
- TCGArg a0, a1, a2;
+ TCGArg a0, a1, a2, a3;
a0 = args[0];
a1 = args[1];
@@ -3028,6 +3028,22 @@ static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
sub = 0xdd; /* orB!C */
goto gen_simd_imm8;
+ case INDEX_op_bitsel_vec:
+ insn = OPC_VPTERNLOGQ;
+ a3 = args[3];
+ if (a0 == a1) {
+ a1 = a2;
+ a2 = a3;
+ sub = 0xca; /* A?B:C */
+ } else if (a0 == a2) {
+ a2 = a3;
+ sub = 0xe2; /* B?A:C */
+ } else {
+ tcg_out_mov(s, type, a0, a3);
+ sub = 0xb8; /* B?C:A */
+ }
+ goto gen_simd_imm8;
+
gen_simd_imm8:
tcg_debug_assert(insn != OPC_UD2);
if (type == TCG_TYPE_V256) {
@@ -3296,6 +3312,7 @@ static TCGConstraintSetIndex tcg_target_op_def(TCGOpcode
op)
case INDEX_op_x86_vpshrdv_vec:
return C_O1_I3(x, 0, x, x);
+ case INDEX_op_bitsel_vec:
case INDEX_op_x86_vpblendvb_vec:
return C_O1_I3(x, x, x, x);
@@ -3318,6 +3335,7 @@ int tcg_can_emit_vec_op(TCGOpcode opc, TCGType type,
unsigned vece)
case INDEX_op_nor_vec:
case INDEX_op_eqv_vec:
case INDEX_op_not_vec:
+ case INDEX_op_bitsel_vec:
return 1;
case INDEX_op_cmp_vec:
case INDEX_op_cmpsel_vec:
--
2.25.1
- [PATCH 09/20] tcg/i386: Implement avx512 scalar shift, (continued)
- [PATCH 09/20] tcg/i386: Implement avx512 scalar shift, Richard Henderson, 2021/12/18
- [PATCH 11/20] tcg/i386: Implement avx512 immediate rotate, Richard Henderson, 2021/12/18
- [PATCH 13/20] tcg/i386: Support avx512vbmi2 vector shift-double instructions, Richard Henderson, 2021/12/18
- [PATCH 10/20] tcg/i386: Implement avx512 immediate sari shift, Richard Henderson, 2021/12/18
- [PATCH 12/20] tcg/i386: Implement avx512 variable rotate, Richard Henderson, 2021/12/18
- [PATCH 15/20] tcg/i386: Remove rotls_vec from tcg_target_op_def, Richard Henderson, 2021/12/18
- [PATCH 14/20] tcg/i386: Expand vector word rotate as avx512vbmi2 shift-double, Richard Henderson, 2021/12/18
- [PATCH 19/20] tcg/i386: Implement more logical operations for avx512, Richard Henderson, 2021/12/18
- [PATCH 17/20] tcg/i386: Implement avx512 min/max/abs, Richard Henderson, 2021/12/18
- [PATCH 18/20] tcg/i386: Implement avx512 multiply, Richard Henderson, 2021/12/18
- [PATCH 20/20] tcg/i386: Implement bitsel for avx512,
Richard Henderson <=
- [PATCH 16/20] tcg/i386: Expand scalar rotate with avx512 insns, Richard Henderson, 2021/12/18