1. Patchew
  2. QEMU
  3. [PATCH v4 00/12] tcg/riscv: Add support for vector
  4. View patch

[PATCH v4 11/12] tcg/riscv: Implement vector roti/v/x ops

LIU Zhiwei posted 12 patches 2 months, 1 week ago
There is a newer version of this series
[PATCH v4 11/12] tcg/riscv: Implement vector roti/v/x ops
Posted by LIU Zhiwei 2 months, 1 week ago 
From: TANG Tiancheng <tangtiancheng.ttc@alibaba-inc.com>

Signed-off-by: TANG Tiancheng <tangtiancheng.ttc@alibaba-inc.com>
Reviewed-by: Liu Zhiwei <zhiwei_liu@linux.alibaba.com>
---
 tcg/riscv/tcg-target.c.inc | 35 +++++++++++++++++++++++++++++++++++
 tcg/riscv/tcg-target.h     |  6 +++---
 2 files changed, 38 insertions(+), 3 deletions(-)

diff --git a/tcg/riscv/tcg-target.c.inc b/tcg/riscv/tcg-target.c.inc
index 16785ebe8e..afc9747780 100644
--- a/tcg/riscv/tcg-target.c.inc
+++ b/tcg/riscv/tcg-target.c.inc
@@ -2494,6 +2494,33 @@ static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
         set_vtype_len_sew(s, type, vece);
         tcg_out_vshifti(s, OPC_VSRA_VI, OPC_VSRA_VX, a0, a1, a2);
         break;
+    case INDEX_op_rotli_vec:
+        set_vtype_len_sew(s, type, vece);
+        tcg_out_vshifti(s, OPC_VSLL_VI, OPC_VSLL_VX, TCG_REG_V0, a1, a2);
+        tcg_out_vshifti(s, OPC_VSRL_VI, OPC_VSRL_VX, a0, a1, -a2);
+        tcg_out_opc_vv(s, OPC_VOR_VV, a0, a0, TCG_REG_V0, true);
+        break;
+    case INDEX_op_rotls_vec:
+        set_vtype_len_sew(s, type, vece);
+        tcg_out_opc_vx(s, OPC_VSLL_VX, TCG_REG_V0, a1, a2, true);
+        tcg_out_opc_reg(s, OPC_SUBW, a2, TCG_REG_ZERO, a2);
+        tcg_out_opc_vx(s, OPC_VSRL_VX, a0, a1, a2, true);
+        tcg_out_opc_vv(s, OPC_VOR_VV, a0, a0, TCG_REG_V0, true);
+        break;
+    case INDEX_op_rotlv_vec:
+        set_vtype_len_sew(s, type, vece);
+        tcg_out_opc_vv(s, OPC_VSLL_VV, TCG_REG_V0, a1, a2, true);
+        tcg_out_opc_vi(s, OPC_VRSUB_VI, TCG_REG_V0, a2, 0, true);
+        tcg_out_opc_vv(s, OPC_VSRL_VV, a0, a1, TCG_REG_V0, true);
+        tcg_out_opc_vv(s, OPC_VOR_VV, a0, a0, TCG_REG_V0, true);
+        break;
+    case INDEX_op_rotrv_vec:
+        set_vtype_len_sew(s, type, vece);
+        tcg_out_opc_vi(s, OPC_VRSUB_VI, TCG_REG_V0, a2, 0, true);
+        tcg_out_opc_vv(s, OPC_VSLL_VV, TCG_REG_V0, a1, TCG_REG_V0, true);
+        tcg_out_opc_vv(s, OPC_VSRL_VV, a0, a1, a2, true);
+        tcg_out_opc_vv(s, OPC_VOR_VV, a0, a0, TCG_REG_V0, true);
+        break;
     case INDEX_op_cmpsel_vec:
         tcg_out_cmpsel(s, type, vece, args[5], a0, a1, a2, const_args[2],
                        args[3], const_args[3], args[4], const_args[4]);
@@ -2556,6 +2583,10 @@ int tcg_can_emit_vec_op(TCGOpcode opc, TCGType type, unsigned vece)
     case INDEX_op_shri_vec:
     case INDEX_op_shli_vec:
     case INDEX_op_sari_vec:
+    case INDEX_op_rotls_vec:
+    case INDEX_op_rotlv_vec:
+    case INDEX_op_rotrv_vec:
+    case INDEX_op_rotli_vec:
     case INDEX_op_cmpsel_vec:
         return 1;
     case INDEX_op_cmp_vec:
@@ -2715,6 +2746,7 @@ static TCGConstraintSetIndex tcg_target_op_def(TCGOpcode op)
     case INDEX_op_shli_vec:
     case INDEX_op_shri_vec:
     case INDEX_op_sari_vec:
+    case INDEX_op_rotli_vec:
         return C_O1_I1(v, v);
     case INDEX_op_add_vec:
     case INDEX_op_sub_vec:
@@ -2733,10 +2765,13 @@ static TCGConstraintSetIndex tcg_target_op_def(TCGOpcode op)
     case INDEX_op_shlv_vec:
     case INDEX_op_shrv_vec:
     case INDEX_op_sarv_vec:
+    case INDEX_op_rotlv_vec:
+    case INDEX_op_rotrv_vec:
         return C_O1_I2(v, v, v);
     case INDEX_op_shls_vec:
     case INDEX_op_shrs_vec:
     case INDEX_op_sars_vec:
+    case INDEX_op_rotls_vec:
         return C_O1_I2(v, v, r);
     case INDEX_op_cmp_vec:
         return C_O1_I2(v, v, vL);
diff --git a/tcg/riscv/tcg-target.h b/tcg/riscv/tcg-target.h
index 76d30e789b..e6d66cd1b9 100644
--- a/tcg/riscv/tcg-target.h
+++ b/tcg/riscv/tcg-target.h
@@ -154,9 +154,9 @@ typedef enum {
 #define TCG_TARGET_HAS_not_vec          1
 #define TCG_TARGET_HAS_neg_vec          1
 #define TCG_TARGET_HAS_abs_vec          0
-#define TCG_TARGET_HAS_roti_vec         0
-#define TCG_TARGET_HAS_rots_vec         0
-#define TCG_TARGET_HAS_rotv_vec         0
+#define TCG_TARGET_HAS_roti_vec         1
+#define TCG_TARGET_HAS_rots_vec         1
+#define TCG_TARGET_HAS_rotv_vec         1
 #define TCG_TARGET_HAS_shi_vec          1
 #define TCG_TARGET_HAS_shs_vec          1
 #define TCG_TARGET_HAS_shv_vec          1
-- 
2.43.0
Re: [PATCH v4 11/12] tcg/riscv: Implement vector roti/v/x ops
Posted by Richard Henderson 2 months, 1 week ago 
On 9/11/24 06:26, LIU Zhiwei wrote:
> From: TANG Tiancheng <tangtiancheng.ttc@alibaba-inc.com>
> 
> Signed-off-by: TANG Tiancheng <tangtiancheng.ttc@alibaba-inc.com>
> Reviewed-by: Liu Zhiwei <zhiwei_liu@linux.alibaba.com>
> ---
>   tcg/riscv/tcg-target.c.inc | 35 +++++++++++++++++++++++++++++++++++
>   tcg/riscv/tcg-target.h     |  6 +++---
>   2 files changed, 38 insertions(+), 3 deletions(-)
> 
> diff --git a/tcg/riscv/tcg-target.c.inc b/tcg/riscv/tcg-target.c.inc
> index 16785ebe8e..afc9747780 100644
> --- a/tcg/riscv/tcg-target.c.inc
> +++ b/tcg/riscv/tcg-target.c.inc
> @@ -2494,6 +2494,33 @@ static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
>           set_vtype_len_sew(s, type, vece);
>           tcg_out_vshifti(s, OPC_VSRA_VI, OPC_VSRA_VX, a0, a1, a2);
>           break;
> +    case INDEX_op_rotli_vec:
> +        set_vtype_len_sew(s, type, vece);
> +        tcg_out_vshifti(s, OPC_VSLL_VI, OPC_VSLL_VX, TCG_REG_V0, a1, a2);
> +        tcg_out_vshifti(s, OPC_VSRL_VI, OPC_VSRL_VX, a0, a1, -a2);

You will want to mask -a2, because otherwise it will always fail to match imm < 32 within 
tcg_out_vshifti:

     -a2 & ((8 << vece) - 1)

> +    case INDEX_op_rotlv_vec:
> +        set_vtype_len_sew(s, type, vece);
> +        tcg_out_opc_vv(s, OPC_VSLL_VV, TCG_REG_V0, a1, a2, true);
> +        tcg_out_opc_vi(s, OPC_VRSUB_VI, TCG_REG_V0, a2, 0, true);
> +        tcg_out_opc_vv(s, OPC_VSRL_VV, a0, a1, TCG_REG_V0, true);

You have written to V0 twice, clobbering the result.
Need to swap the shifts:

	vrsub.vi	v0, a2, 0
	vsrl.vv		v0, a1, v0
	vsll.vv		a0, a1, a2
	vor.vv		a0, a0, v0


r~

Patchew 2.1-devel-b67e4c2

© 2016 - 2024 Red Hat, Inc.