[Qemu-devel] [PATCH 4/7] tcg-ia64: Implement deposit operation.

2011-01-10 Thread Richard Henderson
Signed-off-by: Richard Henderson 
---
 tcg/ia64/tcg-target.c |  115 +
 tcg/ia64/tcg-target.h |2 +
 2 files changed, 117 insertions(+), 0 deletions(-)

diff --git a/tcg/ia64/tcg-target.c b/tcg/ia64/tcg-target.c
index e2e44f7..5d19a61 100644
--- a/tcg/ia64/tcg-target.c
+++ b/tcg/ia64/tcg-target.c
@@ -237,6 +237,8 @@ enum {
 OPC_CMP4_LT_A6= 0x184ull,
 OPC_CMP4_LTU_A6   = 0x1a4ull,
 OPC_CMP4_EQ_A6= 0x1c4ull,
+OPC_DEP_I14   = 0x0aeull,
+OPC_DEP_I15   = 0x080ull,
 OPC_DEP_Z_I12 = 0x0a6ull,
 OPC_EXTR_I11  = 0x0a42000ull,
 OPC_EXTR_U_I11= 0x0a4ull,
@@ -508,6 +510,32 @@ static inline uint64_t tcg_opc_i12(int qp, uint64_t opc, 
int r1,
| (qp & 0x3f);
 }
 
+static inline uint64_t tcg_opc_i14(int qp, uint64_t opc, int r1, int imm1,
+   int r3, uint64_t len,
+   uint64_t cpos)
+{
+return opc
+   | ((imm1 & 1LL) << 36)
+   | ((len & 0x0f) << 27)
+   | ((r3 & 0x7f) << 20)
+   | ((cpos & 0x3f) << 14)
+   | ((r1 & 0x7f) << 6)
+   | (qp & 0x3f);
+}
+
+static inline uint64_t tcg_opc_i15(int qp, uint64_t opc, int r1,
+   int r2, int r3, uint64_t len,
+   uint64_t cpos)
+{
+return opc
+   | ((cpos & 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
@@ -1335,6 +1363,84 @@ 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 void tcg_out_deposit_i32(TCGContext *s, TCGArg out, TCGArg in,
+TCGArg val, unsigned ofs, unsigned len)
+{
+uint64_t nop_m = tcg_opc_m48(TCG_REG_P0, OPC_NOP_M48, 0);
+
+if (in == 0) {
+tcg_out_bundle(s, mmI, nop_m, nop_m,
+   tcg_opc_i12(TCG_REG_P0, OPC_DEP_Z_I12, out, val,
+   len - 1, 63 - ofs));
+} else if (val == 0) {
+tcg_out_bundle(s, mmI, nop_m, nop_m,
+   tcg_opc_i14(TCG_REG_P0, OPC_DEP_I14, out, val,
+   in, len - 1, 63 - ofs));
+} else if (len <= 16) {
+tcg_out_bundle(s, mmI, nop_m, nop_m,
+   tcg_opc_i15(TCG_REG_P0, OPC_DEP_I15, out, val, in,
+   len - 1, 63 - ofs));
+} else {
+/* Perform the 17- to 32-bit deposit in two parts.  At the same
+   time we perform the first deposit, extract the high 16-bits
+   into a scratch register.  */
+tcg_out_bundle(s, miI, nop_m,
+   tcg_opc_i11(TCG_REG_P0, OPC_EXTR_U_I11, TCG_REG_R2,
+   val, 16, 31 - 16),
+   tcg_opc_i15(TCG_REG_P0, OPC_DEP_I15, out, val, in,
+   16 - 1, 63 - ofs));
+tcg_out_bundle(s, mmI, nop_m, nop_m,
+   tcg_opc_i15(TCG_REG_P0, OPC_DEP_I15, out, TCG_REG_R2,
+   out, len - 16 - 1, 63 - (ofs + 16)));
+}
+}
+
+static void tcg_out_deposit_i64(TCGContext *s, TCGArg out, TCGArg in,
+TCGArg val, unsigned ofs, unsigned len)
+{
+uint64_t nop_m = tcg_opc_m48(TCG_REG_P0, OPC_NOP_M48, 0);
+
+if (in == 0) {
+tcg_out_bundle(s, mmI, nop_m, nop_m,
+   tcg_opc_i12(TCG_REG_P0, OPC_DEP_Z_I12, out, in,
+   len - 1, 63 - ofs));
+} else if (val == 0) {
+tcg_out_bundle(s, mmI, nop_m, nop_m,
+   tcg_opc_i14(TCG_REG_P0, OPC_DEP_I14, out, val,
+   in, len - 1, 63 - ofs));
+} else if (len <= 16) {
+tcg_out_bundle(s, mmI, nop_m, nop_m,
+   tcg_opc_i15(TCG_REG_P0, OPC_DEP_I15, out, val, in,
+   len - 1, 63 - ofs));
+} else {
+uint64_t ror = 0, shrp, rol = 0;
+
+if (ofs) {
+   ror = tcg_opc_i10(TCG_REG_P0, OPC_SHRP_I10, TCG_REG_R2,
+  in, in, ofs);
+in = TCG_REG_R2;
+}
+
+shrp = tcg_opc_i10(TCG_REG_P0, OPC_SHRP_I10, out, in, val, len);
+
+ofs = (ofs - len) & 63;
+if (ofs) {
+rol = tcg_opc_i10(TCG_REG_P0, OPC_SHRP_I10, out, out, out, 64-ofs);
+}
+
+if (ror) {
+tcg_out_bundle(s, mII, nop_m, ror, shrp);
+if (rol) {
+tcg_out_bundle(s, mmI, nop_m, nop_m, rol);
+}
+} else if (rol) {
+   

[Qemu-devel] [PATCH 4/7] tcg-ia64: Implement deposit operation.

2011-01-07 Thread Richard Henderson
Signed-off-by: Richard Henderson 
---
 tcg/ia64/tcg-target.c |   92 +
 tcg/ia64/tcg-target.h |2 +
 2 files changed, 94 insertions(+), 0 deletions(-)

diff --git a/tcg/ia64/tcg-target.c b/tcg/ia64/tcg-target.c
index 3ddf434..2708d55 100644
--- a/tcg/ia64/tcg-target.c
+++ b/tcg/ia64/tcg-target.c
@@ -237,6 +237,7 @@ enum {
 OPC_CMP4_LT_A6= 0x184ull,
 OPC_CMP4_LTU_A6   = 0x1a4ull,
 OPC_CMP4_EQ_A6= 0x1c4ull,
+OPC_DEP_I15   = 0x080ull,
 OPC_DEP_Z_I12 = 0x0a6ull,
 OPC_EXTR_I11  = 0x0a42000ull,
 OPC_EXTR_U_I11= 0x0a4ull,
@@ -508,6 +509,19 @@ static inline uint64_t tcg_opc_i12(int qp, uint64_t opc, 
int r1,
| (qp & 0x3f);
 }
 
+static inline uint64_t tcg_opc_i15(int qp, uint64_t opc, int r1,
+   int r2, int r3, int len, uint64_t len,
+   uint64_t cpos)
+{
+return opc
+   | ((cpos & 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
@@ -1335,6 +1349,73 @@ 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 void tcg_out_deposit_i32(TCGContext *s, TCGArg out, TCGArg in,
+TCGArg val, unsigned ofs, unsigned len)
+{
+uint64_t nop_m = tcg_opc_m48(TCG_REG_P0, OPC_NOP_M48, 0);
+
+if (val == 0) {
+tcg_out_bundle(s, mmI, nop_m, nop_m,
+   tcg_opc_i12(TCG_REG_P0, OPC_DEP_Z_I12, out, in,
+   len - 1, 63 - ofs));
+} else if (len <= 16) {
+tcg_out_bundle(s, mmI, nop_m, nop_m,
+   tcg_opc_i15(TCG_REG_P0, OPC_DEP_I15, out, in, val,
+   len - 1, 63 - ofs));
+} else {
+tcg_out_bundle(s, miI, nop_m,
+   tcg_opc_i11(TCG_REG_P0, OPC_EXTR_U_I11, TCG_REG_R2,
+   val, 16, 31 - 16),
+   tcg_opc_i15(TCG_REG_P0, OPC_DEP_I15, out, in, val,
+   16 - 1, 63 - ofs));
+tcg_out_bundle(s, mmI, nop_m, nop_m,
+   tcg_opc_i15(TCG_REG_P0, OPC_DEP_I15, out, out,
+   TCG_REG_R2, len - 16 - 1, 63 - (ofs + 16)));
+}
+}
+
+static void tcg_out_deposit_i64(TCGContext *s, TCGArg out, TCGArg in,
+TCGArg val, unsigned ofs, unsigned len)
+{
+uint64_t nop_m = tcg_opc_m48(TCG_REG_P0, OPC_NOP_M48, 0);
+
+if (val == 0) {
+tcg_out_bundle(s, mmI, nop_m, nop_m,
+   tcg_opc_i12(TCG_REG_P0, OPC_DEP_Z_I12, out, in,
+   len - 1, 63 - ofs));
+} else if (len <= 16) {
+tcg_out_bundle(s, mmI, nop_m, nop_m,
+   tcg_opc_i15(TCG_REG_P0, OPC_DEP_I15, out, in, val,
+   len - 1, 63 - ofs));
+} else {
+uint64_t ror = 0, shrp, rol = 0;
+
+if (ofs) {
+   ror = tcg_opc_i10(TCG_REG_P0, OPC_SHRP_I10, TCG_REG_R2,
+  in, in, ofs);
+in = TCG_REG_R2;
+}
+
+shrp = tcg_opc_i10(TCG_REG_P0, OPC_SHRP_I10, out, in, val, len);
+
+ofs = (ofs - len) & 63;
+if (ofs) {
+rol = tcg_opc_i10(TCG_REG_P0, OPC_SHRP_I10, out, out, out, 64-ofs);
+}
+
+if (ror) {
+tcg_out_bundle(s, mII, nop_m, ror, shrp);
+if (rol) {
+tcg_out_bundle(s, mmI, nop_m, nop_m, rol);
+}
+} else if (rol) {
+tcg_out_bundle(s, mII, nop_m, shrp, rol);
+} else {
+tcg_out_bundle(s, mmI, nop_m, nop_m, shrp);
+}
+}
+}
+
 static inline uint64_t tcg_opc_cmp_a(int qp, TCGCond cond, TCGArg arg1,
  TCGArg arg2, int cmp4)
 {
@@ -2063,6 +2144,15 @@ static inline void tcg_out_op(TCGContext *s, TCGOpcode 
opc,
 tcg_out_rotr_i64(s, args[0], args[1], args[2], const_args[2]);
 break;
 
+case INDEX_op_deposit_i32:
+tcg_out_deposit_i32(s, args[0], args[1], args[2],
+(args[3] >> 8) & 31, args[3] & 31);
+break;
+case INDEX_op_deposit_i64:
+tcg_out_deposit_i64(s, args[0], args[1], args[2],
+(args[3] >> 8) & 63, args[3] & 63);
+break;
+
 case INDEX_op_ext8s_i32:
 case INDEX_op_ext8s_i64:
 tcg_out_ext(s, OPC_SXT1_I29, args[0], args[1]);
@@ -2192,6 +2282,7 @@ static const TCGTargetOpDef ia64_op_defs[] = {
 { INDEX_op_shr_i32, { "r", "rZ", "ri" } },
 { IN