* [Qemu-devel] [PATCH] tcg-ia64: Implement deposit
@ 2012-10-10 23:01 Richard Henderson
2012-10-16 23:26 ` Aurelien Jarno
0 siblings, 1 reply; 2+ messages in thread
From: Richard Henderson @ 2012-10-10 23:01 UTC (permalink / raw)
To: qemu-devel; +Cc: Aurelien Jarno
Note that in the general reg=reg,reg case we're restricted
to 16-bit insertions. This makes it easy to allow "any"
constant as input, as post-truncation it will fit into the
constant load insn for which we have room in the bundle.
Signed-off-by: Richard Henderson <rth@twiddle.net>
---
tcg/ia64/tcg-target.c | 53 +++++++++++++++++++++++++++++++++++++++++++++++++++
tcg/ia64/tcg-target.h | 7 +++++--
2 files changed, 58 insertions(+), 2 deletions(-)
Depends on Aurelien's patch set due to the use of dep_i14.
r~
diff --git a/tcg/ia64/tcg-target.c b/tcg/ia64/tcg-target.c
index 6f018f4..06570be 100644
--- a/tcg/ia64/tcg-target.c
+++ b/tcg/ia64/tcg-target.c
@@ -231,6 +231,7 @@ enum {
OPC_CMP4_LTU_A6 = 0x1a400000000ull,
OPC_CMP4_EQ_A6 = 0x1c400000000ull,
OPC_DEP_I14 = 0x0ae00000000ull,
+ OPC_DEP_I15 = 0x08000000000ull,
OPC_DEP_Z_I12 = 0x0a600000000ull,
OPC_EXTR_I11 = 0x0a400002000ull,
OPC_EXTR_U_I11 = 0x0a400000000ull,
@@ -514,6 +515,18 @@ static inline uint64_t tcg_opc_i14(int qp, uint64_t opc, int r1, uint64_t imm,
| (qp & 0x3f);
}
+static inline uint64_t tcg_opc_i15(int qp, uint64_t opc, int r1, int r2,
+ int r3, uint64_t pos, uint64_t len)
+{
+ return opc
+ | ((pos & 0x3f) << 31)
+ | ((len & 0x0f) << 27)
+ | ((r3 & 0x7f) << 20)
+ | ((r2 & 0x7f) << 13)
+ | ((r1 & 0x7f) << 6)
+ | (qp & 0x3f);
+}
+
static inline uint64_t tcg_opc_i18(int qp, uint64_t opc, uint64_t imm)
{
return opc
@@ -1325,6 +1338,37 @@ static inline void tcg_out_bswap64(TCGContext *s, TCGArg ret, TCGArg arg)
tcg_opc_i3 (TCG_REG_P0, OPC_MUX1_I3, ret, arg, 0xb));
}
+static inline void tcg_out_deposit(TCGContext *s, TCGArg ret, TCGArg a1,
+ TCGArg a2, int const_a2, int pos, int len)
+{
+ uint64_t i1 = 0, i2 = 0;
+ int cpos = 63 - pos, lm1 = len - 1;
+
+ if (const_a2) {
+ /* Truncate the value of a constant a2 to the width of the field. */
+ int mask = (1u << len) - 1;
+ a2 &= mask;
+
+ if (a2 == 0 || a2 == mask) {
+ /* 1-bit signed constant inserted into register. */
+ i2 = tcg_opc_i14(TCG_REG_P0, OPC_DEP_I14, ret, a2, a1, cpos, lm1);
+ } else {
+ /* Otherwise, load any constant into a temporary. Do this into
+ the first I slot to help out with cross-unit delays. */
+ i1 = tcg_opc_a5(TCG_REG_P0, OPC_ADDL_A5,
+ TCG_REG_R2, a2, TCG_REG_R0);
+ a2 = TCG_REG_R2;
+ }
+ }
+ if (i2 == 0) {
+ i2 = tcg_opc_i15(TCG_REG_P0, OPC_DEP_I15, ret, a2, a1, cpos, lm1);
+ }
+ tcg_out_bundle(s, (i1 ? mII : miI),
+ tcg_opc_m48(TCG_REG_P0, OPC_NOP_M48, 0),
+ i1 ? i1 : tcg_opc_i18(TCG_REG_P0, OPC_NOP_I18, 0),
+ i2);
+}
+
static inline uint64_t tcg_opc_cmp_a(int qp, TCGCond cond, TCGArg arg1,
TCGArg arg2, int cmp4)
{
@@ -2130,6 +2174,12 @@ static inline void tcg_out_op(TCGContext *s, TCGOpcode opc,
tcg_out_bswap64(s, args[0], args[1]);
break;
+ case INDEX_op_deposit_i32:
+ case INDEX_op_deposit_i64:
+ tcg_out_deposit(s, args[0], args[1], args[2], const_args[2],
+ args[3], args[4]);
+ break;
+
case INDEX_op_brcond_i32:
tcg_out_brcond(s, args[2], args[0], const_args[0],
args[1], const_args[1], args[3], 1);
@@ -2294,6 +2344,9 @@ static const TCGTargetOpDef ia64_op_defs[] = {
{ INDEX_op_setcond_i64, { "r", "rZ", "rZ" } },
{ INDEX_op_movcond_i64, { "r", "rZ", "rZ", "rI", "rI" } },
+ { INDEX_op_deposit_i32, { "r", "rZ", "ri" } },
+ { INDEX_op_deposit_i64, { "r", "rZ", "ri" } },
+
{ INDEX_op_qemu_ld8u, { "r", "r" } },
{ INDEX_op_qemu_ld8s, { "r", "r" } },
{ INDEX_op_qemu_ld16u, { "r", "r" } },
diff --git a/tcg/ia64/tcg-target.h b/tcg/ia64/tcg-target.h
index 5e7d970..4255ca5 100644
--- a/tcg/ia64/tcg-target.h
+++ b/tcg/ia64/tcg-target.h
@@ -133,8 +133,11 @@ typedef enum {
#define TCG_TARGET_HAS_rot_i64 1
#define TCG_TARGET_HAS_movcond_i32 1
#define TCG_TARGET_HAS_movcond_i64 1
-#define TCG_TARGET_HAS_deposit_i32 0
-#define TCG_TARGET_HAS_deposit_i64 0
+#define TCG_TARGET_HAS_deposit_i32 1
+#define TCG_TARGET_HAS_deposit_i64 1
+
+#define TCG_TARGET_deposit_i32_valid(ofs, len) ((len) <= 16)
+#define TCG_TARGET_deposit_i64_valid(ofs, len) ((len) <= 16)
/* optional instructions automatically implemented */
#define TCG_TARGET_HAS_neg_i32 0 /* sub r1, r0, r3 */
--
1.7.11.4
^ permalink raw reply related [flat|nested] 2+ messages in thread
* Re: [Qemu-devel] [PATCH] tcg-ia64: Implement deposit
2012-10-10 23:01 [Qemu-devel] [PATCH] tcg-ia64: Implement deposit Richard Henderson
@ 2012-10-16 23:26 ` Aurelien Jarno
0 siblings, 0 replies; 2+ messages in thread
From: Aurelien Jarno @ 2012-10-16 23:26 UTC (permalink / raw)
To: Richard Henderson; +Cc: qemu-devel
On Wed, Oct 10, 2012 at 04:01:35PM -0700, Richard Henderson wrote:
> Note that in the general reg=reg,reg case we're restricted
> to 16-bit insertions. This makes it easy to allow "any"
> constant as input, as post-truncation it will fit into the
> constant load insn for which we have room in the bundle.
>
> Signed-off-by: Richard Henderson <rth@twiddle.net>
> ---
> tcg/ia64/tcg-target.c | 53 +++++++++++++++++++++++++++++++++++++++++++++++++++
> tcg/ia64/tcg-target.h | 7 +++++--
> 2 files changed, 58 insertions(+), 2 deletions(-)
>
> Depends on Aurelien's patch set due to the use of dep_i14.
>
>
> r~
>
>
> diff --git a/tcg/ia64/tcg-target.c b/tcg/ia64/tcg-target.c
> index 6f018f4..06570be 100644
> --- a/tcg/ia64/tcg-target.c
> +++ b/tcg/ia64/tcg-target.c
> @@ -231,6 +231,7 @@ enum {
> OPC_CMP4_LTU_A6 = 0x1a400000000ull,
> OPC_CMP4_EQ_A6 = 0x1c400000000ull,
> OPC_DEP_I14 = 0x0ae00000000ull,
> + OPC_DEP_I15 = 0x08000000000ull,
> OPC_DEP_Z_I12 = 0x0a600000000ull,
> OPC_EXTR_I11 = 0x0a400002000ull,
> OPC_EXTR_U_I11 = 0x0a400000000ull,
> @@ -514,6 +515,18 @@ static inline uint64_t tcg_opc_i14(int qp, uint64_t opc, int r1, uint64_t imm,
> | (qp & 0x3f);
> }
>
> +static inline uint64_t tcg_opc_i15(int qp, uint64_t opc, int r1, int r2,
> + int r3, uint64_t pos, uint64_t len)
> +{
> + return opc
> + | ((pos & 0x3f) << 31)
> + | ((len & 0x0f) << 27)
> + | ((r3 & 0x7f) << 20)
> + | ((r2 & 0x7f) << 13)
> + | ((r1 & 0x7f) << 6)
> + | (qp & 0x3f);
> +}
> +
> static inline uint64_t tcg_opc_i18(int qp, uint64_t opc, uint64_t imm)
> {
> return opc
> @@ -1325,6 +1338,37 @@ static inline void tcg_out_bswap64(TCGContext *s, TCGArg ret, TCGArg arg)
> tcg_opc_i3 (TCG_REG_P0, OPC_MUX1_I3, ret, arg, 0xb));
> }
>
> +static inline void tcg_out_deposit(TCGContext *s, TCGArg ret, TCGArg a1,
> + TCGArg a2, int const_a2, int pos, int len)
> +{
> + uint64_t i1 = 0, i2 = 0;
> + int cpos = 63 - pos, lm1 = len - 1;
> +
> + if (const_a2) {
> + /* Truncate the value of a constant a2 to the width of the field. */
> + int mask = (1u << len) - 1;
> + a2 &= mask;
> +
> + if (a2 == 0 || a2 == mask) {
> + /* 1-bit signed constant inserted into register. */
> + i2 = tcg_opc_i14(TCG_REG_P0, OPC_DEP_I14, ret, a2, a1, cpos, lm1);
> + } else {
> + /* Otherwise, load any constant into a temporary. Do this into
> + the first I slot to help out with cross-unit delays. */
> + i1 = tcg_opc_a5(TCG_REG_P0, OPC_ADDL_A5,
> + TCG_REG_R2, a2, TCG_REG_R0);
> + a2 = TCG_REG_R2;
> + }
> + }
> + if (i2 == 0) {
> + i2 = tcg_opc_i15(TCG_REG_P0, OPC_DEP_I15, ret, a2, a1, cpos, lm1);
> + }
> + tcg_out_bundle(s, (i1 ? mII : miI),
> + tcg_opc_m48(TCG_REG_P0, OPC_NOP_M48, 0),
> + i1 ? i1 : tcg_opc_i18(TCG_REG_P0, OPC_NOP_I18, 0),
> + i2);
> +}
> +
> static inline uint64_t tcg_opc_cmp_a(int qp, TCGCond cond, TCGArg arg1,
> TCGArg arg2, int cmp4)
> {
> @@ -2130,6 +2174,12 @@ static inline void tcg_out_op(TCGContext *s, TCGOpcode opc,
> tcg_out_bswap64(s, args[0], args[1]);
> break;
>
> + case INDEX_op_deposit_i32:
> + case INDEX_op_deposit_i64:
> + tcg_out_deposit(s, args[0], args[1], args[2], const_args[2],
> + args[3], args[4]);
> + break;
> +
> case INDEX_op_brcond_i32:
> tcg_out_brcond(s, args[2], args[0], const_args[0],
> args[1], const_args[1], args[3], 1);
> @@ -2294,6 +2344,9 @@ static const TCGTargetOpDef ia64_op_defs[] = {
> { INDEX_op_setcond_i64, { "r", "rZ", "rZ" } },
> { INDEX_op_movcond_i64, { "r", "rZ", "rZ", "rI", "rI" } },
>
> + { INDEX_op_deposit_i32, { "r", "rZ", "ri" } },
> + { INDEX_op_deposit_i64, { "r", "rZ", "ri" } },
> +
> { INDEX_op_qemu_ld8u, { "r", "r" } },
> { INDEX_op_qemu_ld8s, { "r", "r" } },
> { INDEX_op_qemu_ld16u, { "r", "r" } },
> diff --git a/tcg/ia64/tcg-target.h b/tcg/ia64/tcg-target.h
> index 5e7d970..4255ca5 100644
> --- a/tcg/ia64/tcg-target.h
> +++ b/tcg/ia64/tcg-target.h
> @@ -133,8 +133,11 @@ typedef enum {
> #define TCG_TARGET_HAS_rot_i64 1
> #define TCG_TARGET_HAS_movcond_i32 1
> #define TCG_TARGET_HAS_movcond_i64 1
> -#define TCG_TARGET_HAS_deposit_i32 0
> -#define TCG_TARGET_HAS_deposit_i64 0
> +#define TCG_TARGET_HAS_deposit_i32 1
> +#define TCG_TARGET_HAS_deposit_i64 1
> +
> +#define TCG_TARGET_deposit_i32_valid(ofs, len) ((len) <= 16)
> +#define TCG_TARGET_deposit_i64_valid(ofs, len) ((len) <= 16)
>
> /* optional instructions automatically implemented */
> #define TCG_TARGET_HAS_neg_i32 0 /* sub r1, r0, r3 */
Reviewed-by: Aurelien Jarno <aurelien@aurel32.net>
--
Aurelien Jarno GPG: 1024D/F1BCDB73
aurelien@aurel32.net http://www.aurel32.net
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2012-10-16 23:26 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2012-10-10 23:01 [Qemu-devel] [PATCH] tcg-ia64: Implement deposit Richard Henderson
2012-10-16 23:26 ` Aurelien Jarno
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).