[Qemu-devel] [PATCH 0/6] tcg sparc improvements

[Qemu-devel] [PATCH 0/6] tcg sparc improvements

[Richard Henderson]

Three of these patches are related to the new instructions added
in 2011 with the VIS3 instruction set, as present in the T4 cpus.


r~


Richard Henderson (6):
  tcg-sparc: Support addsub2_i64
  tcg-sparc: Use ADDXC in addsub2_i64
  tcg-sparc: Fix setcond_i32 uninitialized value
  tcg-sparc: Use ADDXC in setcond_i64
  tcg-sparc: Rename ADDX/SUBX insns
  tcg-sparc: Use UMULXHI instruction

 disas/sparc.c          |  34 ++++++-------
 include/elf.h          |  37 +++++++++++---
 tcg/sparc/tcg-target.c | 129 +++++++++++++++++++++++++++++++++++++++++++------
 tcg/sparc/tcg-target.h |  12 +++--
 4 files changed, 167 insertions(+), 45 deletions(-)

-- 
1.9.3

[Qemu-devel] [PATCH 1/6] tcg-sparc: Support addsub2_i64

[Richard Henderson]

Signed-off-by: Richard Henderson <rth@twiddle.net>
---
 tcg/sparc/tcg-target.c | 72 +++++++++++++++++++++++++++++++++++++++++++++-----
 tcg/sparc/tcg-target.h |  4 +--
 2 files changed, 67 insertions(+), 9 deletions(-)

diff --git a/tcg/sparc/tcg-target.c b/tcg/sparc/tcg-target.c
index 40f2ec1..21981d8 100644
--- a/tcg/sparc/tcg-target.c
+++ b/tcg/sparc/tcg-target.c
@@ -719,9 +719,9 @@ static void tcg_out_setcond_i64(TCGContext *s, TCGCond cond, TCGReg ret,
     }
 }
 
-static void tcg_out_addsub2(TCGContext *s, TCGReg rl, TCGReg rh,
-                            TCGReg al, TCGReg ah, int32_t bl, int blconst,
-                            int32_t bh, int bhconst, int opl, int oph)
+static void tcg_out_addsub2_i32(TCGContext *s, TCGReg rl, TCGReg rh,
+                                TCGReg al, TCGReg ah, int32_t bl, int blconst,
+                                int32_t bh, int bhconst, int opl, int oph)
 {
     TCGReg tmp = TCG_REG_T1;
 
@@ -735,6 +735,51 @@ static void tcg_out_addsub2(TCGContext *s, TCGReg rl, TCGReg rh,
     tcg_out_mov(s, TCG_TYPE_I32, rl, tmp);
 }
 
+static void tcg_out_addsub2_i64(TCGContext *s, TCGReg rl, TCGReg rh,
+                                TCGReg al, TCGReg ah, int32_t bl, int blconst,
+                                int32_t bh, int bhconst, bool is_sub)
+{
+    TCGReg tmp = TCG_REG_T1;
+
+    /* Note that the low parts are fully consumed before tmp is set.  */
+    if (rl != ah && (bhconst || rl != bh)) {
+        tmp = rl;
+    }
+
+    tcg_out_arithc(s, tmp, al, bl, blconst, is_sub ? ARITH_SUBCC : ARITH_ADDCC);
+
+    /* Note that ADDX/SUBX take the carry-in from %icc, the 32-bit carry,
+       while we want %xcc, the 64-bit carry.  */
+    /* ??? There is a 2011 VIS3 ADDXC insn that does take a 64-bit carry.  */
+
+    if (bh == TCG_REG_G0) {
+	/* If we have a zero, we can perform the operation in two insns,
+           with the arithmetic first, and a conditional move into place.  */
+	if (rh == ah) {
+            tcg_out_arithi(s, TCG_REG_T2, ah, 1,
+			   is_sub ? ARITH_SUB : ARITH_ADD);
+            tcg_out_movcc(s, TCG_COND_LTU, MOVCC_XCC, rh, TCG_REG_T2, 0);
+	} else {
+            tcg_out_arithi(s, rh, ah, 1, is_sub ? ARITH_SUB : ARITH_ADD);
+	    tcg_out_movcc(s, TCG_COND_GEU, MOVCC_XCC, rh, ah, 0);
+	}
+    } else {
+        /* Otherwise adjust BH as if there is carry into T2 ... */
+        if (bhconst) {
+            tcg_out_movi(s, TCG_TYPE_I64, TCG_REG_T2, bh + (is_sub ? -1 : 1));
+        } else {
+            tcg_out_arithi(s, TCG_REG_T2, bh, 1,
+                           is_sub ? ARITH_SUB : ARITH_ADD);
+        }
+        /* ... smoosh T2 back to original BH if carry is clear ... */
+        tcg_out_movcc(s, TCG_COND_GEU, MOVCC_XCC, TCG_REG_T2, bh, bhconst);
+	/* ... and finally perform the arithmetic with the new operand.  */
+        tcg_out_arith(s, rh, ah, TCG_REG_T2, is_sub ? ARITH_SUB : ARITH_ADD);
+    }
+
+    tcg_out_mov(s, TCG_TYPE_I64, rl, tmp);
+}
+
 static void tcg_out_call_nodelay(TCGContext *s, tcg_insn_unit *dest)
 {
     ptrdiff_t disp = tcg_pcrel_diff(s, dest);
@@ -1264,12 +1309,14 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,
         break;
 
     case INDEX_op_add2_i32:
-        tcg_out_addsub2(s, a0, a1, a2, args[3], args[4], const_args[4],
-                        args[5], const_args[5], ARITH_ADDCC, ARITH_ADDX);
+        tcg_out_addsub2_i32(s, args[0], args[1], args[2], args[3],
+                            args[4], const_args[4], args[5], const_args[5],
+                            ARITH_ADDCC, ARITH_ADDX);
         break;
     case INDEX_op_sub2_i32:
-        tcg_out_addsub2(s, a0, a1, a2, args[3], args[4], const_args[4],
-                        args[5], const_args[5], ARITH_SUBCC, ARITH_SUBX);
+        tcg_out_addsub2_i32(s, args[0], args[1], args[2], args[3],
+                            args[4], const_args[4], args[5], const_args[5],
+                            ARITH_SUBCC, ARITH_SUBX);
         break;
     case INDEX_op_mulu2_i32:
         c = ARITH_UMUL;
@@ -1351,6 +1398,14 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,
     case INDEX_op_movcond_i64:
         tcg_out_movcond_i64(s, args[5], a0, a1, a2, c2, args[3], const_args[3]);
         break;
+    case INDEX_op_add2_i64:
+        tcg_out_addsub2_i64(s, args[0], args[1], args[2], args[3], args[4],
+                            const_args[4], args[5], const_args[5], false);
+        break;
+    case INDEX_op_sub2_i64:
+        tcg_out_addsub2_i64(s, args[0], args[1], args[2], args[3], args[4],
+                            const_args[4], args[5], const_args[5], true);
+        break;
 
     gen_arith:
         tcg_out_arithc(s, a0, a1, a2, c2, c);
@@ -1449,6 +1504,9 @@ static const TCGTargetOpDef sparc_op_defs[] = {
     { INDEX_op_setcond_i64, { "R", "RZ", "RJ" } },
     { INDEX_op_movcond_i64, { "R", "RZ", "RJ", "RI", "0" } },
 
+    { INDEX_op_add2_i64, { "R", "R", "RZ", "RZ", "RJ", "RI" } },
+    { INDEX_op_sub2_i64, { "R", "R", "RZ", "RZ", "RJ", "RI" } },
+
     { INDEX_op_qemu_ld_i32, { "r", "A" } },
     { INDEX_op_qemu_ld_i64, { "R", "A" } },
     { INDEX_op_qemu_st_i32, { "sZ", "A" } },
diff --git a/tcg/sparc/tcg-target.h b/tcg/sparc/tcg-target.h
index 089f976..a44d34f 100644
--- a/tcg/sparc/tcg-target.h
+++ b/tcg/sparc/tcg-target.h
@@ -133,8 +133,8 @@ typedef enum {
 #define TCG_TARGET_HAS_nor_i64          0
 #define TCG_TARGET_HAS_deposit_i64      0
 #define TCG_TARGET_HAS_movcond_i64      1
-#define TCG_TARGET_HAS_add2_i64         0
-#define TCG_TARGET_HAS_sub2_i64         0
+#define TCG_TARGET_HAS_add2_i64         1
+#define TCG_TARGET_HAS_sub2_i64         1
 #define TCG_TARGET_HAS_mulu2_i64        0
 #define TCG_TARGET_HAS_muls2_i64        0
 #define TCG_TARGET_HAS_muluh_i64        0
-- 
1.9.3

[Qemu-devel] [PATCH 2/6] tcg-sparc: Use ADDXC in addsub2_i64

[Richard Henderson]

On T4 and newer Sparc chips we have an add-with-carry insn
that takes its input from %xcc instead of %icc.

Signed-off-by: Richard Henderson <rth@twiddle.net>
---
 disas/sparc.c          |  3 +++
 include/elf.h          | 37 +++++++++++++++++++++++++++++--------
 tcg/sparc/tcg-target.c | 28 +++++++++++++++++++++++-----
 tcg/sparc/tcg-target.h |  6 ++++++
 4 files changed, 61 insertions(+), 13 deletions(-)

diff --git a/disas/sparc.c b/disas/sparc.c
index 8eb22e6..092e1b6 100644
--- a/disas/sparc.c
+++ b/disas/sparc.c
@@ -2042,6 +2042,9 @@ IMPDEP ("impdep2", 0x37),
 
 #undef IMPDEP
 
+{ "addxc", F3F(2, 0x36, 0x011), F3F(~2, ~0x36, ~0x011), "1,2,d", 0, v9b },
+{ "addxccc", F3F(2, 0x36, 0x013), F3F(~2, ~0x36, ~0x013), "1,2,d", 0, v9b },
+
 };
 
 static const int sparc_num_opcodes = ((sizeof sparc_opcodes)/(sizeof sparc_opcodes[0]));
diff --git a/include/elf.h b/include/elf.h
index e88d52f..d7730a2 100644
--- a/include/elf.h
+++ b/include/elf.h
@@ -471,14 +471,35 @@ typedef struct {
 #define PPC_FEATURE_TRUE_LE             0x00000002
 #define PPC_FEATURE_PPC_LE              0x00000001
 
-/* Bits present in AT_HWCAP, primarily for Sparc32.  */
-
-#define HWCAP_SPARC_FLUSH       1    /* CPU supports flush instruction. */
-#define HWCAP_SPARC_STBAR       2
-#define HWCAP_SPARC_SWAP        4
-#define HWCAP_SPARC_MULDIV      8
-#define HWCAP_SPARC_V9		16
-#define HWCAP_SPARC_ULTRA3	32
+/* Bits present in AT_HWCAP for Sparc.  */
+
+#define HWCAP_SPARC_FLUSH               0x00000001
+#define HWCAP_SPARC_STBAR               0x00000002
+#define HWCAP_SPARC_SWAP                0x00000004
+#define HWCAP_SPARC_MULDIV              0x00000008
+#define HWCAP_SPARC_V9                  0x00000010
+#define HWCAP_SPARC_ULTRA3              0x00000020
+#define HWCAP_SPARC_BLKINIT             0x00000040
+#define HWCAP_SPARC_N2                  0x00000080
+#define HWCAP_SPARC_MUL32               0x00000100
+#define HWCAP_SPARC_DIV32               0x00000200
+#define HWCAP_SPARC_FSMULD              0x00000400
+#define HWCAP_SPARC_V8PLUS              0x00000800
+#define HWCAP_SPARC_POPC                0x00001000
+#define HWCAP_SPARC_VIS                 0x00002000
+#define HWCAP_SPARC_VIS2                0x00004000
+#define HWCAP_SPARC_ASI_BLK_INIT        0x00008000
+#define HWCAP_SPARC_FMAF                0x00010000
+#define HWCAP_SPARC_VIS3                0x00020000
+#define HWCAP_SPARC_HPC                 0x00040000
+#define HWCAP_SPARC_RANDOM              0x00080000
+#define HWCAP_SPARC_TRANS               0x00100000
+#define HWCAP_SPARC_FJFMAU              0x00200000
+#define HWCAP_SPARC_IMA                 0x00400000
+#define HWCAP_SPARC_ASI_CACHE_SPARING   0x00800000
+#define HWCAP_SPARC_PAUSE               0x01000000
+#define HWCAP_SPARC_CBCOND              0x02000000
+#define HWCAP_SPARC_CRYPTO              0x04000000
 
 /* Bits present in AT_HWCAP for s390.  */
 
diff --git a/tcg/sparc/tcg-target.c b/tcg/sparc/tcg-target.c
index 21981d8..08ca482 100644
--- a/tcg/sparc/tcg-target.c
+++ b/tcg/sparc/tcg-target.c
@@ -209,6 +209,8 @@ static const int tcg_target_call_oarg_regs[] = {
 #define ARITH_MOVCC (INSN_OP(2) | INSN_OP3(0x2c))
 #define ARITH_MOVR (INSN_OP(2) | INSN_OP3(0x2f))
 
+#define ARITH_ADDXC (INSN_OP(2) | INSN_OP3(0x36) | INSN_OPF(0x11))
+
 #define SHIFT_SLL  (INSN_OP(2) | INSN_OP3(0x25))
 #define SHIFT_SRL  (INSN_OP(2) | INSN_OP3(0x26))
 #define SHIFT_SRA  (INSN_OP(2) | INSN_OP3(0x27))
@@ -262,6 +264,10 @@ static const int tcg_target_call_oarg_regs[] = {
 #define STW_LE     (STWA  | INSN_ASI(ASI_PRIMARY_LITTLE))
 #define STX_LE     (STXA  | INSN_ASI(ASI_PRIMARY_LITTLE))
 
+#ifndef use_vis3_instructions
+bool use_vis3_instructions;
+#endif
+
 static inline int check_fit_i64(int64_t val, unsigned int bits)
 {
     return val == sextract64(val, 0, bits);
@@ -748,11 +754,14 @@ static void tcg_out_addsub2_i64(TCGContext *s, TCGReg rl, TCGReg rh,
 
     tcg_out_arithc(s, tmp, al, bl, blconst, is_sub ? ARITH_SUBCC : ARITH_ADDCC);
 
-    /* Note that ADDX/SUBX take the carry-in from %icc, the 32-bit carry,
-       while we want %xcc, the 64-bit carry.  */
-    /* ??? There is a 2011 VIS3 ADDXC insn that does take a 64-bit carry.  */
-
-    if (bh == TCG_REG_G0) {
+    if (use_vis3_instructions && !is_sub) {
+        /* Note that ADDXC doesn't accept immediates.  */
+        if (bhconst && bh != 0) {
+           tcg_out_movi(s, TCG_TYPE_I64, TCG_REG_T2, bh);
+           bh = TCG_REG_T2;
+        }
+        tcg_out_arith(s, rh, ah, bh, ARITH_ADDXC);
+    } else if (bh == TCG_REG_G0) {
 	/* If we have a zero, we can perform the operation in two insns,
            with the arithmetic first, and a conditional move into place.  */
 	if (rh == ah) {
@@ -1517,6 +1526,15 @@ static const TCGTargetOpDef sparc_op_defs[] = {
 
 static void tcg_target_init(TCGContext *s)
 {
+    /* Only probe for the platform and capabilities if we havn't already
+       determined maximum values at compile time.  */
+#ifndef use_vis3_instructions
+    {
+        unsigned long hwcap = qemu_getauxval(AT_HWCAP);
+        use_vis3_instructions = (hwcap & HWCAP_SPARC_VIS3) != 0;
+    }
+#endif
+
     tcg_regset_set32(tcg_target_available_regs[TCG_TYPE_I32], 0, 0xffffffff);
     tcg_regset_set32(tcg_target_available_regs[TCG_TYPE_I64], 0, ALL_64);
 
diff --git a/tcg/sparc/tcg-target.h b/tcg/sparc/tcg-target.h
index a44d34f..099b308 100644
--- a/tcg/sparc/tcg-target.h
+++ b/tcg/sparc/tcg-target.h
@@ -85,6 +85,12 @@ typedef enum {
 #define TCG_TARGET_EXTEND_ARGS 1
 #endif
 
+#if defined(__VIS__) && __VIS__ >= 0x300
+#define use_vis3_instructions  1
+#else
+extern bool use_vis3_instructions;
+#endif
+
 /* optional instructions */
 #define TCG_TARGET_HAS_div_i32		1
 #define TCG_TARGET_HAS_rem_i32		0
-- 
1.9.3

[Qemu-devel] [PATCH 3/6] tcg-sparc: Fix setcond_i32 uninitialized value

[Richard Henderson]

We failed to swap c1 and c2 correctly for NE c2 == 0.

Signed-off-by: Richard Henderson <rth@twiddle.net>
---
 tcg/sparc/tcg-target.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/tcg/sparc/tcg-target.c b/tcg/sparc/tcg-target.c
index 08ca482..3b232d6 100644
--- a/tcg/sparc/tcg-target.c
+++ b/tcg/sparc/tcg-target.c
@@ -674,9 +674,12 @@ static void tcg_out_setcond_i32(TCGContext *s, TCGCond cond, TCGReg ret,
     case TCG_COND_NE:
         /* For equality, we can transform to inequality vs zero.  */
         if (c2 != 0) {
-            tcg_out_arithc(s, ret, c1, c2, c2const, ARITH_XOR);
+            tcg_out_arithc(s, TCG_REG_T1, c1, c2, c2const, ARITH_XOR);
+            c2 = TCG_REG_T1;
+        } else {
+            c2 = c1;
         }
-        c1 = TCG_REG_G0, c2 = ret, c2const = 0;
+        c1 = TCG_REG_G0, c2const = 0;
         cond = (cond == TCG_COND_EQ ? TCG_COND_GEU : TCG_COND_LTU);
 	break;
 
-- 
1.9.3

[Qemu-devel] [PATCH 4/6] tcg-sparc: Use ADDXC in setcond_i64

[Richard Henderson]

Similar to the ADDC tricks we use in setcond_i32.

Signed-off-by: Richard Henderson <rth@twiddle.net>
---
 tcg/sparc/tcg-target.c | 17 +++++++++++++++++
 1 file changed, 17 insertions(+)

diff --git a/tcg/sparc/tcg-target.c b/tcg/sparc/tcg-target.c
index 3b232d6..d0bd08c 100644
--- a/tcg/sparc/tcg-target.c
+++ b/tcg/sparc/tcg-target.c
@@ -716,6 +716,23 @@ static void tcg_out_setcond_i32(TCGContext *s, TCGCond cond, TCGReg ret,
 static void tcg_out_setcond_i64(TCGContext *s, TCGCond cond, TCGReg ret,
                                 TCGReg c1, int32_t c2, int c2const)
 {
+    if (use_vis3_instructions) {
+        switch (cond) {
+        case TCG_COND_NE:
+            if (c2 != 0) {
+                break;
+            }
+            c2 = c1, c2const = 0, c1 = TCG_REG_G0;
+            /* FALLTHRU */
+        case TCG_COND_LTU:
+            tcg_out_cmp(s, c1, c2, c2const);
+            tcg_out_arith(s, ret, TCG_REG_G0, TCG_REG_G0, ARITH_ADDXC);
+            return;
+        default:
+            break;
+        }
+    }
+
     /* For 64-bit signed comparisons vs zero, we can avoid the compare
        if the input does not overlap the output.  */
     if (c2 == 0 && !is_unsigned_cond(cond) && c1 != ret) {
-- 
1.9.3

[Qemu-devel] [PATCH 5/6] tcg-sparc: Rename ADDX/SUBX insns

[Richard Henderson]

The pre-v9 ADDX/SUBX insns were renamed ADDC/SUBC for v9.
Standardizing on the v9 name makes things less confusing.

Signed-off-by: Richard Henderson <rth@twiddle.net>
---
 disas/sparc.c          | 32 +++++++++++---------------------
 tcg/sparc/tcg-target.c | 14 +++++++-------
 2 files changed, 18 insertions(+), 28 deletions(-)

diff --git a/disas/sparc.c b/disas/sparc.c
index 092e1b6..22ceac3 100644
--- a/disas/sparc.c
+++ b/disas/sparc.c
@@ -1175,15 +1175,11 @@ static const struct sparc_opcode sparc_opcodes[] = {
 { "subcc",      F3(2, 0x14, 0), F3(~2, ~0x14, ~0)|ASI(~0),      "1,2,d", 0, v6 },
 { "subcc",      F3(2, 0x14, 1), F3(~2, ~0x14, ~1),              "1,i,d", 0, v6 },
 
-{ "subx",       F3(2, 0x0c, 0), F3(~2, ~0x0c, ~0)|ASI(~0),      "1,2,d", 0, v6notv9 },
-{ "subx",       F3(2, 0x0c, 1), F3(~2, ~0x0c, ~1),              "1,i,d", 0, v6notv9 },
-{ "subc",       F3(2, 0x0c, 0), F3(~2, ~0x0c, ~0)|ASI(~0),      "1,2,d", 0, v9 },
-{ "subc",       F3(2, 0x0c, 1), F3(~2, ~0x0c, ~1),              "1,i,d", 0, v9 },
+{ "subc",       F3(2, 0x0c, 0), F3(~2, ~0x0c, ~0)|ASI(~0),      "1,2,d", 0, v6 },
+{ "subc",       F3(2, 0x0c, 1), F3(~2, ~0x0c, ~1),              "1,i,d", 0, v6 },
 
-{ "subxcc",     F3(2, 0x1c, 0), F3(~2, ~0x1c, ~0)|ASI(~0),      "1,2,d", 0, v6notv9 },
-{ "subxcc",     F3(2, 0x1c, 1), F3(~2, ~0x1c, ~1),              "1,i,d", 0, v6notv9 },
-{ "subccc",     F3(2, 0x1c, 0), F3(~2, ~0x1c, ~0)|ASI(~0),      "1,2,d", 0, v9 },
-{ "subccc",     F3(2, 0x1c, 1), F3(~2, ~0x1c, ~1),              "1,i,d", 0, v9 },
+{ "subccc",     F3(2, 0x1c, 0), F3(~2, ~0x1c, ~0)|ASI(~0),      "1,2,d", 0, v6 },
+{ "subccc",     F3(2, 0x1c, 1), F3(~2, ~0x1c, ~1),              "1,i,d", 0, v6 },
 
 { "and",        F3(2, 0x01, 0), F3(~2, ~0x01, ~0)|ASI(~0),      "1,2,d", 0, v6 },
 { "and",        F3(2, 0x01, 1), F3(~2, ~0x01, ~1),              "1,i,d", 0, v6 },
@@ -1215,19 +1211,13 @@ static const struct sparc_opcode sparc_opcodes[] = {
 { "addcc",      F3(2, 0x10, 1), F3(~2, ~0x10, ~1),              "1,i,d", 0, v6 },
 { "addcc",      F3(2, 0x10, 1), F3(~2, ~0x10, ~1),              "i,1,d", 0, v6 },
 
-{ "addx",       F3(2, 0x08, 0), F3(~2, ~0x08, ~0)|ASI(~0),      "1,2,d", 0, v6notv9 },
-{ "addx",       F3(2, 0x08, 1), F3(~2, ~0x08, ~1),              "1,i,d", 0, v6notv9 },
-{ "addx",       F3(2, 0x08, 1), F3(~2, ~0x08, ~1),              "i,1,d", 0, v6notv9 },
-{ "addc",       F3(2, 0x08, 0), F3(~2, ~0x08, ~0)|ASI(~0),      "1,2,d", 0, v9 },
-{ "addc",       F3(2, 0x08, 1), F3(~2, ~0x08, ~1),              "1,i,d", 0, v9 },
-{ "addc",       F3(2, 0x08, 1), F3(~2, ~0x08, ~1),              "i,1,d", 0, v9 },
-
-{ "addxcc",     F3(2, 0x18, 0), F3(~2, ~0x18, ~0)|ASI(~0),      "1,2,d", 0, v6notv9 },
-{ "addxcc",     F3(2, 0x18, 1), F3(~2, ~0x18, ~1),              "1,i,d", 0, v6notv9 },
-{ "addxcc",     F3(2, 0x18, 1), F3(~2, ~0x18, ~1),              "i,1,d", 0, v6notv9 },
-{ "addccc",     F3(2, 0x18, 0), F3(~2, ~0x18, ~0)|ASI(~0),      "1,2,d", 0, v9 },
-{ "addccc",     F3(2, 0x18, 1), F3(~2, ~0x18, ~1),              "1,i,d", 0, v9 },
-{ "addccc",     F3(2, 0x18, 1), F3(~2, ~0x18, ~1),              "i,1,d", 0, v9 },
+{ "addc",       F3(2, 0x08, 0), F3(~2, ~0x08, ~0)|ASI(~0),      "1,2,d", 0, v6 },
+{ "addc",       F3(2, 0x08, 1), F3(~2, ~0x08, ~1),              "1,i,d", 0, v6 },
+{ "addc",       F3(2, 0x08, 1), F3(~2, ~0x08, ~1),              "i,1,d", 0, v6 },
+
+{ "addccc",     F3(2, 0x18, 0), F3(~2, ~0x18, ~0)|ASI(~0),      "1,2,d", 0, v6 },
+{ "addccc",     F3(2, 0x18, 1), F3(~2, ~0x18, ~1),              "1,i,d", 0, v6 },
+{ "addccc",     F3(2, 0x18, 1), F3(~2, ~0x18, ~1),              "i,1,d", 0, v6 },
 
 { "smul",       F3(2, 0x0b, 0), F3(~2, ~0x0b, ~0)|ASI(~0),      "1,2,d", 0, v8 },
 { "smul",       F3(2, 0x0b, 1), F3(~2, ~0x0b, ~1),              "1,i,d", 0, v8 },
diff --git a/tcg/sparc/tcg-target.c b/tcg/sparc/tcg-target.c
index d0bd08c..0a8c26a 100644
--- a/tcg/sparc/tcg-target.c
+++ b/tcg/sparc/tcg-target.c
@@ -197,8 +197,8 @@ static const int tcg_target_call_oarg_regs[] = {
 #define ARITH_XOR  (INSN_OP(2) | INSN_OP3(0x03))
 #define ARITH_SUB  (INSN_OP(2) | INSN_OP3(0x04))
 #define ARITH_SUBCC (INSN_OP(2) | INSN_OP3(0x14))
-#define ARITH_ADDX (INSN_OP(2) | INSN_OP3(0x08))
-#define ARITH_SUBX (INSN_OP(2) | INSN_OP3(0x0c))
+#define ARITH_ADDC (INSN_OP(2) | INSN_OP3(0x08))
+#define ARITH_SUBC (INSN_OP(2) | INSN_OP3(0x0c))
 #define ARITH_UMUL (INSN_OP(2) | INSN_OP3(0x0a))
 #define ARITH_SMUL (INSN_OP(2) | INSN_OP3(0x0b))
 #define ARITH_UDIV (INSN_OP(2) | INSN_OP3(0x0e))
@@ -663,7 +663,7 @@ static void tcg_out_movcond_i64(TCGContext *s, TCGCond cond, TCGReg ret,
 static void tcg_out_setcond_i32(TCGContext *s, TCGCond cond, TCGReg ret,
                                 TCGReg c1, int32_t c2, int c2const)
 {
-    /* For 32-bit comparisons, we can play games with ADDX/SUBX.  */
+    /* For 32-bit comparisons, we can play games with ADDC/SUBC.  */
     switch (cond) {
     case TCG_COND_LTU:
     case TCG_COND_GEU:
@@ -707,9 +707,9 @@ static void tcg_out_setcond_i32(TCGContext *s, TCGCond cond, TCGReg ret,
 
     tcg_out_cmp(s, c1, c2, c2const);
     if (cond == TCG_COND_LTU) {
-        tcg_out_arithi(s, ret, TCG_REG_G0, 0, ARITH_ADDX);
+        tcg_out_arithi(s, ret, TCG_REG_G0, 0, ARITH_ADDC);
     } else {
-        tcg_out_arithi(s, ret, TCG_REG_G0, -1, ARITH_SUBX);
+        tcg_out_arithi(s, ret, TCG_REG_G0, -1, ARITH_SUBC);
     }
 }
 
@@ -1340,12 +1340,12 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,
     case INDEX_op_add2_i32:
         tcg_out_addsub2_i32(s, args[0], args[1], args[2], args[3],
                             args[4], const_args[4], args[5], const_args[5],
-                            ARITH_ADDCC, ARITH_ADDX);
+                            ARITH_ADDCC, ARITH_ADDC);
         break;
     case INDEX_op_sub2_i32:
         tcg_out_addsub2_i32(s, args[0], args[1], args[2], args[3],
                             args[4], const_args[4], args[5], const_args[5],
-                            ARITH_SUBCC, ARITH_SUBX);
+                            ARITH_SUBCC, ARITH_SUBC);
         break;
     case INDEX_op_mulu2_i32:
         c = ARITH_UMUL;
-- 
1.9.3

[Qemu-devel] [PATCH 6/6] tcg-sparc: Use UMULXHI instruction

[Richard Henderson]

Signed-off-by: Richard Henderson <rth@twiddle.net>
---
 disas/sparc.c          | 1 +
 tcg/sparc/tcg-target.c | 5 +++++
 tcg/sparc/tcg-target.h | 2 +-
 3 files changed, 7 insertions(+), 1 deletion(-)

diff --git a/disas/sparc.c b/disas/sparc.c
index 22ceac3..8e755d1 100644
--- a/disas/sparc.c
+++ b/disas/sparc.c
@@ -2034,6 +2034,7 @@ IMPDEP ("impdep2", 0x37),
 
 { "addxc", F3F(2, 0x36, 0x011), F3F(~2, ~0x36, ~0x011), "1,2,d", 0, v9b },
 { "addxccc", F3F(2, 0x36, 0x013), F3F(~2, ~0x36, ~0x013), "1,2,d", 0, v9b },
+{ "umulxhi", F3F(2, 0x36, 0x016), F3F(~2, ~0x36, ~0x016), "1,2,d", 0, v9b },
 
 };
 
diff --git a/tcg/sparc/tcg-target.c b/tcg/sparc/tcg-target.c
index 0a8c26a..0c4b028 100644
--- a/tcg/sparc/tcg-target.c
+++ b/tcg/sparc/tcg-target.c
@@ -210,6 +210,7 @@ static const int tcg_target_call_oarg_regs[] = {
 #define ARITH_MOVR (INSN_OP(2) | INSN_OP3(0x2f))
 
 #define ARITH_ADDXC (INSN_OP(2) | INSN_OP3(0x36) | INSN_OPF(0x11))
+#define ARITH_UMULXHI (INSN_OP(2) | INSN_OP3(0x36) | INSN_OPF(0x16))
 
 #define SHIFT_SLL  (INSN_OP(2) | INSN_OP3(0x25))
 #define SHIFT_SRL  (INSN_OP(2) | INSN_OP3(0x26))
@@ -1435,6 +1436,9 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc,
         tcg_out_addsub2_i64(s, args[0], args[1], args[2], args[3], args[4],
                             const_args[4], args[5], const_args[5], true);
         break;
+    case INDEX_op_muluh_i64:
+        tcg_out_arith(s, args[0], args[1], args[2], ARITH_UMULXHI);
+        break;
 
     gen_arith:
         tcg_out_arithc(s, a0, a1, a2, c2, c);
@@ -1535,6 +1539,7 @@ static const TCGTargetOpDef sparc_op_defs[] = {
 
     { INDEX_op_add2_i64, { "R", "R", "RZ", "RZ", "RJ", "RI" } },
     { INDEX_op_sub2_i64, { "R", "R", "RZ", "RZ", "RJ", "RI" } },
+    { INDEX_op_muluh_i64, { "R", "RZ", "RZ" } },
 
     { INDEX_op_qemu_ld_i32, { "r", "A" } },
     { INDEX_op_qemu_ld_i64, { "R", "A" } },
diff --git a/tcg/sparc/tcg-target.h b/tcg/sparc/tcg-target.h
index 099b308..0c4c8af 100644
--- a/tcg/sparc/tcg-target.h
+++ b/tcg/sparc/tcg-target.h
@@ -143,7 +143,7 @@ extern bool use_vis3_instructions;
 #define TCG_TARGET_HAS_sub2_i64         1
 #define TCG_TARGET_HAS_mulu2_i64        0
 #define TCG_TARGET_HAS_muls2_i64        0
-#define TCG_TARGET_HAS_muluh_i64        0
+#define TCG_TARGET_HAS_muluh_i64        use_vis3_instructions
 #define TCG_TARGET_HAS_mulsh_i64        0
 
 #define TCG_AREG0 TCG_REG_I0
-- 
1.9.3