[PATCH 00/22] target/sparc: floating-point cleanup

[PATCH 00/22] target/sparc: floating-point cleanup

[Richard Henderson]

Major changes: 

(1) Get rid of the env->qt[01] temporaries and use TCGv_i128 for float128.
(2) Perform ieee exception check within the helpers, before any writeback
    to the floating point registers.
(3) Split env->fsr into pieces to simplify update, especially compares.


r~


Based-on: 20231101041132.174501-1-richard.henderson@linaro.org
("[PATCH v2 00/21] target/sparc: Cleanup condition codes etc")


Richard Henderson (22):
  target/sparc: Use tcg_gen_qemu_{ld,st}_i128 for ASI_M_BCOPY
  target/sparc: Use tcg_gen_qemu_{ld,st}_i128 for ASI_M_BFILL
  target/sparc: Remove gen_dest_fpr_F
  target/sparc: Introduce gen_{load,store}_fpr_Q
  target/sparc: Inline FNEG, FABS
  target/sparc: Use i128 for FSQRTq
  target/sparc: Use i128 for FADDq, FSUBq, FMULq, FDIVq
  target/sparc: Use i128 for FqTOs, FqTOi
  target/sparc: Use i128 for FqTOd, FqTOx
  target/sparc: Use i128 for FCMPq, FCMPEq
  target/sparc: Use i128 for FsTOq, FiTOq
  target/sparc: Use i128 for FdTOq, FxTOq
  target/sparc: Use i128 for Fdmulq
  target/sparc: Remove qt0, qt1 temporaries
  target/sparc: Introduce cpu_get_fsr, cpu_put_fsr
  target/split: Split ver from env->fsr
  target/sparc: Clear cexc and ftt in do_check_ieee_exceptions
  target/sparc: Merge check_ieee_exceptions with FPop helpers
  target/sparc: Split cexc and ftt from env->fsr
  target/sparc: Remove cpu_fsr
  target/sparc: Split fcc out of env->fsr
  target/sparc: Remove FSR_FTT_NMASK, FSR_FTT_CEXC_NMASK

 target/sparc/cpu.h          |  39 +-
 target/sparc/helper.h       | 116 ++----
 linux-user/sparc/cpu_loop.c |   2 +-
 linux-user/sparc/signal.c   |  14 +-
 target/sparc/cpu.c          |  32 +-
 target/sparc/fop_helper.c   | 510 +++++++++++++----------
 target/sparc/gdbstub.c      |   8 +-
 target/sparc/ldst_helper.c  |   3 -
 target/sparc/machine.c      |  38 +-
 target/sparc/translate.c    | 799 ++++++++++++------------------------
 10 files changed, 680 insertions(+), 881 deletions(-)

-- 
2.34.1

[PATCH 01/22] target/sparc: Use tcg_gen_qemu_{ld, st}_i128 for ASI_M_BCOPY

[Richard Henderson]

Align the operation to the 32-byte cacheline.
Use 2 pair of i128 instead of 8 pair of i32.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/sparc/translate.c | 43 +++++++++++++++++++++++-----------------
 1 file changed, 25 insertions(+), 18 deletions(-)

diff --git a/target/sparc/translate.c b/target/sparc/translate.c
index 6fc333a6b8..5d55856a54 100644
--- a/target/sparc/translate.c
+++ b/target/sparc/translate.c
@@ -1727,28 +1727,35 @@ static void gen_st_asi(DisasContext *dc, DisasASI *da, TCGv src, TCGv addr)
 
     case GET_ASI_BCOPY:
         assert(TARGET_LONG_BITS == 32);
-        /* Copy 32 bytes from the address in SRC to ADDR.  */
-        /* ??? The original qemu code suggests 4-byte alignment, dropping
-           the low bits, but the only place I can see this used is in the
-           Linux kernel with 32 byte alignment, which would make more sense
-           as a cacheline-style operation.  */
+        /*
+         * Copy 32 bytes from the address in SRC to ADDR.
+         *
+         * From Ross RT625 hyperSPARC manual, section 4.6:
+         * "Block Copy and Block Fill will work only on cache line boundaries."
+         *
+         * It does not specify if an unaliged address is truncated or trapped.
+         * Previous qemu behaviour was to truncate to 4 byte alignment, which
+         * is obviously wrong.  The only place I can see this used is in the
+         * Linux kernel which begins with page alignment, advancing by 32,
+         * so is always aligned.  Assume truncation as the simpler option.
+         *
+         * Since the loads and stores are paired, allow the copy to happen
+         * in the host endianness.  The copy need not be atomic.
+         */
         {
+            MemOp mop = MO_128 | MO_ATOM_IFALIGN_PAIR;
             TCGv saddr = tcg_temp_new();
             TCGv daddr = tcg_temp_new();
-            TCGv four = tcg_constant_tl(4);
-            TCGv_i32 tmp = tcg_temp_new_i32();
-            int i;
+            TCGv_i128 tmp = tcg_temp_new_i128();
 
-            tcg_gen_andi_tl(saddr, src, -4);
-            tcg_gen_andi_tl(daddr, addr, -4);
-            for (i = 0; i < 32; i += 4) {
-                /* Since the loads and stores are paired, allow the
-                   copy to happen in the host endianness.  */
-                tcg_gen_qemu_ld_i32(tmp, saddr, da->mem_idx, MO_UL);
-                tcg_gen_qemu_st_i32(tmp, daddr, da->mem_idx, MO_UL);
-                tcg_gen_add_tl(saddr, saddr, four);
-                tcg_gen_add_tl(daddr, daddr, four);
-            }
+            tcg_gen_andi_tl(saddr, src, -32);
+            tcg_gen_andi_tl(daddr, addr, -32);
+            tcg_gen_qemu_ld_i128(tmp, saddr, da->mem_idx, mop);
+            tcg_gen_qemu_st_i128(tmp, daddr, da->mem_idx, mop);
+            tcg_gen_addi_tl(saddr, saddr, 16);
+            tcg_gen_addi_tl(daddr, daddr, 16);
+            tcg_gen_qemu_ld_i128(tmp, saddr, da->mem_idx, mop);
+            tcg_gen_qemu_st_i128(tmp, daddr, da->mem_idx, mop);
         }
         break;
 
-- 
2.34.1

[PATCH 02/22] target/sparc: Use tcg_gen_qemu_{ld, st}_i128 for ASI_M_BFILL

[Richard Henderson]

Align the operation to the 32-byte cacheline.
Use 2 i128 instead of 4 i64.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/sparc/translate.c | 29 ++++++++++++++---------------
 1 file changed, 14 insertions(+), 15 deletions(-)

diff --git a/target/sparc/translate.c b/target/sparc/translate.c
index 5d55856a54..713ac5bbae 100644
--- a/target/sparc/translate.c
+++ b/target/sparc/translate.c
@@ -2172,23 +2172,22 @@ static void gen_stda_asi(DisasContext *dc, DisasASI *da, TCGv addr, int rd)
 
     case GET_ASI_BFILL:
         assert(TARGET_LONG_BITS == 32);
-        /* Store 32 bytes of T64 to ADDR.  */
-        /* ??? The original qemu code suggests 8-byte alignment, dropping
-           the low bits, but the only place I can see this used is in the
-           Linux kernel with 32 byte alignment, which would make more sense
-           as a cacheline-style operation.  */
+        /*
+         * Store 32 bytes of [rd:rd+1] to ADDR.
+         * See comments for GET_ASI_COPY above.
+         */
         {
-            TCGv_i64 t64 = tcg_temp_new_i64();
-            TCGv d_addr = tcg_temp_new();
-            TCGv eight = tcg_constant_tl(8);
-            int i;
+            MemOp mop = MO_TE | MO_128 | MO_ATOM_IFALIGN_PAIR;
+            TCGv_i64 t8 = tcg_temp_new_i64();
+            TCGv_i128 t16 = tcg_temp_new_i128();
+            TCGv daddr = tcg_temp_new();
 
-            tcg_gen_concat_tl_i64(t64, lo, hi);
-            tcg_gen_andi_tl(d_addr, addr, -8);
-            for (i = 0; i < 32; i += 8) {
-                tcg_gen_qemu_st_i64(t64, d_addr, da->mem_idx, da->memop);
-                tcg_gen_add_tl(d_addr, d_addr, eight);
-            }
+            tcg_gen_concat_tl_i64(t8, lo, hi);
+            tcg_gen_concat_i64_i128(t16, t8, t8);
+            tcg_gen_andi_tl(daddr, addr, -32);
+            tcg_gen_qemu_st_i128(t16, daddr, da->mem_idx, mop);
+            tcg_gen_addi_tl(daddr, daddr, 16);
+            tcg_gen_qemu_st_i128(t16, daddr, da->mem_idx, mop);
         }
         break;
 
-- 
2.34.1

[PATCH 03/22] target/sparc: Remove gen_dest_fpr_F

[Richard Henderson]

Replace with tcg_temp_new_i32.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/sparc/translate.c | 17 ++++++-----------
 1 file changed, 6 insertions(+), 11 deletions(-)

diff --git a/target/sparc/translate.c b/target/sparc/translate.c
index 713ac5bbae..cdc85be807 100644
--- a/target/sparc/translate.c
+++ b/target/sparc/translate.c
@@ -246,11 +246,6 @@ static void gen_store_fpr_F(DisasContext *dc, unsigned int dst, TCGv_i32 v)
     gen_update_fprs_dirty(dc, dst);
 }
 
-static TCGv_i32 gen_dest_fpr_F(DisasContext *dc)
-{
-    return tcg_temp_new_i32();
-}
-
 static TCGv_i64 gen_load_fpr_D(DisasContext *dc, unsigned int src)
 {
     src = DFPREG(src);
@@ -1873,7 +1868,7 @@ static void gen_ldf_asi(DisasContext *dc, DisasASI *da, MemOp orig_size,
         memop |= MO_ALIGN_4;
         switch (size) {
         case MO_32:
-            d32 = gen_dest_fpr_F(dc);
+            d32 = tcg_temp_new_i32();
             tcg_gen_qemu_ld_i32(d32, addr, da->mem_idx, memop);
             gen_store_fpr_F(dc, rd, d32);
             break;
@@ -1938,7 +1933,7 @@ static void gen_ldf_asi(DisasContext *dc, DisasASI *da, MemOp orig_size,
             case MO_32:
                 d64 = tcg_temp_new_i64();
                 gen_helper_ld_asi(d64, tcg_env, addr, r_asi, r_mop);
-                d32 = gen_dest_fpr_F(dc);
+                d32 = tcg_temp_new_i32();
                 tcg_gen_extrl_i64_i32(d32, d64);
                 gen_store_fpr_F(dc, rd, d32);
                 break;
@@ -2228,7 +2223,7 @@ static void gen_fmovs(DisasContext *dc, DisasCompare *cmp, int rd, int rs)
 
     s1 = gen_load_fpr_F(dc, rs);
     s2 = gen_load_fpr_F(dc, rd);
-    dst = gen_dest_fpr_F(dc);
+    dst = tcg_temp_new_i32();
     zero = tcg_constant_i32(0);
 
     tcg_gen_movcond_i32(TCG_COND_NE, dst, c32, zero, s1, s2);
@@ -4497,7 +4492,7 @@ static bool do_fd(DisasContext *dc, arg_r_r *a,
         return true;
     }
 
-    dst = gen_dest_fpr_F(dc);
+    dst = tcg_temp_new_i32();
     src = gen_load_fpr_D(dc, a->rs);
     func(dst, src);
     gen_store_fpr_F(dc, a->rd, dst);
@@ -4539,7 +4534,7 @@ static bool do_env_fd(DisasContext *dc, arg_r_r *a,
     }
 
     gen_op_clear_ieee_excp_and_FTT();
-    dst = gen_dest_fpr_F(dc);
+    dst = tcg_temp_new_i32();
     src = gen_load_fpr_D(dc, a->rs);
     func(dst, tcg_env, src);
     gen_helper_check_ieee_exceptions(cpu_fsr, tcg_env);
@@ -4697,7 +4692,7 @@ static bool do_env_fq(DisasContext *dc, arg_r_r *a,
 
     gen_op_clear_ieee_excp_and_FTT();
     gen_op_load_fpr_QT1(QFPREG(a->rs));
-    dst = gen_dest_fpr_F(dc);
+    dst = tcg_temp_new_i32();
     func(dst, tcg_env);
     gen_helper_check_ieee_exceptions(cpu_fsr, tcg_env);
     gen_store_fpr_F(dc, a->rd, dst);
-- 
2.34.1

Re: [PATCH 03/22] target/sparc: Remove gen_dest_fpr_F

[Philippe Mathieu-Daudé]

On 3/11/23 18:38, Richard Henderson wrote:
> Replace with tcg_temp_new_i32.
> 
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
>   target/sparc/translate.c | 17 ++++++-----------
>   1 file changed, 6 insertions(+), 11 deletions(-)

Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>

[PATCH 04/22] target/sparc: Introduce gen_{load,store}_fpr_Q

[Richard Henderson]

Use them for trans_FMOVq.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/sparc/translate.c | 25 +++++++++++++++++++------
 1 file changed, 19 insertions(+), 6 deletions(-)

diff --git a/target/sparc/translate.c b/target/sparc/translate.c
index cdc85be807..0e494d3ebd 100644
--- a/target/sparc/translate.c
+++ b/target/sparc/translate.c
@@ -264,6 +264,22 @@ static TCGv_i64 gen_dest_fpr_D(DisasContext *dc, unsigned int dst)
     return cpu_fpr[DFPREG(dst) / 2];
 }
 
+static TCGv_i128 gen_load_fpr_Q(DisasContext *dc, unsigned int src)
+{
+    TCGv_i128 ret = tcg_temp_new_i128();
+
+    src = QFPREG(src);
+    tcg_gen_concat_i64_i128(ret, cpu_fpr[src / 2 + 1], cpu_fpr[src / 2]);
+    return ret;
+}
+
+static void gen_store_fpr_Q(DisasContext *dc, unsigned int dst, TCGv_i128 v)
+{
+    dst = DFPREG(dst);
+    tcg_gen_extr_i128_i64(cpu_fpr[dst / 2 + 1], cpu_fpr[dst / 2], v);
+    gen_update_fprs_dirty(dc, dst);
+}
+
 static void gen_op_load_fpr_QT0(unsigned int src)
 {
     tcg_gen_st_i64(cpu_fpr[src / 2], tcg_env, offsetof(CPUSPARCState, qt0) +
@@ -4615,7 +4631,7 @@ TRANS(FsTOx, 64, do_env_df, a, gen_helper_fstox)
 
 static bool trans_FMOVq(DisasContext *dc, arg_FMOVq *a)
 {
-    int rd, rs;
+    TCGv_i128 t;
 
     if (!avail_64(dc)) {
         return false;
@@ -4628,11 +4644,8 @@ static bool trans_FMOVq(DisasContext *dc, arg_FMOVq *a)
     }
 
     gen_op_clear_ieee_excp_and_FTT();
-    rd = QFPREG(a->rd);
-    rs = QFPREG(a->rs);
-    tcg_gen_mov_i64(cpu_fpr[rd / 2], cpu_fpr[rs / 2]);
-    tcg_gen_mov_i64(cpu_fpr[rd / 2 + 1], cpu_fpr[rs / 2 + 1]);
-    gen_update_fprs_dirty(dc, rd);
+    t = gen_load_fpr_Q(dc, a->rs);
+    gen_store_fpr_Q(dc, a->rd, t);
     return advance_pc(dc);
 }
 
-- 
2.34.1

Re: [PATCH 04/22] target/sparc: Introduce gen_{load,store}_fpr_Q

[Philippe Mathieu-Daudé]

On 3/11/23 18:38, Richard Henderson wrote:
> Use them for trans_FMOVq.
> 
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
>   target/sparc/translate.c | 25 +++++++++++++++++++------
>   1 file changed, 19 insertions(+), 6 deletions(-)

Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>

[PATCH 05/22] target/sparc: Inline FNEG, FABS

[Richard Henderson]

These are simple bit manipulation insns.
Begin using i128 for float128.
Implement FMOVq with do_qq.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/sparc/helper.h     |  6 ----
 target/sparc/fop_helper.c | 34 ---------------------
 target/sparc/translate.c  | 62 +++++++++++++++++++--------------------
 3 files changed, 30 insertions(+), 72 deletions(-)

diff --git a/target/sparc/helper.h b/target/sparc/helper.h
index 55eff66283..74a1575d21 100644
--- a/target/sparc/helper.h
+++ b/target/sparc/helper.h
@@ -37,7 +37,6 @@ DEF_HELPER_FLAGS_5(st_asi, TCG_CALL_NO_WG, void, env, tl, i64, int, i32)
 #endif
 DEF_HELPER_FLAGS_1(check_ieee_exceptions, TCG_CALL_NO_WG, tl, env)
 DEF_HELPER_FLAGS_2(set_fsr, TCG_CALL_NO_RWG, void, env, tl)
-DEF_HELPER_FLAGS_1(fabss, TCG_CALL_NO_RWG_SE, f32, f32)
 DEF_HELPER_FLAGS_2(fsqrts, TCG_CALL_NO_RWG, f32, env, f32)
 DEF_HELPER_FLAGS_2(fsqrtd, TCG_CALL_NO_RWG, f64, env, f64)
 DEF_HELPER_FLAGS_3(fcmps, TCG_CALL_NO_WG, tl, env, f32, f32)
@@ -48,7 +47,6 @@ DEF_HELPER_FLAGS_1(fsqrtq, TCG_CALL_NO_RWG, void, env)
 DEF_HELPER_FLAGS_1(fcmpq, TCG_CALL_NO_WG, tl, env)
 DEF_HELPER_FLAGS_1(fcmpeq, TCG_CALL_NO_WG, tl, env)
 #ifdef TARGET_SPARC64
-DEF_HELPER_FLAGS_1(fabsd, TCG_CALL_NO_RWG_SE, f64, f64)
 DEF_HELPER_FLAGS_3(fcmps_fcc1, TCG_CALL_NO_WG, tl, env, f32, f32)
 DEF_HELPER_FLAGS_3(fcmps_fcc2, TCG_CALL_NO_WG, tl, env, f32, f32)
 DEF_HELPER_FLAGS_3(fcmps_fcc3, TCG_CALL_NO_WG, tl, env, f32, f32)
@@ -61,7 +59,6 @@ DEF_HELPER_FLAGS_3(fcmpes_fcc3, TCG_CALL_NO_WG, tl, env, f32, f32)
 DEF_HELPER_FLAGS_3(fcmped_fcc1, TCG_CALL_NO_WG, tl, env, f64, f64)
 DEF_HELPER_FLAGS_3(fcmped_fcc2, TCG_CALL_NO_WG, tl, env, f64, f64)
 DEF_HELPER_FLAGS_3(fcmped_fcc3, TCG_CALL_NO_WG, tl, env, f64, f64)
-DEF_HELPER_FLAGS_1(fabsq, TCG_CALL_NO_RWG, void, env)
 DEF_HELPER_FLAGS_1(fcmpq_fcc1, TCG_CALL_NO_WG, tl, env)
 DEF_HELPER_FLAGS_1(fcmpq_fcc2, TCG_CALL_NO_WG, tl, env)
 DEF_HELPER_FLAGS_1(fcmpq_fcc3, TCG_CALL_NO_WG, tl, env)
@@ -90,15 +87,12 @@ DEF_HELPER_FLAGS_3(fdivs, TCG_CALL_NO_RWG, f32, env, f32, f32)
 DEF_HELPER_FLAGS_3(fsmuld, TCG_CALL_NO_RWG, f64, env, f32, f32)
 DEF_HELPER_FLAGS_3(fdmulq, TCG_CALL_NO_RWG, void, env, f64, f64)
 
-DEF_HELPER_FLAGS_1(fnegs, TCG_CALL_NO_RWG_SE, f32, f32)
 DEF_HELPER_FLAGS_2(fitod, TCG_CALL_NO_RWG_SE, f64, env, s32)
 DEF_HELPER_FLAGS_2(fitoq, TCG_CALL_NO_RWG, void, env, s32)
 
 DEF_HELPER_FLAGS_2(fitos, TCG_CALL_NO_RWG, f32, env, s32)
 
 #ifdef TARGET_SPARC64
-DEF_HELPER_FLAGS_1(fnegd, TCG_CALL_NO_RWG_SE, f64, f64)
-DEF_HELPER_FLAGS_1(fnegq, TCG_CALL_NO_RWG, void, env)
 DEF_HELPER_FLAGS_2(fxtos, TCG_CALL_NO_RWG, f32, env, s64)
 DEF_HELPER_FLAGS_2(fxtod, TCG_CALL_NO_RWG, f64, env, s64)
 DEF_HELPER_FLAGS_2(fxtoq, TCG_CALL_NO_RWG, void, env, s64)
diff --git a/target/sparc/fop_helper.c b/target/sparc/fop_helper.c
index 0f8aa3abcd..d6fb769769 100644
--- a/target/sparc/fop_helper.c
+++ b/target/sparc/fop_helper.c
@@ -114,23 +114,6 @@ void helper_fdmulq(CPUSPARCState *env, float64 src1, float64 src2)
                        &env->fp_status);
 }
 
-float32 helper_fnegs(float32 src)
-{
-    return float32_chs(src);
-}
-
-#ifdef TARGET_SPARC64
-float64 helper_fnegd(float64 src)
-{
-    return float64_chs(src);
-}
-
-F_HELPER(neg, q)
-{
-    QT0 = float128_chs(QT1);
-}
-#endif
-
 /* Integer to float conversion.  */
 float32 helper_fitos(CPUSPARCState *env, int32_t src)
 {
@@ -229,23 +212,6 @@ int64_t helper_fqtox(CPUSPARCState *env)
 }
 #endif
 
-float32 helper_fabss(float32 src)
-{
-    return float32_abs(src);
-}
-
-#ifdef TARGET_SPARC64
-float64 helper_fabsd(float64 src)
-{
-    return float64_abs(src);
-}
-
-void helper_fabsq(CPUSPARCState *env)
-{
-    QT0 = float128_abs(QT1);
-}
-#endif
-
 float32 helper_fsqrts(CPUSPARCState *env, float32 src)
 {
     return float32_sqrt(src, &env->fp_status);
diff --git a/target/sparc/translate.c b/target/sparc/translate.c
index 0e494d3ebd..254f185b83 100644
--- a/target/sparc/translate.c
+++ b/target/sparc/translate.c
@@ -43,9 +43,7 @@
 #else
 # define gen_helper_clear_softint(E, S)         qemu_build_not_reached()
 # define gen_helper_done(E)                     qemu_build_not_reached()
-# define gen_helper_fabsd(D, S)                 qemu_build_not_reached()
 # define gen_helper_flushw(E)                   qemu_build_not_reached()
-# define gen_helper_fnegd(D, S)                 qemu_build_not_reached()
 # define gen_helper_rdccr(D, E)                 qemu_build_not_reached()
 # define gen_helper_rdcwp(D, E)                 qemu_build_not_reached()
 # define gen_helper_restored(E)                 qemu_build_not_reached()
@@ -61,7 +59,6 @@
 # define gen_helper_write_softint(E, S)         qemu_build_not_reached()
 # define gen_helper_wrpil(E, S)                 qemu_build_not_reached()
 # define gen_helper_wrpstate(E, S)              qemu_build_not_reached()
-# define gen_helper_fabsq                ({ qemu_build_not_reached(); NULL; })
 # define gen_helper_fcmpeq16             ({ qemu_build_not_reached(); NULL; })
 # define gen_helper_fcmpeq32             ({ qemu_build_not_reached(); NULL; })
 # define gen_helper_fcmpgt16             ({ qemu_build_not_reached(); NULL; })
@@ -79,7 +76,6 @@
 # define gen_helper_fmul8x16             ({ qemu_build_not_reached(); NULL; })
 # define gen_helper_fmuld8sux16          ({ qemu_build_not_reached(); NULL; })
 # define gen_helper_fmuld8ulx16          ({ qemu_build_not_reached(); NULL; })
-# define gen_helper_fnegq                ({ qemu_build_not_reached(); NULL; })
 # define gen_helper_fpmerge              ({ qemu_build_not_reached(); NULL; })
 # define gen_helper_fqtox                ({ qemu_build_not_reached(); NULL; })
 # define gen_helper_fstox                ({ qemu_build_not_reached(); NULL; })
@@ -1239,13 +1235,13 @@ static void gen_op_fmovs(TCGv_i32 dst, TCGv_i32 src)
 static void gen_op_fnegs(TCGv_i32 dst, TCGv_i32 src)
 {
     gen_op_clear_ieee_excp_and_FTT();
-    gen_helper_fnegs(dst, src);
+    tcg_gen_xori_i32(dst, src, 1u << 31);
 }
 
 static void gen_op_fabss(TCGv_i32 dst, TCGv_i32 src)
 {
     gen_op_clear_ieee_excp_and_FTT();
-    gen_helper_fabss(dst, src);
+    tcg_gen_andi_i32(dst, src, ~(1u << 31));
 }
 
 static void gen_op_fmovd(TCGv_i64 dst, TCGv_i64 src)
@@ -1257,13 +1253,33 @@ static void gen_op_fmovd(TCGv_i64 dst, TCGv_i64 src)
 static void gen_op_fnegd(TCGv_i64 dst, TCGv_i64 src)
 {
     gen_op_clear_ieee_excp_and_FTT();
-    gen_helper_fnegd(dst, src);
+    tcg_gen_xori_i64(dst, src, 1ull << 63);
 }
 
 static void gen_op_fabsd(TCGv_i64 dst, TCGv_i64 src)
 {
     gen_op_clear_ieee_excp_and_FTT();
-    gen_helper_fabsd(dst, src);
+    tcg_gen_andi_i64(dst, src, ~(1ull << 63));
+}
+
+static void gen_op_fnegq(TCGv_i128 dst, TCGv_i128 src)
+{
+    TCGv_i64 l = tcg_temp_new_i64();
+    TCGv_i64 h = tcg_temp_new_i64();
+
+    tcg_gen_extr_i128_i64(l, h, src);
+    tcg_gen_xori_i64(h, h, 1ull << 63);
+    tcg_gen_concat_i64_i128(dst, l, h);
+}
+
+static void gen_op_fabsq(TCGv_i128 dst, TCGv_i128 src)
+{
+    TCGv_i64 l = tcg_temp_new_i64();
+    TCGv_i64 h = tcg_temp_new_i64();
+
+    tcg_gen_extr_i128_i64(l, h, src);
+    tcg_gen_andi_i64(h, h, ~(1ull << 63));
+    tcg_gen_concat_i64_i128(dst, l, h);
 }
 
 #ifdef TARGET_SPARC64
@@ -4629,13 +4645,11 @@ TRANS(FiTOd, ALL, do_env_df, a, gen_helper_fitod)
 TRANS(FsTOd, ALL, do_env_df, a, gen_helper_fstod)
 TRANS(FsTOx, 64, do_env_df, a, gen_helper_fstox)
 
-static bool trans_FMOVq(DisasContext *dc, arg_FMOVq *a)
+static bool do_qq(DisasContext *dc, arg_r_r *a,
+                  void (*func)(TCGv_i128, TCGv_i128))
 {
     TCGv_i128 t;
 
-    if (!avail_64(dc)) {
-        return false;
-    }
     if (gen_trap_ifnofpu(dc)) {
         return true;
     }
@@ -4645,30 +4659,14 @@ static bool trans_FMOVq(DisasContext *dc, arg_FMOVq *a)
 
     gen_op_clear_ieee_excp_and_FTT();
     t = gen_load_fpr_Q(dc, a->rs);
+    func(t, t);
     gen_store_fpr_Q(dc, a->rd, t);
     return advance_pc(dc);
 }
 
-static bool do_qq(DisasContext *dc, arg_r_r *a,
-                  void (*func)(TCGv_env))
-{
-    if (gen_trap_ifnofpu(dc)) {
-        return true;
-    }
-    if (gen_trap_float128(dc)) {
-        return true;
-    }
-
-    gen_op_clear_ieee_excp_and_FTT();
-    gen_op_load_fpr_QT1(QFPREG(a->rs));
-    func(tcg_env);
-    gen_op_store_QT0_fpr(QFPREG(a->rd));
-    gen_update_fprs_dirty(dc, QFPREG(a->rd));
-    return advance_pc(dc);
-}
-
-TRANS(FNEGq, 64, do_qq, a, gen_helper_fnegq)
-TRANS(FABSq, 64, do_qq, a, gen_helper_fabsq)
+TRANS(FMOVq, 64, do_qq, a, tcg_gen_mov_i128)
+TRANS(FNEGq, 64, do_qq, a, gen_op_fnegq)
+TRANS(FABSq, 64, do_qq, a, gen_op_fabsq)
 
 static bool do_env_qq(DisasContext *dc, arg_r_r *a,
                        void (*func)(TCGv_env))
-- 
2.34.1

Re: [PATCH 05/22] target/sparc: Inline FNEG, FABS

[Philippe Mathieu-Daudé]

On 3/11/23 18:38, Richard Henderson wrote:
> These are simple bit manipulation insns.
> Begin using i128 for float128.
> Implement FMOVq with do_qq.
> 
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
>   target/sparc/helper.h     |  6 ----
>   target/sparc/fop_helper.c | 34 ---------------------
>   target/sparc/translate.c  | 62 +++++++++++++++++++--------------------
>   3 files changed, 30 insertions(+), 72 deletions(-)

Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>

Re: [PATCH 05/22] target/sparc: Inline FNEG, FABS

[Philippe Mathieu-Daudé]

Hi Richard,

On 3/11/23 18:38, Richard Henderson wrote:
> These are simple bit manipulation insns.
> Begin using i128 for float128.
> Implement FMOVq with do_qq.
> 
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
>   target/sparc/helper.h     |  6 ----
>   target/sparc/fop_helper.c | 34 ---------------------
>   target/sparc/translate.c  | 62 +++++++++++++++++++--------------------
>   3 files changed, 30 insertions(+), 72 deletions(-)


> @@ -1239,13 +1235,13 @@ static void gen_op_fmovs(TCGv_i32 dst, TCGv_i32 src)
>   static void gen_op_fnegs(TCGv_i32 dst, TCGv_i32 src)
>   {
>       gen_op_clear_ieee_excp_and_FTT();
> -    gen_helper_fnegs(dst, src);
> +    tcg_gen_xori_i32(dst, src, 1u << 31);
>   }
>   
>   static void gen_op_fabss(TCGv_i32 dst, TCGv_i32 src)
>   {
>       gen_op_clear_ieee_excp_and_FTT();
> -    gen_helper_fabss(dst, src);
> +    tcg_gen_andi_i32(dst, src, ~(1u << 31));
>   }
>   
>   static void gen_op_fmovd(TCGv_i64 dst, TCGv_i64 src)
> @@ -1257,13 +1253,33 @@ static void gen_op_fmovd(TCGv_i64 dst, TCGv_i64 src)
>   static void gen_op_fnegd(TCGv_i64 dst, TCGv_i64 src)
>   {
>       gen_op_clear_ieee_excp_and_FTT();
> -    gen_helper_fnegd(dst, src);
> +    tcg_gen_xori_i64(dst, src, 1ull << 63);
>   }
>   
>   static void gen_op_fabsd(TCGv_i64 dst, TCGv_i64 src)
>   {
>       gen_op_clear_ieee_excp_and_FTT();
> -    gen_helper_fabsd(dst, src);
> +    tcg_gen_andi_i64(dst, src, ~(1ull << 63));
> +}
> +
> +static void gen_op_fnegq(TCGv_i128 dst, TCGv_i128 src)
> +{
> +    TCGv_i64 l = tcg_temp_new_i64();
> +    TCGv_i64 h = tcg_temp_new_i64();
> +
> +    tcg_gen_extr_i128_i64(l, h, src);
> +    tcg_gen_xori_i64(h, h, 1ull << 63);
> +    tcg_gen_concat_i64_i128(dst, l, h);
> +}
> +
> +static void gen_op_fabsq(TCGv_i128 dst, TCGv_i128 src)
> +{
> +    TCGv_i64 l = tcg_temp_new_i64();
> +    TCGv_i64 h = tcg_temp_new_i64();
> +
> +    tcg_gen_extr_i128_i64(l, h, src);
> +    tcg_gen_andi_i64(h, h, ~(1ull << 63));
> +    tcg_gen_concat_i64_i128(dst, l, h);
>   }

Why not extract these as generic TCG FPU helpers?

$ git grep -wE 'float...?_(chs|abs)' target/
target/arm/tcg/helper-a64.c:214:    a = float16_chs(a);
target/arm/tcg/helper-a64.c:229:    a = float32_chs(a);
target/arm/tcg/helper-a64.c:244:    a = float64_chs(a);
target/arm/tcg/helper-a64.c:259:    a = float16_chs(a);
target/arm/tcg/helper-a64.c:274:    a = float32_chs(a);
target/arm/tcg/helper-a64.c:289:    a = float64_chs(a);
target/arm/tcg/helper-a64.c:632:    float16 f0 = float16_abs(a);
target/arm/tcg/helper-a64.c:633:    float16 f1 = float16_abs(b);
target/arm/tcg/helper-a64.c:642:    float16 f0 = float16_abs(a);
target/arm/tcg/helper-a64.c:643:    float16 f1 = float16_abs(b);
target/arm/tcg/mve_helper.c:2840:    return float16_abs(float16_sub(a, 
b, s));
target/arm/tcg/mve_helper.c:2845:    return float32_abs(float32_sub(a, 
b, s));
target/arm/tcg/mve_helper.c:2854:    return 
float16_maxnum(float16_abs(a), float16_abs(b), s);
target/arm/tcg/mve_helper.c:2859:    return 
float32_maxnum(float32_abs(a), float32_abs(b), s);
target/arm/tcg/mve_helper.c:2864:    return 
float16_minnum(float16_abs(a), float16_abs(b), s);
target/arm/tcg/mve_helper.c:2869:    return 
float32_minnum(float32_abs(a), float32_abs(b), s);
target/arm/tcg/neon_helper.c:1513:    float32 f0 = 
float32_abs(make_float32(a));
target/arm/tcg/neon_helper.c:1514:    float32 f1 = 
float32_abs(make_float32(b));
target/arm/tcg/neon_helper.c:1521:    float32 f0 = 
float32_abs(make_float32(a));
target/arm/tcg/neon_helper.c:1522:    float32 f1 = 
float32_abs(make_float32(b));
target/arm/tcg/neon_helper.c:1529:    float64 f0 = 
float64_abs(make_float64(a));
target/arm/tcg/neon_helper.c:1530:    float64 f1 = 
float64_abs(make_float64(b));
target/arm/tcg/neon_helper.c:1537:    float64 f0 = 
float64_abs(make_float64(a));
target/arm/tcg/neon_helper.c:1538:    float64 f1 = 
float64_abs(make_float64(b));
target/arm/tcg/sve_helper.c:4227:DO_REDUCE(sve_fmaxv_h, float16, H1_2, 
max, float16_chs(float16_infinity))
target/arm/tcg/sve_helper.c:4228:DO_REDUCE(sve_fmaxv_s, float32, H1_4, 
max, float32_chs(float32_infinity))
target/arm/tcg/sve_helper.c:4229:DO_REDUCE(sve_fmaxv_d, float64, H1_8, 
max, float64_chs(float64_infinity))
target/arm/tcg/sve_helper.c:4345:    return float16_abs(float16_sub(a, 
b, s));
target/arm/tcg/sve_helper.c:4350:    return float32_abs(float32_sub(a, 
b, s));
target/arm/tcg/sve_helper.c:4355:    return float64_abs(float64_sub(a, 
b, s));
target/arm/tcg/sve_helper.c:4997:            mm = float16_abs(mm);
target/arm/tcg/sve_helper.c:5019:            mm = float32_abs(mm);
target/arm/tcg/sve_helper.c:5045:            mm = float64_abs(mm);
target/arm/tcg/sve_helper.c:5062:    float16 neg_real = 
float16_chs(neg_imag);
target/arm/tcg/sve_helper.c:5094:    float32 neg_real = 
float32_chs(neg_imag);
target/arm/tcg/sve_helper.c:5126:    float64 neg_real = 
float64_chs(neg_imag);
target/arm/tcg/vec_helper.c:996:    return -float16_le(float16_abs(op2), 
float16_abs(op1), stat);
target/arm/tcg/vec_helper.c:1001:    return 
-float32_le(float32_abs(op2), float32_abs(op1), stat);
target/arm/tcg/vec_helper.c:1006:    return 
-float16_lt(float16_abs(op2), float16_abs(op1), stat);
target/arm/tcg/vec_helper.c:1011:    return 
-float32_lt(float32_abs(op2), float32_abs(op1), stat);
target/arm/tcg/vec_helper.c:1124:    return float16_abs(float16_sub(op1, 
op2, stat));
target/arm/tcg/vec_helper.c:1129:    return float32_abs(float32_sub(op1, 
op2, stat));
target/arm/tcg/vec_helper.c:1304:    return 
float16_muladd(float16_chs(op1), op2, dest, 0, stat);
target/arm/tcg/vec_helper.c:1310:    return 
float32_muladd(float32_chs(op1), op2, dest, 0, stat);
target/arm/vfp_helper.c:286:    return float16_chs(a);
target/arm/vfp_helper.c:291:    return float32_chs(a);
target/arm/vfp_helper.c:296:    return float64_chs(a);
target/arm/vfp_helper.c:301:    return float16_abs(a);
target/arm/vfp_helper.c:306:    return float32_abs(a);
target/arm/vfp_helper.c:311:    return float64_abs(a);
target/arm/vfp_helper.c:688:    } else if (float16_abs(f16) < (1 << 8)) {
target/arm/vfp_helper.c:738:    } else if (float32_abs(f32) < (1ULL << 
21)) {
target/arm/vfp_helper.c:1133:    if (value == float64_chs(float64_zero)) {
target/i386/tcg/fpu_helper.c:591:    ST0 = floatx80_chs(ST0);
target/i386/tcg/fpu_helper.c:596:    ST0 = floatx80_abs(ST0);
target/i386/tcg/fpu_helper.c:781:        tmp = floatx80_chs(tmp);
target/i386/tcg/fpu_helper.c:1739:        ST0 = 
floatx80_div(floatx80_chs(floatx80_one), floatx80_zero,
target/i386/tcg/fpu_helper.c:2104:            ST1 = floatx80_chs(ST1);
target/i386/tcg/fpu_helper.c:2119:            ST1 = floatx80_chs(ST0);
target/i386/tcg/fpu_helper.c:2135:            ST1 = floatx80_chs(ST1);
target/i386/tcg/fpu_helper.c:2140:            ST1 = 
floatx80_chs(floatx80_zero);
target/i386/tcg/fpu_helper.c:2276: 
floatx80_chs(floatx80_zero) :
target/i386/tcg/fpu_helper.c:2285: 
floatx80_chs(floatx80_infinity) :
target/m68k/fpu_helper.c:212:    res->d = 
floatx80_round(floatx80_abs(val->d), &env->fp_status);
target/m68k/fpu_helper.c:218:    res->d = 
floatx80_round(floatx80_abs(val->d), &env->fp_status);
target/m68k/fpu_helper.c:225:    res->d = 
floatx80_round(floatx80_abs(val->d), &env->fp_status);
target/m68k/fpu_helper.c:231:    res->d = 
floatx80_round(floatx80_chs(val->d), &env->fp_status);
target/m68k/fpu_helper.c:237:    res->d = 
floatx80_round(floatx80_chs(val->d), &env->fp_status);
target/m68k/fpu_helper.c:244:    res->d = 
floatx80_round(floatx80_chs(val->d), &env->fp_status);
target/m68k/fpu_helper.c:557:        quotient = 
floatx80_to_int32(floatx80_abs(fp_quot.d), &env->fp_status);
target/m68k/softfloat.c:2714:            fp0 = floatx80_abs(a); /* Y = 
|X| */
target/m68k/softfloat.c:2734:        fp0 = floatx80_abs(a); /* Y = |X| */
target/mips/tcg/fpu_helper.c:977:   return float64_abs(fdt0);
target/mips/tcg/fpu_helper.c:982:    return float32_abs(fst0);
target/mips/tcg/fpu_helper.c:990:    wt0 = float32_abs(fdt0 & 0XFFFFFFFF);
target/mips/tcg/fpu_helper.c:991:    wth0 = float32_abs(fdt0 >> 32);
target/mips/tcg/fpu_helper.c:997:   return float64_chs(fdt0);
target/mips/tcg/fpu_helper.c:1002:    return float32_chs(fst0);
target/mips/tcg/fpu_helper.c:1010:    wt0 = float32_chs(fdt0 & 0XFFFFFFFF);
target/mips/tcg/fpu_helper.c:1011:    wth0 = float32_chs(fdt0 >> 32);
target/mips/tcg/fpu_helper.c:1365:    fdt2 = 
float64_chs(float64_sub(fdt2, float64_one,
target/mips/tcg/fpu_helper.c:1374:    fst2 = 
float32_chs(float32_sub(fst2, float32_one,
target/mips/tcg/fpu_helper.c:1389:    fstl2 = 
float32_chs(float32_sub(fstl2, float32_one,
target/mips/tcg/fpu_helper.c:1391:    fsth2 = 
float32_chs(float32_sub(fsth2, float32_one,
target/mips/tcg/fpu_helper.c:1401:    fdt2 = 
float64_chs(float64_div(fdt2, FLOAT_TWO64,
target/mips/tcg/fpu_helper.c:1411:    fst2 = 
float32_chs(float32_div(fst2, FLOAT_TWO32,
target/mips/tcg/fpu_helper.c:1428:    fstl2 = 
float32_chs(float32_div(fstl2, FLOAT_TWO32,
target/mips/tcg/fpu_helper.c:1430:    fsth2 = 
float32_chs(float32_div(fsth2, FLOAT_TWO32,
target/mips/tcg/fpu_helper.c:1633:    fst0 = float64_chs(fst0);
target/mips/tcg/fpu_helper.c:1644:    fst0 = float32_chs(fst0);
target/mips/tcg/fpu_helper.c:1662:    fstl0 = float32_chs(fstl0);
target/mips/tcg/fpu_helper.c:1665:    fsth0 = float32_chs(fsth0);
target/mips/tcg/fpu_helper.c:1676:    fst0 = float64_chs(fst0);
target/mips/tcg/fpu_helper.c:1687:    fst0 = float32_chs(fst0);
target/mips/tcg/fpu_helper.c:1705:    fstl0 = float32_chs(fstl0);
target/mips/tcg/fpu_helper.c:1708:    fsth0 = float32_chs(fsth0);
target/mips/tcg/fpu_helper.c:1781:    fdt0 = float64_abs(fdt0); 
                         \
target/mips/tcg/fpu_helper.c:1782:    fdt1 = float64_abs(fdt1); 
                         \
target/mips/tcg/fpu_helper.c:1860:    fst0 = float32_abs(fst0); 
                         \
target/mips/tcg/fpu_helper.c:1861:    fst1 = float32_abs(fst1); 
                         \
target/mips/tcg/fpu_helper.c:1950:    fst0 = float32_abs(fdt0 & 
0XFFFFFFFF);                      \
target/mips/tcg/fpu_helper.c:1951:    fsth0 = float32_abs(fdt0 >> 32); 
                          \
target/mips/tcg/fpu_helper.c:1952:    fst1 = float32_abs(fdt1 & 
0XFFFFFFFF);                      \
target/mips/tcg/fpu_helper.c:1953:    fsth1 = float32_abs(fdt1 >> 32); 
                          \
target/ppc/fpu_helper.c:44:        return float32_chs(a);
target/ppc/int_helper.c:694:            float32 bneg = 
float32_chs(b->f32[i]);
target/s390x/tcg/vec_fpu_helper.c:922:                a = float32_abs(a);
target/s390x/tcg/vec_fpu_helper.c:923:                b = float32_abs(b);
target/s390x/tcg/vec_fpu_helper.c:984:                a = float64_abs(a);
target/s390x/tcg/vec_fpu_helper.c:985:                b = float64_abs(b);
target/s390x/tcg/vec_fpu_helper.c:1042:            a = float128_abs(a);
target/s390x/tcg/vec_fpu_helper.c:1043:            b = float128_abs(b);
target/sparc/fop_helper.c:119:    return float32_chs(src);
target/sparc/fop_helper.c:125:    return float64_chs(src);
target/sparc/fop_helper.c:130:    QT0 = float128_chs(QT1);
target/sparc/fop_helper.c:234:    return float32_abs(src);
target/sparc/fop_helper.c:240:    return float64_abs(src);
target/sparc/fop_helper.c:245:    QT0 = float128_abs(QT1);
target/xtensa/fpu_helper.c:126:    return float64_abs(v);
target/xtensa/fpu_helper.c:131:    return float32_abs(v);
target/xtensa/fpu_helper.c:136:    return float64_chs(v);
target/xtensa/fpu_helper.c:141:    return float32_chs(v);

Re: [PATCH 05/22] target/sparc: Inline FNEG, FABS

[Richard Henderson]

On 1/30/24 18:40, Philippe Mathieu-Daudé wrote:
> Hi Richard,
> 
> On 3/11/23 18:38, Richard Henderson wrote:
>> These are simple bit manipulation insns.
>> Begin using i128 for float128.
>> Implement FMOVq with do_qq.
>>
>> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
>> ---
>>   target/sparc/helper.h     |  6 ----
>>   target/sparc/fop_helper.c | 34 ---------------------
>>   target/sparc/translate.c  | 62 +++++++++++++++++++--------------------
>>   3 files changed, 30 insertions(+), 72 deletions(-)
> 
> 
>> @@ -1239,13 +1235,13 @@ static void gen_op_fmovs(TCGv_i32 dst, TCGv_i32 src)
>>   static void gen_op_fnegs(TCGv_i32 dst, TCGv_i32 src)
>>   {
>>       gen_op_clear_ieee_excp_and_FTT();
>> -    gen_helper_fnegs(dst, src);
>> +    tcg_gen_xori_i32(dst, src, 1u << 31);
>>   }
>>   static void gen_op_fabss(TCGv_i32 dst, TCGv_i32 src)
>>   {
>>       gen_op_clear_ieee_excp_and_FTT();
>> -    gen_helper_fabss(dst, src);
>> +    tcg_gen_andi_i32(dst, src, ~(1u << 31));
>>   }
>>   static void gen_op_fmovd(TCGv_i64 dst, TCGv_i64 src)
>> @@ -1257,13 +1253,33 @@ static void gen_op_fmovd(TCGv_i64 dst, TCGv_i64 src)
>>   static void gen_op_fnegd(TCGv_i64 dst, TCGv_i64 src)
>>   {
>>       gen_op_clear_ieee_excp_and_FTT();
>> -    gen_helper_fnegd(dst, src);
>> +    tcg_gen_xori_i64(dst, src, 1ull << 63);
>>   }
>>   static void gen_op_fabsd(TCGv_i64 dst, TCGv_i64 src)
>>   {
>>       gen_op_clear_ieee_excp_and_FTT();
>> -    gen_helper_fabsd(dst, src);
>> +    tcg_gen_andi_i64(dst, src, ~(1ull << 63));
>> +}
>> +
>> +static void gen_op_fnegq(TCGv_i128 dst, TCGv_i128 src)
>> +{
>> +    TCGv_i64 l = tcg_temp_new_i64();
>> +    TCGv_i64 h = tcg_temp_new_i64();
>> +
>> +    tcg_gen_extr_i128_i64(l, h, src);
>> +    tcg_gen_xori_i64(h, h, 1ull << 63);
>> +    tcg_gen_concat_i64_i128(dst, l, h);
>> +}
>> +
>> +static void gen_op_fabsq(TCGv_i128 dst, TCGv_i128 src)
>> +{
>> +    TCGv_i64 l = tcg_temp_new_i64();
>> +    TCGv_i64 h = tcg_temp_new_i64();
>> +
>> +    tcg_gen_extr_i128_i64(l, h, src);
>> +    tcg_gen_andi_i64(h, h, ~(1ull << 63));
>> +    tcg_gen_concat_i64_i128(dst, l, h);
>>   }
> 
> Why not extract these as generic TCG FPU helpers?

The representation of floating-point registers varies wildly between targets.  Sparc would 
be the only one to (a) have float128 and (b) represent them in TCGv_i128.

Even considering float32, is the representation TCGv_i32 or TCGv_i64?
Should the result be nan-boxed (riscv and loongarch)?

We already provide tcg_gen_xori_i{32,64}, so, really that's enough for any target.

> $ git grep -wE 'float...?_(chs|abs)' target/
> target/arm/tcg/helper-a64.c:214:    a = float16_chs(a);
> target/arm/tcg/helper-a64.c:229:    a = float32_chs(a);
> target/arm/tcg/helper-a64.c:244:    a = float64_chs(a);
> target/arm/tcg/helper-a64.c:259:    a = float16_chs(a);
> target/arm/tcg/helper-a64.c:274:    a = float32_chs(a);
> target/arm/tcg/helper-a64.c:289:    a = float64_chs(a);
> target/arm/tcg/helper-a64.c:632:    float16 f0 = float16_abs(a);
> target/arm/tcg/helper-a64.c:633:    float16 f1 = float16_abs(b);
> target/arm/tcg/helper-a64.c:642:    float16 f0 = float16_abs(a);
> target/arm/tcg/helper-a64.c:643:    float16 f1 = float16_abs(b);
> target/arm/tcg/mve_helper.c:2840:    return float16_abs(float16_sub(a, b, s));
> target/arm/tcg/mve_helper.c:2845:    return float32_abs(float32_sub(a, b, s));
> target/arm/tcg/mve_helper.c:2854:    return float16_maxnum(float16_abs(a), float16_abs(b), 
> s);
> target/arm/tcg/mve_helper.c:2859:    return float32_maxnum(float32_abs(a), float32_abs(b), 
> s);
> target/arm/tcg/mve_helper.c:2864:    return float16_minnum(float16_abs(a), float16_abs(b), 
> s);
> target/arm/tcg/mve_helper.c:2869:    return float32_minnum(float32_abs(a), float32_abs(b), 
> s);
> target/arm/tcg/neon_helper.c:1513:    float32 f0 = float32_abs(make_float32(a));
> target/arm/tcg/neon_helper.c:1514:    float32 f1 = float32_abs(make_float32(b));
> target/arm/tcg/neon_helper.c:1521:    float32 f0 = float32_abs(make_float32(a));
> target/arm/tcg/neon_helper.c:1522:    float32 f1 = float32_abs(make_float32(b));
> target/arm/tcg/neon_helper.c:1529:    float64 f0 = float64_abs(make_float64(a));
> target/arm/tcg/neon_helper.c:1530:    float64 f1 = float64_abs(make_float64(b));
> target/arm/tcg/neon_helper.c:1537:    float64 f0 = float64_abs(make_float64(a));
> target/arm/tcg/neon_helper.c:1538:    float64 f1 = float64_abs(make_float64(b));
> target/arm/tcg/sve_helper.c:4227:DO_REDUCE(sve_fmaxv_h, float16, H1_2, max, 
> float16_chs(float16_infinity))
> target/arm/tcg/sve_helper.c:4228:DO_REDUCE(sve_fmaxv_s, float32, H1_4, max, 
> float32_chs(float32_infinity))
> target/arm/tcg/sve_helper.c:4229:DO_REDUCE(sve_fmaxv_d, float64, H1_8, max, 
> float64_chs(float64_infinity))
> target/arm/tcg/sve_helper.c:4345:    return float16_abs(float16_sub(a, b, s));
> target/arm/tcg/sve_helper.c:4350:    return float32_abs(float32_sub(a, b, s));
> target/arm/tcg/sve_helper.c:4355:    return float64_abs(float64_sub(a, b, s));
> target/arm/tcg/sve_helper.c:4997:            mm = float16_abs(mm);
> target/arm/tcg/sve_helper.c:5019:            mm = float32_abs(mm);
> target/arm/tcg/sve_helper.c:5045:            mm = float64_abs(mm);
> target/arm/tcg/sve_helper.c:5062:    float16 neg_real = float16_chs(neg_imag);
> target/arm/tcg/sve_helper.c:5094:    float32 neg_real = float32_chs(neg_imag);
> target/arm/tcg/sve_helper.c:5126:    float64 neg_real = float64_chs(neg_imag);
> target/arm/tcg/vec_helper.c:996:    return -float16_le(float16_abs(op2), float16_abs(op1), 
> stat);
> target/arm/tcg/vec_helper.c:1001:    return -float32_le(float32_abs(op2), 
> float32_abs(op1), stat);
> target/arm/tcg/vec_helper.c:1006:    return -float16_lt(float16_abs(op2), 
> float16_abs(op1), stat);
> target/arm/tcg/vec_helper.c:1011:    return -float32_lt(float32_abs(op2), 
> float32_abs(op1), stat);
> target/arm/tcg/vec_helper.c:1124:    return float16_abs(float16_sub(op1, op2, stat));
> target/arm/tcg/vec_helper.c:1129:    return float32_abs(float32_sub(op1, op2, stat));
> target/arm/tcg/vec_helper.c:1304:    return float16_muladd(float16_chs(op1), op2, dest, 0, 
> stat);
> target/arm/tcg/vec_helper.c:1310:    return float32_muladd(float32_chs(op1), op2, dest, 0, 
> stat);
> target/arm/vfp_helper.c:286:    return float16_chs(a);
> target/arm/vfp_helper.c:291:    return float32_chs(a);
> target/arm/vfp_helper.c:296:    return float64_chs(a);
> target/arm/vfp_helper.c:301:    return float16_abs(a);
> target/arm/vfp_helper.c:306:    return float32_abs(a);
> target/arm/vfp_helper.c:311:    return float64_abs(a);
> target/arm/vfp_helper.c:688:    } else if (float16_abs(f16) < (1 << 8)) {
> target/arm/vfp_helper.c:738:    } else if (float32_abs(f32) < (1ULL << 21)) {
> target/arm/vfp_helper.c:1133:    if (value == float64_chs(float64_zero)) {
> target/i386/tcg/fpu_helper.c:591:    ST0 = floatx80_chs(ST0);
> target/i386/tcg/fpu_helper.c:596:    ST0 = floatx80_abs(ST0);
> target/i386/tcg/fpu_helper.c:781:        tmp = floatx80_chs(tmp);
> target/i386/tcg/fpu_helper.c:1739:        ST0 = floatx80_div(floatx80_chs(floatx80_one), 
> floatx80_zero,
> target/i386/tcg/fpu_helper.c:2104:            ST1 = floatx80_chs(ST1);
> target/i386/tcg/fpu_helper.c:2119:            ST1 = floatx80_chs(ST0);
> target/i386/tcg/fpu_helper.c:2135:            ST1 = floatx80_chs(ST1);
> target/i386/tcg/fpu_helper.c:2140:            ST1 = floatx80_chs(floatx80_zero);
> target/i386/tcg/fpu_helper.c:2276: floatx80_chs(floatx80_zero) :
> target/i386/tcg/fpu_helper.c:2285: floatx80_chs(floatx80_infinity) :
> target/m68k/fpu_helper.c:212:    res->d = floatx80_round(floatx80_abs(val->d), 
> &env->fp_status);
> target/m68k/fpu_helper.c:218:    res->d = floatx80_round(floatx80_abs(val->d), 
> &env->fp_status);
> target/m68k/fpu_helper.c:225:    res->d = floatx80_round(floatx80_abs(val->d), 
> &env->fp_status);
> target/m68k/fpu_helper.c:231:    res->d = floatx80_round(floatx80_chs(val->d), 
> &env->fp_status);
> target/m68k/fpu_helper.c:237:    res->d = floatx80_round(floatx80_chs(val->d), 
> &env->fp_status);
> target/m68k/fpu_helper.c:244:    res->d = floatx80_round(floatx80_chs(val->d), 
> &env->fp_status);
> target/m68k/fpu_helper.c:557:        quotient = floatx80_to_int32(floatx80_abs(fp_quot.d), 
> &env->fp_status);
> target/m68k/softfloat.c:2714:            fp0 = floatx80_abs(a); /* Y = |X| */
> target/m68k/softfloat.c:2734:        fp0 = floatx80_abs(a); /* Y = |X| */
> target/mips/tcg/fpu_helper.c:977:   return float64_abs(fdt0);
> target/mips/tcg/fpu_helper.c:982:    return float32_abs(fst0);
> target/mips/tcg/fpu_helper.c:990:    wt0 = float32_abs(fdt0 & 0XFFFFFFFF);
> target/mips/tcg/fpu_helper.c:991:    wth0 = float32_abs(fdt0 >> 32);
> target/mips/tcg/fpu_helper.c:997:   return float64_chs(fdt0);
> target/mips/tcg/fpu_helper.c:1002:    return float32_chs(fst0);
> target/mips/tcg/fpu_helper.c:1010:    wt0 = float32_chs(fdt0 & 0XFFFFFFFF);
> target/mips/tcg/fpu_helper.c:1011:    wth0 = float32_chs(fdt0 >> 32);
> target/mips/tcg/fpu_helper.c:1365:    fdt2 = float64_chs(float64_sub(fdt2, float64_one,
> target/mips/tcg/fpu_helper.c:1374:    fst2 = float32_chs(float32_sub(fst2, float32_one,
> target/mips/tcg/fpu_helper.c:1389:    fstl2 = float32_chs(float32_sub(fstl2, float32_one,
> target/mips/tcg/fpu_helper.c:1391:    fsth2 = float32_chs(float32_sub(fsth2, float32_one,
> target/mips/tcg/fpu_helper.c:1401:    fdt2 = float64_chs(float64_div(fdt2, FLOAT_TWO64,
> target/mips/tcg/fpu_helper.c:1411:    fst2 = float32_chs(float32_div(fst2, FLOAT_TWO32,
> target/mips/tcg/fpu_helper.c:1428:    fstl2 = float32_chs(float32_div(fstl2, FLOAT_TWO32,
> target/mips/tcg/fpu_helper.c:1430:    fsth2 = float32_chs(float32_div(fsth2, FLOAT_TWO32,
> target/mips/tcg/fpu_helper.c:1633:    fst0 = float64_chs(fst0);
> target/mips/tcg/fpu_helper.c:1644:    fst0 = float32_chs(fst0);
> target/mips/tcg/fpu_helper.c:1662:    fstl0 = float32_chs(fstl0);
> target/mips/tcg/fpu_helper.c:1665:    fsth0 = float32_chs(fsth0);
> target/mips/tcg/fpu_helper.c:1676:    fst0 = float64_chs(fst0);
> target/mips/tcg/fpu_helper.c:1687:    fst0 = float32_chs(fst0);
> target/mips/tcg/fpu_helper.c:1705:    fstl0 = float32_chs(fstl0);
> target/mips/tcg/fpu_helper.c:1708:    fsth0 = float32_chs(fsth0);
> target/mips/tcg/fpu_helper.c:1781:    fdt0 = float64_abs(fdt0);                         \
> target/mips/tcg/fpu_helper.c:1782:    fdt1 = float64_abs(fdt1);                         \
> target/mips/tcg/fpu_helper.c:1860:    fst0 = float32_abs(fst0);                         \
> target/mips/tcg/fpu_helper.c:1861:    fst1 = float32_abs(fst1);                         \
> target/mips/tcg/fpu_helper.c:1950:    fst0 = float32_abs(fdt0 & 
> 0XFFFFFFFF);                      \
> target/mips/tcg/fpu_helper.c:1951:    fsth0 = float32_abs(fdt0 >> 32); 
>                           \
> target/mips/tcg/fpu_helper.c:1952:    fst1 = float32_abs(fdt1 & 
> 0XFFFFFFFF);                      \
> target/mips/tcg/fpu_helper.c:1953:    fsth1 = float32_abs(fdt1 >> 32); 
>                           \
> target/ppc/fpu_helper.c:44:        return float32_chs(a);
> target/ppc/int_helper.c:694:            float32 bneg = float32_chs(b->f32[i]);
> target/s390x/tcg/vec_fpu_helper.c:922:                a = float32_abs(a);
> target/s390x/tcg/vec_fpu_helper.c:923:                b = float32_abs(b);
> target/s390x/tcg/vec_fpu_helper.c:984:                a = float64_abs(a);
> target/s390x/tcg/vec_fpu_helper.c:985:                b = float64_abs(b);
> target/s390x/tcg/vec_fpu_helper.c:1042:            a = float128_abs(a);
> target/s390x/tcg/vec_fpu_helper.c:1043:            b = float128_abs(b);
> target/sparc/fop_helper.c:119:    return float32_chs(src);
> target/sparc/fop_helper.c:125:    return float64_chs(src);
> target/sparc/fop_helper.c:130:    QT0 = float128_chs(QT1);
> target/sparc/fop_helper.c:234:    return float32_abs(src);
> target/sparc/fop_helper.c:240:    return float64_abs(src);
> target/sparc/fop_helper.c:245:    QT0 = float128_abs(QT1);
> target/xtensa/fpu_helper.c:126:    return float64_abs(v);
> target/xtensa/fpu_helper.c:131:    return float32_abs(v);
> target/xtensa/fpu_helper.c:136:    return float64_chs(v);
> target/xtensa/fpu_helper.c:141:    return float32_chs(v);

With only a few exceptions, most of of these results are part of a larger out-of-line 
operation.


r~

[PATCH 06/22] target/sparc: Use i128 for FSQRTq

[Richard Henderson]

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/sparc/helper.h     |  2 +-
 target/sparc/fop_helper.c | 26 ++++++++++++++++++++++++--
 target/sparc/translate.c  | 12 +++++++-----
 3 files changed, 32 insertions(+), 8 deletions(-)

diff --git a/target/sparc/helper.h b/target/sparc/helper.h
index 74a1575d21..eea2fa570c 100644
--- a/target/sparc/helper.h
+++ b/target/sparc/helper.h
@@ -43,7 +43,7 @@ DEF_HELPER_FLAGS_3(fcmps, TCG_CALL_NO_WG, tl, env, f32, f32)
 DEF_HELPER_FLAGS_3(fcmpd, TCG_CALL_NO_WG, tl, env, f64, f64)
 DEF_HELPER_FLAGS_3(fcmpes, TCG_CALL_NO_WG, tl, env, f32, f32)
 DEF_HELPER_FLAGS_3(fcmped, TCG_CALL_NO_WG, tl, env, f64, f64)
-DEF_HELPER_FLAGS_1(fsqrtq, TCG_CALL_NO_RWG, void, env)
+DEF_HELPER_FLAGS_2(fsqrtq, TCG_CALL_NO_RWG, i128, env, i128)
 DEF_HELPER_FLAGS_1(fcmpq, TCG_CALL_NO_WG, tl, env)
 DEF_HELPER_FLAGS_1(fcmpeq, TCG_CALL_NO_WG, tl, env)
 #ifdef TARGET_SPARC64
diff --git a/target/sparc/fop_helper.c b/target/sparc/fop_helper.c
index d6fb769769..d639e50965 100644
--- a/target/sparc/fop_helper.c
+++ b/target/sparc/fop_helper.c
@@ -26,6 +26,28 @@
 #define QT0 (env->qt0)
 #define QT1 (env->qt1)
 
+static inline float128 f128_in(Int128 i)
+{
+    union {
+        Int128 i;
+        float128 f;
+    } u;
+
+    u.i = i;
+    return u.f;
+}
+
+static inline Int128 f128_ret(float128 f)
+{
+    union {
+        Int128 i;
+        float128 f;
+    } u;
+
+    u.f = f;
+    return u.i;
+}
+
 static target_ulong do_check_ieee_exceptions(CPUSPARCState *env, uintptr_t ra)
 {
     target_ulong status = get_float_exception_flags(&env->fp_status);
@@ -222,9 +244,9 @@ float64 helper_fsqrtd(CPUSPARCState *env, float64 src)
     return float64_sqrt(src, &env->fp_status);
 }
 
-void helper_fsqrtq(CPUSPARCState *env)
+Int128 helper_fsqrtq(CPUSPARCState *env, Int128 src)
 {
-    QT0 = float128_sqrt(QT1, &env->fp_status);
+    return f128_ret(float128_sqrt(f128_in(src), &env->fp_status));
 }
 
 #define GEN_FCMP(name, size, reg1, reg2, FS, E)                         \
diff --git a/target/sparc/translate.c b/target/sparc/translate.c
index 254f185b83..6c08bf909e 100644
--- a/target/sparc/translate.c
+++ b/target/sparc/translate.c
@@ -4669,8 +4669,10 @@ TRANS(FNEGq, 64, do_qq, a, gen_op_fnegq)
 TRANS(FABSq, 64, do_qq, a, gen_op_fabsq)
 
 static bool do_env_qq(DisasContext *dc, arg_r_r *a,
-                       void (*func)(TCGv_env))
+                      void (*func)(TCGv_i128, TCGv_env, TCGv_i128))
 {
+    TCGv_i128 t;
+
     if (gen_trap_ifnofpu(dc)) {
         return true;
     }
@@ -4679,11 +4681,11 @@ static bool do_env_qq(DisasContext *dc, arg_r_r *a,
     }
 
     gen_op_clear_ieee_excp_and_FTT();
-    gen_op_load_fpr_QT1(QFPREG(a->rs));
-    func(tcg_env);
+
+    t = gen_load_fpr_Q(dc, a->rs);
+    func(t, tcg_env, t);
     gen_helper_check_ieee_exceptions(cpu_fsr, tcg_env);
-    gen_op_store_QT0_fpr(QFPREG(a->rd));
-    gen_update_fprs_dirty(dc, QFPREG(a->rd));
+    gen_store_fpr_Q(dc, a->rd, t);
     return advance_pc(dc);
 }
 
-- 
2.34.1

[PATCH 07/22] target/sparc: Use i128 for FADDq, FSUBq, FMULq, FDIVq

[Richard Henderson]

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/sparc/helper.h     | 12 +++++-------
 target/sparc/fop_helper.c | 29 ++++++++++++++---------------
 target/sparc/translate.c  | 13 +++++++------
 3 files changed, 26 insertions(+), 28 deletions(-)

diff --git a/target/sparc/helper.h b/target/sparc/helper.h
index eea2fa570c..0a030fc908 100644
--- a/target/sparc/helper.h
+++ b/target/sparc/helper.h
@@ -67,17 +67,16 @@ DEF_HELPER_FLAGS_1(fcmpeq_fcc2, TCG_CALL_NO_WG, tl, env)
 DEF_HELPER_FLAGS_1(fcmpeq_fcc3, TCG_CALL_NO_WG, tl, env)
 #endif
 DEF_HELPER_2(raise_exception, noreturn, env, int)
-#define F_HELPER_0_1(name) \
-  DEF_HELPER_FLAGS_1(f ## name, TCG_CALL_NO_RWG, void, env)
 
 DEF_HELPER_FLAGS_3(faddd, TCG_CALL_NO_RWG, f64, env, f64, f64)
 DEF_HELPER_FLAGS_3(fsubd, TCG_CALL_NO_RWG, f64, env, f64, f64)
 DEF_HELPER_FLAGS_3(fmuld, TCG_CALL_NO_RWG, f64, env, f64, f64)
 DEF_HELPER_FLAGS_3(fdivd, TCG_CALL_NO_RWG, f64, env, f64, f64)
-F_HELPER_0_1(addq)
-F_HELPER_0_1(subq)
-F_HELPER_0_1(mulq)
-F_HELPER_0_1(divq)
+
+DEF_HELPER_FLAGS_3(faddq, TCG_CALL_NO_RWG, i128, env, i128, i128)
+DEF_HELPER_FLAGS_3(fsubq, TCG_CALL_NO_RWG, i128, env, i128, i128)
+DEF_HELPER_FLAGS_3(fmulq, TCG_CALL_NO_RWG, i128, env, i128, i128)
+DEF_HELPER_FLAGS_3(fdivq, TCG_CALL_NO_RWG, i128, env, i128, i128)
 
 DEF_HELPER_FLAGS_3(fadds, TCG_CALL_NO_RWG, f32, env, f32, f32)
 DEF_HELPER_FLAGS_3(fsubs, TCG_CALL_NO_RWG, f32, env, f32, f32)
@@ -135,6 +134,5 @@ VIS_CMPHELPER(cmpeq)
 VIS_CMPHELPER(cmple)
 VIS_CMPHELPER(cmpne)
 #endif
-#undef F_HELPER_0_1
 #undef VIS_HELPER
 #undef VIS_CMPHELPER
diff --git a/target/sparc/fop_helper.c b/target/sparc/fop_helper.c
index d639e50965..ceb64d802f 100644
--- a/target/sparc/fop_helper.c
+++ b/target/sparc/fop_helper.c
@@ -98,22 +98,22 @@ target_ulong helper_check_ieee_exceptions(CPUSPARCState *env)
     return do_check_ieee_exceptions(env, GETPC());
 }
 
-#define F_HELPER(name, p) void helper_f##name##p(CPUSPARCState *env)
-
-#define F_BINOP(name)                                           \
+#define F_BINOP(name)                                                \
     float32 helper_f ## name ## s (CPUSPARCState *env, float32 src1, \
-                                   float32 src2)                \
-    {                                                           \
-        return float32_ ## name (src1, src2, &env->fp_status);  \
-    }                                                           \
+                                   float32 src2)                     \
+    {                                                                \
+        return float32_ ## name (src1, src2, &env->fp_status);       \
+    }                                                                \
     float64 helper_f ## name ## d (CPUSPARCState * env, float64 src1,\
-                                   float64 src2)                \
-    {                                                           \
-        return float64_ ## name (src1, src2, &env->fp_status);  \
-    }                                                           \
-    F_HELPER(name, q)                                           \
-    {                                                           \
-        QT0 = float128_ ## name (QT0, QT1, &env->fp_status);    \
+                                   float64 src2)                     \
+    {                                                                \
+        return float64_ ## name (src1, src2, &env->fp_status);       \
+    }                                                                \
+    Int128 helper_f ## name ## q(CPUSPARCState * env, Int128 src1,   \
+                                 Int128 src2)                        \
+    {                                                                \
+        return f128_ret(float128_ ## name (f128_in(src1), f128_in(src2), \
+                                           &env->fp_status));        \
     }
 
 F_BINOP(add);
@@ -168,7 +168,6 @@ void helper_fxtoq(CPUSPARCState *env, int64_t src)
     QT0 = int64_to_float128(src, &env->fp_status);
 }
 #endif
-#undef F_HELPER
 
 /* floating point conversion */
 float32 helper_fdtos(CPUSPARCState *env, float64 src)
diff --git a/target/sparc/translate.c b/target/sparc/translate.c
index 6c08bf909e..437f54ff19 100644
--- a/target/sparc/translate.c
+++ b/target/sparc/translate.c
@@ -4976,8 +4976,10 @@ static bool do_dddd(DisasContext *dc, arg_r_r_r *a,
 TRANS(PDIST, VIS1, do_dddd, a, gen_helper_pdist)
 
 static bool do_env_qqq(DisasContext *dc, arg_r_r_r *a,
-                       void (*func)(TCGv_env))
+                       void (*func)(TCGv_i128, TCGv_env, TCGv_i128, TCGv_i128))
 {
+    TCGv_i128 src1, src2;
+
     if (gen_trap_ifnofpu(dc)) {
         return true;
     }
@@ -4986,12 +4988,11 @@ static bool do_env_qqq(DisasContext *dc, arg_r_r_r *a,
     }
 
     gen_op_clear_ieee_excp_and_FTT();
-    gen_op_load_fpr_QT0(QFPREG(a->rs1));
-    gen_op_load_fpr_QT1(QFPREG(a->rs2));
-    func(tcg_env);
+    src1 = gen_load_fpr_Q(dc, a->rs1);
+    src2 = gen_load_fpr_Q(dc, a->rs2);
+    func(src1, tcg_env, src1, src2);
     gen_helper_check_ieee_exceptions(cpu_fsr, tcg_env);
-    gen_op_store_QT0_fpr(QFPREG(a->rd));
-    gen_update_fprs_dirty(dc, QFPREG(a->rd));
+    gen_store_fpr_Q(dc, a->rd, src1);
     return advance_pc(dc);
 }
 
-- 
2.34.1

[PATCH 08/22] target/sparc: Use i128 for FqTOs, FqTOi

[Richard Henderson]

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/sparc/helper.h     | 4 ++--
 target/sparc/fop_helper.c | 8 ++++----
 target/sparc/translate.c  | 7 ++++---
 3 files changed, 10 insertions(+), 9 deletions(-)

diff --git a/target/sparc/helper.h b/target/sparc/helper.h
index 0a030fc908..e770107eb0 100644
--- a/target/sparc/helper.h
+++ b/target/sparc/helper.h
@@ -98,13 +98,13 @@ DEF_HELPER_FLAGS_2(fxtoq, TCG_CALL_NO_RWG, void, env, s64)
 #endif
 DEF_HELPER_FLAGS_2(fdtos, TCG_CALL_NO_RWG, f32, env, f64)
 DEF_HELPER_FLAGS_2(fstod, TCG_CALL_NO_RWG, f64, env, f32)
-DEF_HELPER_FLAGS_1(fqtos, TCG_CALL_NO_RWG, f32, env)
+DEF_HELPER_FLAGS_2(fqtos, TCG_CALL_NO_RWG, f32, env, i128)
 DEF_HELPER_FLAGS_2(fstoq, TCG_CALL_NO_RWG, void, env, f32)
 DEF_HELPER_FLAGS_1(fqtod, TCG_CALL_NO_RWG, f64, env)
 DEF_HELPER_FLAGS_2(fdtoq, TCG_CALL_NO_RWG, void, env, f64)
 DEF_HELPER_FLAGS_2(fstoi, TCG_CALL_NO_RWG, s32, env, f32)
 DEF_HELPER_FLAGS_2(fdtoi, TCG_CALL_NO_RWG, s32, env, f64)
-DEF_HELPER_FLAGS_1(fqtoi, TCG_CALL_NO_RWG, s32, env)
+DEF_HELPER_FLAGS_2(fqtoi, TCG_CALL_NO_RWG, s32, env, i128)
 #ifdef TARGET_SPARC64
 DEF_HELPER_FLAGS_2(fstox, TCG_CALL_NO_RWG, s64, env, f32)
 DEF_HELPER_FLAGS_2(fdtox, TCG_CALL_NO_RWG, s64, env, f64)
diff --git a/target/sparc/fop_helper.c b/target/sparc/fop_helper.c
index ceb64d802f..657a14575d 100644
--- a/target/sparc/fop_helper.c
+++ b/target/sparc/fop_helper.c
@@ -180,9 +180,9 @@ float64 helper_fstod(CPUSPARCState *env, float32 src)
     return float32_to_float64(src, &env->fp_status);
 }
 
-float32 helper_fqtos(CPUSPARCState *env)
+float32 helper_fqtos(CPUSPARCState *env, Int128 src)
 {
-    return float128_to_float32(QT1, &env->fp_status);
+    return float128_to_float32(f128_in(src), &env->fp_status);
 }
 
 void helper_fstoq(CPUSPARCState *env, float32 src)
@@ -211,9 +211,9 @@ int32_t helper_fdtoi(CPUSPARCState *env, float64 src)
     return float64_to_int32_round_to_zero(src, &env->fp_status);
 }
 
-int32_t helper_fqtoi(CPUSPARCState *env)
+int32_t helper_fqtoi(CPUSPARCState *env, Int128 src)
 {
-    return float128_to_int32_round_to_zero(QT1, &env->fp_status);
+    return float128_to_int32_round_to_zero(f128_in(src), &env->fp_status);
 }
 
 #ifdef TARGET_SPARC64
diff --git a/target/sparc/translate.c b/target/sparc/translate.c
index 437f54ff19..34a774c8ef 100644
--- a/target/sparc/translate.c
+++ b/target/sparc/translate.c
@@ -4692,8 +4692,9 @@ static bool do_env_qq(DisasContext *dc, arg_r_r *a,
 TRANS(FSQRTq, ALL, do_env_qq, a, gen_helper_fsqrtq)
 
 static bool do_env_fq(DisasContext *dc, arg_r_r *a,
-                      void (*func)(TCGv_i32, TCGv_env))
+                      void (*func)(TCGv_i32, TCGv_env, TCGv_i128))
 {
+    TCGv_i128 src;
     TCGv_i32 dst;
 
     if (gen_trap_ifnofpu(dc)) {
@@ -4704,9 +4705,9 @@ static bool do_env_fq(DisasContext *dc, arg_r_r *a,
     }
 
     gen_op_clear_ieee_excp_and_FTT();
-    gen_op_load_fpr_QT1(QFPREG(a->rs));
+    src = gen_load_fpr_Q(dc, a->rs);
     dst = tcg_temp_new_i32();
-    func(dst, tcg_env);
+    func(dst, tcg_env, src);
     gen_helper_check_ieee_exceptions(cpu_fsr, tcg_env);
     gen_store_fpr_F(dc, a->rd, dst);
     return advance_pc(dc);
-- 
2.34.1

[PATCH 09/22] target/sparc: Use i128 for FqTOd, FqTOx

[Richard Henderson]

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/sparc/helper.h     | 4 ++--
 target/sparc/fop_helper.c | 8 ++++----
 target/sparc/translate.c  | 7 ++++---
 3 files changed, 10 insertions(+), 9 deletions(-)

diff --git a/target/sparc/helper.h b/target/sparc/helper.h
index e770107eb0..4cb3451878 100644
--- a/target/sparc/helper.h
+++ b/target/sparc/helper.h
@@ -100,7 +100,7 @@ DEF_HELPER_FLAGS_2(fdtos, TCG_CALL_NO_RWG, f32, env, f64)
 DEF_HELPER_FLAGS_2(fstod, TCG_CALL_NO_RWG, f64, env, f32)
 DEF_HELPER_FLAGS_2(fqtos, TCG_CALL_NO_RWG, f32, env, i128)
 DEF_HELPER_FLAGS_2(fstoq, TCG_CALL_NO_RWG, void, env, f32)
-DEF_HELPER_FLAGS_1(fqtod, TCG_CALL_NO_RWG, f64, env)
+DEF_HELPER_FLAGS_2(fqtod, TCG_CALL_NO_RWG, f64, env, i128)
 DEF_HELPER_FLAGS_2(fdtoq, TCG_CALL_NO_RWG, void, env, f64)
 DEF_HELPER_FLAGS_2(fstoi, TCG_CALL_NO_RWG, s32, env, f32)
 DEF_HELPER_FLAGS_2(fdtoi, TCG_CALL_NO_RWG, s32, env, f64)
@@ -108,7 +108,7 @@ DEF_HELPER_FLAGS_2(fqtoi, TCG_CALL_NO_RWG, s32, env, i128)
 #ifdef TARGET_SPARC64
 DEF_HELPER_FLAGS_2(fstox, TCG_CALL_NO_RWG, s64, env, f32)
 DEF_HELPER_FLAGS_2(fdtox, TCG_CALL_NO_RWG, s64, env, f64)
-DEF_HELPER_FLAGS_1(fqtox, TCG_CALL_NO_RWG, s64, env)
+DEF_HELPER_FLAGS_2(fqtox, TCG_CALL_NO_RWG, s64, env, i128)
 
 DEF_HELPER_FLAGS_2(fpmerge, TCG_CALL_NO_RWG_SE, i64, i64, i64)
 DEF_HELPER_FLAGS_2(fmul8x16, TCG_CALL_NO_RWG_SE, i64, i64, i64)
diff --git a/target/sparc/fop_helper.c b/target/sparc/fop_helper.c
index 657a14575d..9f39b933e8 100644
--- a/target/sparc/fop_helper.c
+++ b/target/sparc/fop_helper.c
@@ -190,9 +190,9 @@ void helper_fstoq(CPUSPARCState *env, float32 src)
     QT0 = float32_to_float128(src, &env->fp_status);
 }
 
-float64 helper_fqtod(CPUSPARCState *env)
+float64 helper_fqtod(CPUSPARCState *env, Int128 src)
 {
-    return float128_to_float64(QT1, &env->fp_status);
+    return float128_to_float64(f128_in(src), &env->fp_status);
 }
 
 void helper_fdtoq(CPUSPARCState *env, float64 src)
@@ -227,9 +227,9 @@ int64_t helper_fdtox(CPUSPARCState *env, float64 src)
     return float64_to_int64_round_to_zero(src, &env->fp_status);
 }
 
-int64_t helper_fqtox(CPUSPARCState *env)
+int64_t helper_fqtox(CPUSPARCState *env, Int128 src)
 {
-    return float128_to_int64_round_to_zero(QT1, &env->fp_status);
+    return float128_to_int64_round_to_zero(f128_in(src), &env->fp_status);
 }
 #endif
 
diff --git a/target/sparc/translate.c b/target/sparc/translate.c
index 34a774c8ef..ba0b7f32d2 100644
--- a/target/sparc/translate.c
+++ b/target/sparc/translate.c
@@ -4717,8 +4717,9 @@ TRANS(FqTOs, ALL, do_env_fq, a, gen_helper_fqtos)
 TRANS(FqTOi, ALL, do_env_fq, a, gen_helper_fqtoi)
 
 static bool do_env_dq(DisasContext *dc, arg_r_r *a,
-                      void (*func)(TCGv_i64, TCGv_env))
+                      void (*func)(TCGv_i64, TCGv_env, TCGv_i128))
 {
+    TCGv_i128 src;
     TCGv_i64 dst;
 
     if (gen_trap_ifnofpu(dc)) {
@@ -4729,9 +4730,9 @@ static bool do_env_dq(DisasContext *dc, arg_r_r *a,
     }
 
     gen_op_clear_ieee_excp_and_FTT();
-    gen_op_load_fpr_QT1(QFPREG(a->rs));
+    src = gen_load_fpr_Q(dc, a->rs);
     dst = gen_dest_fpr_D(dc, a->rd);
-    func(dst, tcg_env);
+    func(dst, tcg_env, src);
     gen_helper_check_ieee_exceptions(cpu_fsr, tcg_env);
     gen_store_fpr_D(dc, a->rd, dst);
     return advance_pc(dc);
-- 
2.34.1

[PATCH 10/22] target/sparc: Use i128 for FCMPq, FCMPEq

[Richard Henderson]

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/sparc/helper.h     | 16 ++++++------
 target/sparc/fop_helper.c | 23 +++++++++--------
 target/sparc/translate.c  | 54 +++++++++++++++------------------------
 3 files changed, 41 insertions(+), 52 deletions(-)

diff --git a/target/sparc/helper.h b/target/sparc/helper.h
index 4cb3451878..7caae9a441 100644
--- a/target/sparc/helper.h
+++ b/target/sparc/helper.h
@@ -44,8 +44,8 @@ DEF_HELPER_FLAGS_3(fcmpd, TCG_CALL_NO_WG, tl, env, f64, f64)
 DEF_HELPER_FLAGS_3(fcmpes, TCG_CALL_NO_WG, tl, env, f32, f32)
 DEF_HELPER_FLAGS_3(fcmped, TCG_CALL_NO_WG, tl, env, f64, f64)
 DEF_HELPER_FLAGS_2(fsqrtq, TCG_CALL_NO_RWG, i128, env, i128)
-DEF_HELPER_FLAGS_1(fcmpq, TCG_CALL_NO_WG, tl, env)
-DEF_HELPER_FLAGS_1(fcmpeq, TCG_CALL_NO_WG, tl, env)
+DEF_HELPER_FLAGS_3(fcmpq, TCG_CALL_NO_WG, tl, env, i128, i128)
+DEF_HELPER_FLAGS_3(fcmpeq, TCG_CALL_NO_WG, tl, env, i128, i128)
 #ifdef TARGET_SPARC64
 DEF_HELPER_FLAGS_3(fcmps_fcc1, TCG_CALL_NO_WG, tl, env, f32, f32)
 DEF_HELPER_FLAGS_3(fcmps_fcc2, TCG_CALL_NO_WG, tl, env, f32, f32)
@@ -59,12 +59,12 @@ DEF_HELPER_FLAGS_3(fcmpes_fcc3, TCG_CALL_NO_WG, tl, env, f32, f32)
 DEF_HELPER_FLAGS_3(fcmped_fcc1, TCG_CALL_NO_WG, tl, env, f64, f64)
 DEF_HELPER_FLAGS_3(fcmped_fcc2, TCG_CALL_NO_WG, tl, env, f64, f64)
 DEF_HELPER_FLAGS_3(fcmped_fcc3, TCG_CALL_NO_WG, tl, env, f64, f64)
-DEF_HELPER_FLAGS_1(fcmpq_fcc1, TCG_CALL_NO_WG, tl, env)
-DEF_HELPER_FLAGS_1(fcmpq_fcc2, TCG_CALL_NO_WG, tl, env)
-DEF_HELPER_FLAGS_1(fcmpq_fcc3, TCG_CALL_NO_WG, tl, env)
-DEF_HELPER_FLAGS_1(fcmpeq_fcc1, TCG_CALL_NO_WG, tl, env)
-DEF_HELPER_FLAGS_1(fcmpeq_fcc2, TCG_CALL_NO_WG, tl, env)
-DEF_HELPER_FLAGS_1(fcmpeq_fcc3, TCG_CALL_NO_WG, tl, env)
+DEF_HELPER_FLAGS_3(fcmpq_fcc1, TCG_CALL_NO_WG, tl, env, i128, i128)
+DEF_HELPER_FLAGS_3(fcmpq_fcc2, TCG_CALL_NO_WG, tl, env, i128, i128)
+DEF_HELPER_FLAGS_3(fcmpq_fcc3, TCG_CALL_NO_WG, tl, env, i128, i128)
+DEF_HELPER_FLAGS_3(fcmpeq_fcc1, TCG_CALL_NO_WG, tl, env, i128, i128)
+DEF_HELPER_FLAGS_3(fcmpeq_fcc2, TCG_CALL_NO_WG, tl, env, i128, i128)
+DEF_HELPER_FLAGS_3(fcmpeq_fcc3, TCG_CALL_NO_WG, tl, env, i128, i128)
 #endif
 DEF_HELPER_2(raise_exception, noreturn, env, int)
 
diff --git a/target/sparc/fop_helper.c b/target/sparc/fop_helper.c
index 9f39b933e8..faf75e651f 100644
--- a/target/sparc/fop_helper.c
+++ b/target/sparc/fop_helper.c
@@ -248,9 +248,12 @@ Int128 helper_fsqrtq(CPUSPARCState *env, Int128 src)
     return f128_ret(float128_sqrt(f128_in(src), &env->fp_status));
 }
 
-#define GEN_FCMP(name, size, reg1, reg2, FS, E)                         \
-    target_ulong glue(helper_, name) (CPUSPARCState *env)               \
+#define GEN_FCMP(name, size, FS, E)                                     \
+    target_ulong glue(helper_, name) (CPUSPARCState *env,               \
+                                      Int128 src1, Int128 src2)         \
     {                                                                   \
+        float128 reg1 = f128_in(src1);                                  \
+        float128 reg2 = f128_in(src2);                                  \
         FloatRelation ret;                                              \
         target_ulong fsr;                                               \
         if (E) {                                                        \
@@ -316,33 +319,33 @@ GEN_FCMP_T(fcmpd, float64, 0, 0);
 GEN_FCMP_T(fcmpes, float32, 0, 1);
 GEN_FCMP_T(fcmped, float64, 0, 1);
 
-GEN_FCMP(fcmpq, float128, QT0, QT1, 0, 0);
-GEN_FCMP(fcmpeq, float128, QT0, QT1, 0, 1);
+GEN_FCMP(fcmpq, float128, 0, 0);
+GEN_FCMP(fcmpeq, float128, 0, 1);
 
 #ifdef TARGET_SPARC64
 GEN_FCMP_T(fcmps_fcc1, float32, 22, 0);
 GEN_FCMP_T(fcmpd_fcc1, float64, 22, 0);
-GEN_FCMP(fcmpq_fcc1, float128, QT0, QT1, 22, 0);
+GEN_FCMP(fcmpq_fcc1, float128, 22, 0);
 
 GEN_FCMP_T(fcmps_fcc2, float32, 24, 0);
 GEN_FCMP_T(fcmpd_fcc2, float64, 24, 0);
-GEN_FCMP(fcmpq_fcc2, float128, QT0, QT1, 24, 0);
+GEN_FCMP(fcmpq_fcc2, float128, 24, 0);
 
 GEN_FCMP_T(fcmps_fcc3, float32, 26, 0);
 GEN_FCMP_T(fcmpd_fcc3, float64, 26, 0);
-GEN_FCMP(fcmpq_fcc3, float128, QT0, QT1, 26, 0);
+GEN_FCMP(fcmpq_fcc3, float128, 26, 0);
 
 GEN_FCMP_T(fcmpes_fcc1, float32, 22, 1);
 GEN_FCMP_T(fcmped_fcc1, float64, 22, 1);
-GEN_FCMP(fcmpeq_fcc1, float128, QT0, QT1, 22, 1);
+GEN_FCMP(fcmpeq_fcc1, float128, 22, 1);
 
 GEN_FCMP_T(fcmpes_fcc2, float32, 24, 1);
 GEN_FCMP_T(fcmped_fcc2, float64, 24, 1);
-GEN_FCMP(fcmpeq_fcc2, float128, QT0, QT1, 24, 1);
+GEN_FCMP(fcmpeq_fcc2, float128, 24, 1);
 
 GEN_FCMP_T(fcmpes_fcc3, float32, 26, 1);
 GEN_FCMP_T(fcmped_fcc3, float64, 26, 1);
-GEN_FCMP(fcmpeq_fcc3, float128, QT0, QT1, 26, 1);
+GEN_FCMP(fcmpeq_fcc3, float128, 26, 1);
 #endif
 #undef GEN_FCMP_T
 #undef GEN_FCMP
diff --git a/target/sparc/translate.c b/target/sparc/translate.c
index ba0b7f32d2..d4eea47c33 100644
--- a/target/sparc/translate.c
+++ b/target/sparc/translate.c
@@ -276,22 +276,6 @@ static void gen_store_fpr_Q(DisasContext *dc, unsigned int dst, TCGv_i128 v)
     gen_update_fprs_dirty(dc, dst);
 }
 
-static void gen_op_load_fpr_QT0(unsigned int src)
-{
-    tcg_gen_st_i64(cpu_fpr[src / 2], tcg_env, offsetof(CPUSPARCState, qt0) +
-                   offsetof(CPU_QuadU, ll.upper));
-    tcg_gen_st_i64(cpu_fpr[src/2 + 1], tcg_env, offsetof(CPUSPARCState, qt0) +
-                   offsetof(CPU_QuadU, ll.lower));
-}
-
-static void gen_op_load_fpr_QT1(unsigned int src)
-{
-    tcg_gen_st_i64(cpu_fpr[src / 2], tcg_env, offsetof(CPUSPARCState, qt1) +
-                   offsetof(CPU_QuadU, ll.upper));
-    tcg_gen_st_i64(cpu_fpr[src/2 + 1], tcg_env, offsetof(CPUSPARCState, qt1) +
-                   offsetof(CPU_QuadU, ll.lower));
-}
-
 static void gen_op_store_QT0_fpr(unsigned int dst)
 {
     tcg_gen_ld_i64(cpu_fpr[dst / 2], tcg_env, offsetof(CPUSPARCState, qt0) +
@@ -1319,20 +1303,20 @@ static void gen_op_fcmpd(int fccno, TCGv_i64 r_rs1, TCGv_i64 r_rs2)
     }
 }
 
-static void gen_op_fcmpq(int fccno)
+static void gen_op_fcmpq(int fccno, TCGv_i128 r_rs1, TCGv_i128 r_rs2)
 {
     switch (fccno) {
     case 0:
-        gen_helper_fcmpq(cpu_fsr, tcg_env);
+        gen_helper_fcmpq(cpu_fsr, tcg_env, r_rs1, r_rs2);
         break;
     case 1:
-        gen_helper_fcmpq_fcc1(cpu_fsr, tcg_env);
+        gen_helper_fcmpq_fcc1(cpu_fsr, tcg_env, r_rs1, r_rs2);
         break;
     case 2:
-        gen_helper_fcmpq_fcc2(cpu_fsr, tcg_env);
+        gen_helper_fcmpq_fcc2(cpu_fsr, tcg_env, r_rs1, r_rs2);
         break;
     case 3:
-        gen_helper_fcmpq_fcc3(cpu_fsr, tcg_env);
+        gen_helper_fcmpq_fcc3(cpu_fsr, tcg_env, r_rs1, r_rs2);
         break;
     }
 }
@@ -1373,20 +1357,20 @@ static void gen_op_fcmped(int fccno, TCGv_i64 r_rs1, TCGv_i64 r_rs2)
     }
 }
 
-static void gen_op_fcmpeq(int fccno)
+static void gen_op_fcmpeq(int fccno, TCGv_i128 r_rs1, TCGv_i128 r_rs2)
 {
     switch (fccno) {
     case 0:
-        gen_helper_fcmpeq(cpu_fsr, tcg_env);
+        gen_helper_fcmpeq(cpu_fsr, tcg_env, r_rs1, r_rs2);
         break;
     case 1:
-        gen_helper_fcmpeq_fcc1(cpu_fsr, tcg_env);
+        gen_helper_fcmpeq_fcc1(cpu_fsr, tcg_env, r_rs1, r_rs2);
         break;
     case 2:
-        gen_helper_fcmpeq_fcc2(cpu_fsr, tcg_env);
+        gen_helper_fcmpeq_fcc2(cpu_fsr, tcg_env, r_rs1, r_rs2);
         break;
     case 3:
-        gen_helper_fcmpeq_fcc3(cpu_fsr, tcg_env);
+        gen_helper_fcmpeq_fcc3(cpu_fsr, tcg_env, r_rs1, r_rs2);
         break;
     }
 }
@@ -1403,9 +1387,9 @@ static void gen_op_fcmpd(int fccno, TCGv_i64 r_rs1, TCGv_i64 r_rs2)
     gen_helper_fcmpd(cpu_fsr, tcg_env, r_rs1, r_rs2);
 }
 
-static void gen_op_fcmpq(int fccno)
+static void gen_op_fcmpq(int fccno, TCGv_i128 r_rs1, TCGv_i128 r_rs2)
 {
-    gen_helper_fcmpq(cpu_fsr, tcg_env);
+    gen_helper_fcmpq(cpu_fsr, tcg_env, r_rs1, r_rs2);
 }
 
 static void gen_op_fcmpes(int fccno, TCGv r_rs1, TCGv r_rs2)
@@ -1418,9 +1402,9 @@ static void gen_op_fcmped(int fccno, TCGv_i64 r_rs1, TCGv_i64 r_rs2)
     gen_helper_fcmped(cpu_fsr, tcg_env, r_rs1, r_rs2);
 }
 
-static void gen_op_fcmpeq(int fccno)
+static void gen_op_fcmpeq(int fccno, TCGv_i128 r_rs1, TCGv_i128 r_rs2)
 {
-    gen_helper_fcmpeq(cpu_fsr, tcg_env);
+    gen_helper_fcmpeq(cpu_fsr, tcg_env, r_rs1, r_rs2);
 }
 #endif
 
@@ -5144,6 +5128,8 @@ TRANS(FCMPEd, ALL, do_fcmpd, a, true)
 
 static bool do_fcmpq(DisasContext *dc, arg_FCMPq *a, bool e)
 {
+    TCGv_i128 src1, src2;
+
     if (avail_32(dc) && a->cc != 0) {
         return false;
     }
@@ -5155,12 +5141,12 @@ static bool do_fcmpq(DisasContext *dc, arg_FCMPq *a, bool e)
     }
 
     gen_op_clear_ieee_excp_and_FTT();
-    gen_op_load_fpr_QT0(QFPREG(a->rs1));
-    gen_op_load_fpr_QT1(QFPREG(a->rs2));
+    src1 = gen_load_fpr_Q(dc, a->rs1);
+    src2 = gen_load_fpr_Q(dc, a->rs2);
     if (e) {
-        gen_op_fcmpeq(a->cc);
+        gen_op_fcmpeq(a->cc, src1, src2);
     } else {
-        gen_op_fcmpq(a->cc);
+        gen_op_fcmpq(a->cc, src1, src2);
     }
     return advance_pc(dc);
 }
-- 
2.34.1

[PATCH 11/22] target/sparc: Use i128 for FsTOq, FiTOq

[Richard Henderson]

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/sparc/helper.h     | 4 ++--
 target/sparc/fop_helper.c | 8 ++++----
 target/sparc/translate.c  | 9 +++++----
 3 files changed, 11 insertions(+), 10 deletions(-)

diff --git a/target/sparc/helper.h b/target/sparc/helper.h
index 7caae9a441..5e93342583 100644
--- a/target/sparc/helper.h
+++ b/target/sparc/helper.h
@@ -87,7 +87,7 @@ DEF_HELPER_FLAGS_3(fsmuld, TCG_CALL_NO_RWG, f64, env, f32, f32)
 DEF_HELPER_FLAGS_3(fdmulq, TCG_CALL_NO_RWG, void, env, f64, f64)
 
 DEF_HELPER_FLAGS_2(fitod, TCG_CALL_NO_RWG_SE, f64, env, s32)
-DEF_HELPER_FLAGS_2(fitoq, TCG_CALL_NO_RWG, void, env, s32)
+DEF_HELPER_FLAGS_2(fitoq, TCG_CALL_NO_RWG, i128, env, s32)
 
 DEF_HELPER_FLAGS_2(fitos, TCG_CALL_NO_RWG, f32, env, s32)
 
@@ -99,7 +99,7 @@ DEF_HELPER_FLAGS_2(fxtoq, TCG_CALL_NO_RWG, void, env, s64)
 DEF_HELPER_FLAGS_2(fdtos, TCG_CALL_NO_RWG, f32, env, f64)
 DEF_HELPER_FLAGS_2(fstod, TCG_CALL_NO_RWG, f64, env, f32)
 DEF_HELPER_FLAGS_2(fqtos, TCG_CALL_NO_RWG, f32, env, i128)
-DEF_HELPER_FLAGS_2(fstoq, TCG_CALL_NO_RWG, void, env, f32)
+DEF_HELPER_FLAGS_2(fstoq, TCG_CALL_NO_RWG, i128, env, f32)
 DEF_HELPER_FLAGS_2(fqtod, TCG_CALL_NO_RWG, f64, env, i128)
 DEF_HELPER_FLAGS_2(fdtoq, TCG_CALL_NO_RWG, void, env, f64)
 DEF_HELPER_FLAGS_2(fstoi, TCG_CALL_NO_RWG, s32, env, f32)
diff --git a/target/sparc/fop_helper.c b/target/sparc/fop_helper.c
index faf75e651f..c7dc835d28 100644
--- a/target/sparc/fop_helper.c
+++ b/target/sparc/fop_helper.c
@@ -147,9 +147,9 @@ float64 helper_fitod(CPUSPARCState *env, int32_t src)
     return int32_to_float64(src, &env->fp_status);
 }
 
-void helper_fitoq(CPUSPARCState *env, int32_t src)
+Int128 helper_fitoq(CPUSPARCState *env, int32_t src)
 {
-    QT0 = int32_to_float128(src, &env->fp_status);
+    return f128_ret(int32_to_float128(src, &env->fp_status));
 }
 
 #ifdef TARGET_SPARC64
@@ -185,9 +185,9 @@ float32 helper_fqtos(CPUSPARCState *env, Int128 src)
     return float128_to_float32(f128_in(src), &env->fp_status);
 }
 
-void helper_fstoq(CPUSPARCState *env, float32 src)
+Int128 helper_fstoq(CPUSPARCState *env, float32 src)
 {
-    QT0 = float32_to_float128(src, &env->fp_status);
+    return f128_ret(float32_to_float128(src, &env->fp_status));
 }
 
 float64 helper_fqtod(CPUSPARCState *env, Int128 src)
diff --git a/target/sparc/translate.c b/target/sparc/translate.c
index d4eea47c33..1f96990316 100644
--- a/target/sparc/translate.c
+++ b/target/sparc/translate.c
@@ -4726,9 +4726,10 @@ TRANS(FqTOd, ALL, do_env_dq, a, gen_helper_fqtod)
 TRANS(FqTOx, 64, do_env_dq, a, gen_helper_fqtox)
 
 static bool do_env_qf(DisasContext *dc, arg_r_r *a,
-                      void (*func)(TCGv_env, TCGv_i32))
+                      void (*func)(TCGv_i128, TCGv_env, TCGv_i32))
 {
     TCGv_i32 src;
+    TCGv_i128 dst;
 
     if (gen_trap_ifnofpu(dc)) {
         return true;
@@ -4739,9 +4740,9 @@ static bool do_env_qf(DisasContext *dc, arg_r_r *a,
 
     gen_op_clear_ieee_excp_and_FTT();
     src = gen_load_fpr_F(dc, a->rs);
-    func(tcg_env, src);
-    gen_op_store_QT0_fpr(QFPREG(a->rd));
-    gen_update_fprs_dirty(dc, QFPREG(a->rd));
+    dst = tcg_temp_new_i128();
+    func(dst, tcg_env, src);
+    gen_store_fpr_Q(dc, a->rd, dst);
     return advance_pc(dc);
 }
 
-- 
2.34.1

[PATCH 12/22] target/sparc: Use i128 for FdTOq, FxTOq

[Richard Henderson]

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/sparc/helper.h     | 4 ++--
 target/sparc/fop_helper.c | 8 ++++----
 target/sparc/translate.c  | 9 +++++----
 3 files changed, 11 insertions(+), 10 deletions(-)

diff --git a/target/sparc/helper.h b/target/sparc/helper.h
index 5e93342583..20f67f89b0 100644
--- a/target/sparc/helper.h
+++ b/target/sparc/helper.h
@@ -94,14 +94,14 @@ DEF_HELPER_FLAGS_2(fitos, TCG_CALL_NO_RWG, f32, env, s32)
 #ifdef TARGET_SPARC64
 DEF_HELPER_FLAGS_2(fxtos, TCG_CALL_NO_RWG, f32, env, s64)
 DEF_HELPER_FLAGS_2(fxtod, TCG_CALL_NO_RWG, f64, env, s64)
-DEF_HELPER_FLAGS_2(fxtoq, TCG_CALL_NO_RWG, void, env, s64)
+DEF_HELPER_FLAGS_2(fxtoq, TCG_CALL_NO_RWG, i128, env, s64)
 #endif
 DEF_HELPER_FLAGS_2(fdtos, TCG_CALL_NO_RWG, f32, env, f64)
 DEF_HELPER_FLAGS_2(fstod, TCG_CALL_NO_RWG, f64, env, f32)
 DEF_HELPER_FLAGS_2(fqtos, TCG_CALL_NO_RWG, f32, env, i128)
 DEF_HELPER_FLAGS_2(fstoq, TCG_CALL_NO_RWG, i128, env, f32)
 DEF_HELPER_FLAGS_2(fqtod, TCG_CALL_NO_RWG, f64, env, i128)
-DEF_HELPER_FLAGS_2(fdtoq, TCG_CALL_NO_RWG, void, env, f64)
+DEF_HELPER_FLAGS_2(fdtoq, TCG_CALL_NO_RWG, i128, env, f64)
 DEF_HELPER_FLAGS_2(fstoi, TCG_CALL_NO_RWG, s32, env, f32)
 DEF_HELPER_FLAGS_2(fdtoi, TCG_CALL_NO_RWG, s32, env, f64)
 DEF_HELPER_FLAGS_2(fqtoi, TCG_CALL_NO_RWG, s32, env, i128)
diff --git a/target/sparc/fop_helper.c b/target/sparc/fop_helper.c
index c7dc835d28..9a0110e201 100644
--- a/target/sparc/fop_helper.c
+++ b/target/sparc/fop_helper.c
@@ -163,9 +163,9 @@ float64 helper_fxtod(CPUSPARCState *env, int64_t src)
     return int64_to_float64(src, &env->fp_status);
 }
 
-void helper_fxtoq(CPUSPARCState *env, int64_t src)
+Int128 helper_fxtoq(CPUSPARCState *env, int64_t src)
 {
-    QT0 = int64_to_float128(src, &env->fp_status);
+    return f128_ret(int64_to_float128(src, &env->fp_status));
 }
 #endif
 
@@ -195,9 +195,9 @@ float64 helper_fqtod(CPUSPARCState *env, Int128 src)
     return float128_to_float64(f128_in(src), &env->fp_status);
 }
 
-void helper_fdtoq(CPUSPARCState *env, float64 src)
+Int128 helper_fdtoq(CPUSPARCState *env, float64 src)
 {
-    QT0 = float64_to_float128(src, &env->fp_status);
+    return f128_ret(float64_to_float128(src, &env->fp_status));
 }
 
 /* Float to integer conversion.  */
diff --git a/target/sparc/translate.c b/target/sparc/translate.c
index 1f96990316..e01cbc5bcb 100644
--- a/target/sparc/translate.c
+++ b/target/sparc/translate.c
@@ -4750,9 +4750,10 @@ TRANS(FiTOq, ALL, do_env_qf, a, gen_helper_fitoq)
 TRANS(FsTOq, ALL, do_env_qf, a, gen_helper_fstoq)
 
 static bool do_env_qd(DisasContext *dc, arg_r_r *a,
-                      void (*func)(TCGv_env, TCGv_i64))
+                      void (*func)(TCGv_i128, TCGv_env, TCGv_i64))
 {
     TCGv_i64 src;
+    TCGv_i128 dst;
 
     if (gen_trap_ifnofpu(dc)) {
         return true;
@@ -4763,9 +4764,9 @@ static bool do_env_qd(DisasContext *dc, arg_r_r *a,
 
     gen_op_clear_ieee_excp_and_FTT();
     src = gen_load_fpr_D(dc, a->rs);
-    func(tcg_env, src);
-    gen_op_store_QT0_fpr(QFPREG(a->rd));
-    gen_update_fprs_dirty(dc, QFPREG(a->rd));
+    dst = tcg_temp_new_i128();
+    func(dst, tcg_env, src);
+    gen_store_fpr_Q(dc, a->rd, dst);
     return advance_pc(dc);
 }
 
-- 
2.34.1

[PATCH 13/22] target/sparc: Use i128 for Fdmulq

[Richard Henderson]

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/sparc/helper.h     |  2 +-
 target/sparc/fop_helper.c |  8 ++++----
 target/sparc/translate.c  | 15 ++++-----------
 3 files changed, 9 insertions(+), 16 deletions(-)

diff --git a/target/sparc/helper.h b/target/sparc/helper.h
index 20f67f89b0..f7aeb31169 100644
--- a/target/sparc/helper.h
+++ b/target/sparc/helper.h
@@ -84,7 +84,7 @@ DEF_HELPER_FLAGS_3(fmuls, TCG_CALL_NO_RWG, f32, env, f32, f32)
 DEF_HELPER_FLAGS_3(fdivs, TCG_CALL_NO_RWG, f32, env, f32, f32)
 
 DEF_HELPER_FLAGS_3(fsmuld, TCG_CALL_NO_RWG, f64, env, f32, f32)
-DEF_HELPER_FLAGS_3(fdmulq, TCG_CALL_NO_RWG, void, env, f64, f64)
+DEF_HELPER_FLAGS_3(fdmulq, TCG_CALL_NO_RWG, i128, env, f64, f64)
 
 DEF_HELPER_FLAGS_2(fitod, TCG_CALL_NO_RWG_SE, f64, env, s32)
 DEF_HELPER_FLAGS_2(fitoq, TCG_CALL_NO_RWG, i128, env, s32)
diff --git a/target/sparc/fop_helper.c b/target/sparc/fop_helper.c
index 9a0110e201..cd9b212597 100644
--- a/target/sparc/fop_helper.c
+++ b/target/sparc/fop_helper.c
@@ -129,11 +129,11 @@ float64 helper_fsmuld(CPUSPARCState *env, float32 src1, float32 src2)
                        &env->fp_status);
 }
 
-void helper_fdmulq(CPUSPARCState *env, float64 src1, float64 src2)
+Int128 helper_fdmulq(CPUSPARCState *env, float64 src1, float64 src2)
 {
-    QT0 = float128_mul(float64_to_float128(src1, &env->fp_status),
-                       float64_to_float128(src2, &env->fp_status),
-                       &env->fp_status);
+    return f128_ret(float128_mul(float64_to_float128(src1, &env->fp_status),
+                                 float64_to_float128(src2, &env->fp_status),
+                                 &env->fp_status));
 }
 
 /* Integer to float conversion.  */
diff --git a/target/sparc/translate.c b/target/sparc/translate.c
index e01cbc5bcb..99482df256 100644
--- a/target/sparc/translate.c
+++ b/target/sparc/translate.c
@@ -276,14 +276,6 @@ static void gen_store_fpr_Q(DisasContext *dc, unsigned int dst, TCGv_i128 v)
     gen_update_fprs_dirty(dc, dst);
 }
 
-static void gen_op_store_QT0_fpr(unsigned int dst)
-{
-    tcg_gen_ld_i64(cpu_fpr[dst / 2], tcg_env, offsetof(CPUSPARCState, qt0) +
-                   offsetof(CPU_QuadU, ll.upper));
-    tcg_gen_ld_i64(cpu_fpr[dst/2 + 1], tcg_env, offsetof(CPUSPARCState, qt0) +
-                   offsetof(CPU_QuadU, ll.lower));
-}
-
 /* moves */
 #ifdef CONFIG_USER_ONLY
 #define supervisor(dc) 0
@@ -4992,6 +4984,7 @@ TRANS(FDIVq, ALL, do_env_qqq, a, gen_helper_fdivq)
 static bool trans_FdMULq(DisasContext *dc, arg_r_r_r *a)
 {
     TCGv_i64 src1, src2;
+    TCGv_i128 dst;
 
     if (gen_trap_ifnofpu(dc)) {
         return true;
@@ -5003,10 +4996,10 @@ static bool trans_FdMULq(DisasContext *dc, arg_r_r_r *a)
     gen_op_clear_ieee_excp_and_FTT();
     src1 = gen_load_fpr_D(dc, a->rs1);
     src2 = gen_load_fpr_D(dc, a->rs2);
-    gen_helper_fdmulq(tcg_env, src1, src2);
+    dst = tcg_temp_new_i128();
+    gen_helper_fdmulq(dst, tcg_env, src1, src2);
     gen_helper_check_ieee_exceptions(cpu_fsr, tcg_env);
-    gen_op_store_QT0_fpr(QFPREG(a->rd));
-    gen_update_fprs_dirty(dc, QFPREG(a->rd));
+    gen_store_fpr_Q(dc, a->rd, dst);
     return advance_pc(dc);
 }
 
-- 
2.34.1

[PATCH 14/22] target/sparc: Remove qt0, qt1 temporaries

[Richard Henderson]

These are no longer used for passing data to/from helpers.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/sparc/cpu.h         | 2 --
 target/sparc/fop_helper.c  | 3 ---
 target/sparc/ldst_helper.c | 3 ---
 3 files changed, 8 deletions(-)

diff --git a/target/sparc/cpu.h b/target/sparc/cpu.h
index 3e361a5b75..446b38f3df 100644
--- a/target/sparc/cpu.h
+++ b/target/sparc/cpu.h
@@ -509,8 +509,6 @@ struct CPUArchState {
     uint64_t mmubpregs[4];
     uint64_t prom_addr;
 #endif
-    /* temporary float registers */
-    float128 qt0, qt1;
     float_status fp_status;
 #if defined(TARGET_SPARC64)
 #define MAXTL_MAX 8
diff --git a/target/sparc/fop_helper.c b/target/sparc/fop_helper.c
index cd9b212597..7353a61237 100644
--- a/target/sparc/fop_helper.c
+++ b/target/sparc/fop_helper.c
@@ -23,9 +23,6 @@
 #include "exec/helper-proto.h"
 #include "fpu/softfloat.h"
 
-#define QT0 (env->qt0)
-#define QT1 (env->qt1)
-
 static inline float128 f128_in(Int128 i)
 {
     union {
diff --git a/target/sparc/ldst_helper.c b/target/sparc/ldst_helper.c
index 09066d5487..fe984d44d7 100644
--- a/target/sparc/ldst_helper.c
+++ b/target/sparc/ldst_helper.c
@@ -66,9 +66,6 @@
 #endif
 #endif
 
-#define QT0 (env->qt0)
-#define QT1 (env->qt1)
-
 #if defined(TARGET_SPARC64) && !defined(CONFIG_USER_ONLY)
 /* Calculates TSB pointer value for fault page size
  * UltraSPARC IIi has fixed sizes (8k or 64k) for the page pointers
-- 
2.34.1

[PATCH 15/22] target/sparc: Introduce cpu_get_fsr, cpu_put_fsr

[Richard Henderson]

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/sparc/cpu.h          |  4 +++-
 target/sparc/helper.h       |  1 +
 linux-user/sparc/cpu_loop.c |  2 +-
 linux-user/sparc/signal.c   | 14 +++++++++-----
 target/sparc/cpu.c          |  5 +++--
 target/sparc/fop_helper.c   | 21 ++++++++++++++++++--
 target/sparc/gdbstub.c      |  8 ++++----
 target/sparc/machine.c      | 38 +++++++++++++++++++++++++++++++++++--
 target/sparc/translate.c    |  7 ++++++-
 9 files changed, 82 insertions(+), 18 deletions(-)

diff --git a/target/sparc/cpu.h b/target/sparc/cpu.h
index 446b38f3df..33c7d31fef 100644
--- a/target/sparc/cpu.h
+++ b/target/sparc/cpu.h
@@ -605,7 +605,9 @@ void sparc_restore_state_to_opc(CPUState *cs,
                                 const TranslationBlock *tb,
                                 const uint64_t *data);
 
-/* cpu-exec.c */
+/* fop_helper.c */
+target_ulong cpu_get_fsr(CPUSPARCState *);
+void cpu_put_fsr(CPUSPARCState *, target_ulong);
 
 /* win_helper.c */
 target_ulong cpu_get_psr(CPUSPARCState *env1);
diff --git a/target/sparc/helper.h b/target/sparc/helper.h
index f7aeb31169..cc8db50d75 100644
--- a/target/sparc/helper.h
+++ b/target/sparc/helper.h
@@ -36,6 +36,7 @@ DEF_HELPER_FLAGS_4(ld_asi, TCG_CALL_NO_WG, i64, env, tl, int, i32)
 DEF_HELPER_FLAGS_5(st_asi, TCG_CALL_NO_WG, void, env, tl, i64, int, i32)
 #endif
 DEF_HELPER_FLAGS_1(check_ieee_exceptions, TCG_CALL_NO_WG, tl, env)
+DEF_HELPER_FLAGS_1(get_fsr, TCG_CALL_NO_WG_SE, tl, env)
 DEF_HELPER_FLAGS_2(set_fsr, TCG_CALL_NO_RWG, void, env, tl)
 DEF_HELPER_FLAGS_2(fsqrts, TCG_CALL_NO_RWG, f32, env, f32)
 DEF_HELPER_FLAGS_2(fsqrtd, TCG_CALL_NO_RWG, f64, env, f64)
diff --git a/linux-user/sparc/cpu_loop.c b/linux-user/sparc/cpu_loop.c
index 3c1bde00dd..50424a54df 100644
--- a/linux-user/sparc/cpu_loop.c
+++ b/linux-user/sparc/cpu_loop.c
@@ -293,7 +293,7 @@ void cpu_loop (CPUSPARCState *env)
         case TT_FP_EXCP:
             {
                 int code = TARGET_FPE_FLTUNK;
-                target_ulong fsr = env->fsr;
+                target_ulong fsr = cpu_get_fsr(env);
 
                 if ((fsr & FSR_FTT_MASK) == FSR_FTT_IEEE_EXCP) {
                     if (fsr & FSR_NVC) {
diff --git a/linux-user/sparc/signal.c b/linux-user/sparc/signal.c
index dfcae707e0..c2dc1000e2 100644
--- a/linux-user/sparc/signal.c
+++ b/linux-user/sparc/signal.c
@@ -199,20 +199,21 @@ static void save_fpu(struct target_siginfo_fpu *fpu, CPUSPARCState *env)
     for (i = 0; i < 32; ++i) {
         __put_user(env->fpr[i].ll, &fpu->si_double_regs[i]);
     }
-    __put_user(env->fsr, &fpu->si_fsr);
+    __put_user(cpu_get_fsr(env), &fpu->si_fsr);
     __put_user(env->gsr, &fpu->si_gsr);
     __put_user(env->fprs, &fpu->si_fprs);
 #else
     for (i = 0; i < 16; ++i) {
         __put_user(env->fpr[i].ll, &fpu->si_double_regs[i]);
     }
-    __put_user(env->fsr, &fpu->si_fsr);
+    __put_user(cpu_get_fsr(env), &fpu->si_fsr);
     __put_user(0, &fpu->si_fpqdepth);
 #endif
 }
 
 static void restore_fpu(struct target_siginfo_fpu *fpu, CPUSPARCState *env)
 {
+    target_ulong fsr;
     int i;
 
 #ifdef TARGET_SPARC64
@@ -230,15 +231,16 @@ static void restore_fpu(struct target_siginfo_fpu *fpu, CPUSPARCState *env)
             __get_user(env->fpr[i].ll, &fpu->si_double_regs[i]);
         }
     }
-    __get_user(env->fsr, &fpu->si_fsr);
     __get_user(env->gsr, &fpu->si_gsr);
     env->fprs |= fprs;
 #else
     for (i = 0; i < 16; ++i) {
         __get_user(env->fpr[i].ll, &fpu->si_double_regs[i]);
     }
-    __get_user(env->fsr, &fpu->si_fsr);
 #endif
+
+    __get_user(fsr, &fpu->si_fsr);
+    cpu_put_fsr(env, fsr);
 }
 
 #ifdef TARGET_ARCH_HAS_SETUP_FRAME
@@ -662,6 +664,7 @@ void sparc64_set_context(CPUSPARCState *env)
     __get_user(fenab, &(fpup->mcfpu_enab));
     if (fenab) {
         abi_ulong fprs;
+        abi_ulong fsr;
 
         /*
          * We use the FPRS from the guest only in deciding whether
@@ -690,7 +693,8 @@ void sparc64_set_context(CPUSPARCState *env)
                 __get_user(env->fpr[i].ll, &(fpup->mcfpu_fregs.dregs[i]));
             }
         }
-        __get_user(env->fsr, &(fpup->mcfpu_fsr));
+        __get_user(fsr, &(fpup->mcfpu_fsr));
+        cpu_put_fsr(env, fsr);
         __get_user(env->gsr, &(fpup->mcfpu_gsr));
     }
     unlock_user_struct(ucp, ucp_addr, 0);
diff --git a/target/sparc/cpu.c b/target/sparc/cpu.c
index befa7fc4eb..69dfa1dd4e 100644
--- a/target/sparc/cpu.c
+++ b/target/sparc/cpu.c
@@ -670,7 +670,7 @@ static void sparc_cpu_dump_state(CPUState *cs, FILE *f, int flags)
                  env->cansave, env->canrestore, env->otherwin, env->wstate,
                  env->cleanwin, env->nwindows - 1 - env->cwp);
     qemu_fprintf(f, "fsr: " TARGET_FMT_lx " y: " TARGET_FMT_lx " fprs: %016x\n",
-                 env->fsr, env->y, env->fprs);
+                 cpu_get_fsr(env), env->y, env->fprs);
 
 #else
     qemu_fprintf(f, "psr: %08x (icc: ", cpu_get_psr(env));
@@ -679,7 +679,7 @@ static void sparc_cpu_dump_state(CPUState *cs, FILE *f, int flags)
                  env->psrps ? 'P' : '-', env->psret ? 'E' : '-',
                  env->wim);
     qemu_fprintf(f, "fsr: " TARGET_FMT_lx " y: " TARGET_FMT_lx "\n",
-                 env->fsr, env->y);
+                 cpu_get_fsr(env), env->y);
 #endif
     qemu_fprintf(f, "\n");
 }
@@ -770,6 +770,7 @@ static void sparc_cpu_realizefn(DeviceState *dev, Error **errp)
     env->version |= env->def.maxtl << 8;
     env->version |= env->def.nwindows - 1;
 #endif
+    cpu_put_fsr(env, 0);
 
     cpu_exec_realizefn(cs, &local_err);
     if (local_err != NULL) {
diff --git a/target/sparc/fop_helper.c b/target/sparc/fop_helper.c
index 7353a61237..70b38011d2 100644
--- a/target/sparc/fop_helper.c
+++ b/target/sparc/fop_helper.c
@@ -347,10 +347,22 @@ GEN_FCMP(fcmpeq_fcc3, float128, 26, 1);
 #undef GEN_FCMP_T
 #undef GEN_FCMP
 
-static void set_fsr(CPUSPARCState *env, target_ulong fsr)
+target_ulong cpu_get_fsr(CPUSPARCState *env)
+{
+    return env->fsr;
+}
+
+target_ulong helper_get_fsr(CPUSPARCState *env)
+{
+    return cpu_get_fsr(env);
+}
+
+static void set_fsr_nonsplit(CPUSPARCState *env, target_ulong fsr)
 {
     int rnd_mode;
 
+    env->fsr = fsr;
+
     switch (fsr & FSR_RD_MASK) {
     case FSR_RD_NEAREST:
         rnd_mode = float_round_nearest_even;
@@ -369,7 +381,12 @@ static void set_fsr(CPUSPARCState *env, target_ulong fsr)
     set_float_rounding_mode(rnd_mode, &env->fp_status);
 }
 
+void cpu_put_fsr(CPUSPARCState *env, target_ulong fsr)
+{
+    set_fsr_nonsplit(env, fsr);
+}
+
 void helper_set_fsr(CPUSPARCState *env, target_ulong fsr)
 {
-    set_fsr(env, fsr);
+    set_fsr_nonsplit(env, fsr);
 }
diff --git a/target/sparc/gdbstub.c b/target/sparc/gdbstub.c
index a1c8fdc4d5..d1586b2392 100644
--- a/target/sparc/gdbstub.c
+++ b/target/sparc/gdbstub.c
@@ -64,7 +64,7 @@ int sparc_cpu_gdb_read_register(CPUState *cs, GByteArray *mem_buf, int n)
     case 69:
         return gdb_get_rega(mem_buf, env->npc);
     case 70:
-        return gdb_get_rega(mem_buf, env->fsr);
+        return gdb_get_rega(mem_buf, cpu_get_fsr(env));
     case 71:
         return gdb_get_rega(mem_buf, 0); /* csr */
     default:
@@ -94,7 +94,7 @@ int sparc_cpu_gdb_read_register(CPUState *cs, GByteArray *mem_buf, int n)
                                      ((env->pstate & 0xfff) << 8) |
                                      cpu_get_cwp64(env));
     case 83:
-        return gdb_get_regl(mem_buf, env->fsr);
+        return gdb_get_regl(mem_buf, cpu_get_fsr(env));
     case 84:
         return gdb_get_regl(mem_buf, env->fprs);
     case 85:
@@ -156,7 +156,7 @@ int sparc_cpu_gdb_write_register(CPUState *cs, uint8_t *mem_buf, int n)
             env->npc = tmp;
             break;
         case 70:
-            env->fsr = tmp;
+            cpu_put_fsr(env, tmp);
             break;
         default:
             return 0;
@@ -191,7 +191,7 @@ int sparc_cpu_gdb_write_register(CPUState *cs, uint8_t *mem_buf, int n)
             cpu_put_cwp64(env, tmp & 0xff);
             break;
         case 83:
-            env->fsr = tmp;
+            cpu_put_fsr(env, tmp);
             break;
         case 84:
             env->fprs = tmp;
diff --git a/target/sparc/machine.c b/target/sparc/machine.c
index 44dfc07014..e46f15adb8 100644
--- a/target/sparc/machine.c
+++ b/target/sparc/machine.c
@@ -83,6 +83,34 @@ static const VMStateInfo vmstate_psr = {
     .put = put_psr,
 };
 
+static int get_fsr(QEMUFile *f, void *opaque, size_t size,
+                   const VMStateField *field)
+{
+    SPARCCPU *cpu = opaque;
+    CPUSPARCState *env = &cpu->env;
+    target_ulong val = qemu_get_betl(f);
+
+    cpu_put_fsr(env, val);
+    return 0;
+}
+
+static int put_fsr(QEMUFile *f, void *opaque, size_t size,
+                   const VMStateField *field, JSONWriter *vmdesc)
+{
+    SPARCCPU *cpu = opaque;
+    CPUSPARCState *env = &cpu->env;
+    target_ulong val = cpu_get_fsr(env);
+
+    qemu_put_betl(f, val);
+    return 0;
+}
+
+static const VMStateInfo vmstate_fsr = {
+    .name = "fsr",
+    .get = get_fsr,
+    .put = put_fsr,
+};
+
 #ifdef TARGET_SPARC64
 static int get_xcc(QEMUFile *f, void *opaque, size_t size,
                    const VMStateField *field)
@@ -157,7 +185,6 @@ const VMStateDescription vmstate_sparc_cpu = {
         VMSTATE_UINTTL(env.npc, SPARCCPU),
         VMSTATE_UINTTL(env.y, SPARCCPU),
         {
-
             .name = "psr",
             .version_id = 0,
             .size = sizeof(uint32_t),
@@ -165,7 +192,14 @@ const VMStateDescription vmstate_sparc_cpu = {
             .flags = VMS_SINGLE,
             .offset = 0,
         },
-        VMSTATE_UINTTL(env.fsr, SPARCCPU),
+        {
+            .name = "fsr",
+            .version_id = 0,
+            .size = sizeof(target_ulong),
+            .info = &vmstate_fsr,
+            .flags = VMS_SINGLE,
+            .offset = 0,
+        },
         VMSTATE_UINTTL(env.tbr, SPARCCPU),
         VMSTATE_INT32(env.interrupt_index, SPARCCPU),
         VMSTATE_UINT32(env.pil_in, SPARCCPU),
diff --git a/target/sparc/translate.c b/target/sparc/translate.c
index 99482df256..1d5f36dafc 100644
--- a/target/sparc/translate.c
+++ b/target/sparc/translate.c
@@ -4417,13 +4417,18 @@ TRANS(LDXFSR, 64, do_ldfsr, a, MO_TEUQ, FSR_LDXFSR_MASK, FSR_LDXFSR_OLDMASK)
 static bool do_stfsr(DisasContext *dc, arg_r_r_ri *a, MemOp mop)
 {
     TCGv addr = gen_ldst_addr(dc, a->rs1, a->imm, a->rs2_or_imm);
+    TCGv fsr;
+
     if (addr == NULL) {
         return false;
     }
     if (gen_trap_ifnofpu(dc)) {
         return true;
     }
-    tcg_gen_qemu_st_tl(cpu_fsr, addr, dc->mem_idx, mop | MO_ALIGN);
+
+    fsr = tcg_temp_new();
+    gen_helper_get_fsr(fsr, tcg_env);
+    tcg_gen_qemu_st_tl(fsr, addr, dc->mem_idx, mop | MO_ALIGN);
     return advance_pc(dc);
 }
 
-- 
2.34.1

Re: [PATCH 15/22] target/sparc: Introduce cpu_get_fsr, cpu_put_fsr

[Philippe Mathieu-Daudé]

On 3/11/23 18:38, Richard Henderson wrote:
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
>   target/sparc/cpu.h          |  4 +++-
>   target/sparc/helper.h       |  1 +
>   linux-user/sparc/cpu_loop.c |  2 +-
>   linux-user/sparc/signal.c   | 14 +++++++++-----
>   target/sparc/cpu.c          |  5 +++--
>   target/sparc/fop_helper.c   | 21 ++++++++++++++++++--
>   target/sparc/gdbstub.c      |  8 ++++----
>   target/sparc/machine.c      | 38 +++++++++++++++++++++++++++++++++++--
>   target/sparc/translate.c    |  7 ++++++-
>   9 files changed, 82 insertions(+), 18 deletions(-)


> diff --git a/target/sparc/machine.c b/target/sparc/machine.c
> index 44dfc07014..e46f15adb8 100644
> --- a/target/sparc/machine.c
> +++ b/target/sparc/machine.c
> @@ -83,6 +83,34 @@ static const VMStateInfo vmstate_psr = {
>       .put = put_psr,
>   };
>   
> +static int get_fsr(QEMUFile *f, void *opaque, size_t size,
> +                   const VMStateField *field)
> +{
> +    SPARCCPU *cpu = opaque;
> +    CPUSPARCState *env = &cpu->env;
> +    target_ulong val = qemu_get_betl(f);
> +
> +    cpu_put_fsr(env, val);

Preferably directly using cpu_env(), otherwise:
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>

> +    return 0;
> +}
> +
> +static int put_fsr(QEMUFile *f, void *opaque, size_t size,
> +                   const VMStateField *field, JSONWriter *vmdesc)
> +{
> +    SPARCCPU *cpu = opaque;
> +    CPUSPARCState *env = &cpu->env;
> +    target_ulong val = cpu_get_fsr(env);

(Ditto)

> +
> +    qemu_put_betl(f, val);
> +    return 0;
> +}

[PATCH 16/22] target/split: Split ver from env->fsr

[Richard Henderson]

This field is read-only.  It is easier to store it separately
and merge it only upon read.

While we're at it, use FSR_VER_SHIFT to initialize fpu_version.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/sparc/cpu.h        |  3 +++
 target/sparc/cpu.c        | 27 +++++++++++++--------------
 target/sparc/fop_helper.c |  9 +++++++--
 3 files changed, 23 insertions(+), 16 deletions(-)

diff --git a/target/sparc/cpu.h b/target/sparc/cpu.h
index 33c7d31fef..8ff222595e 100644
--- a/target/sparc/cpu.h
+++ b/target/sparc/cpu.h
@@ -191,6 +191,9 @@ enum {
 #define FSR_NXC   (1ULL << 0)
 #define FSR_CEXC_MASK (FSR_NVC | FSR_OFC | FSR_UFC | FSR_DZC | FSR_NXC)
 
+#define FSR_VER_SHIFT  17
+#define FSR_VER_MASK   (7 << FSR_VER_SHIFT)
+
 #define FSR_FTT2   (1ULL << 16)
 #define FSR_FTT1   (1ULL << 15)
 #define FSR_FTT0   (1ULL << 14)
diff --git a/target/sparc/cpu.c b/target/sparc/cpu.c
index 69dfa1dd4e..bebfdbf313 100644
--- a/target/sparc/cpu.c
+++ b/target/sparc/cpu.c
@@ -368,7 +368,7 @@ static const sparc_def_t sparc_defs[] = {
     {
         .name = "Fujitsu MB86904",
         .iu_version = 0x04 << 24, /* Impl 0, ver 4 */
-        .fpu_version = 4 << 17, /* FPU version 4 (Meiko) */
+        .fpu_version = 4 << FSR_VER_SHIFT, /* FPU version 4 (Meiko) */
         .mmu_version = 0x04 << 24, /* Impl 0, ver 4 */
         .mmu_bm = 0x00004000,
         .mmu_ctpr_mask = 0x00ffffc0,
@@ -381,7 +381,7 @@ static const sparc_def_t sparc_defs[] = {
     {
         .name = "Fujitsu MB86907",
         .iu_version = 0x05 << 24, /* Impl 0, ver 5 */
-        .fpu_version = 4 << 17, /* FPU version 4 (Meiko) */
+        .fpu_version = 4 << FSR_VER_SHIFT, /* FPU version 4 (Meiko) */
         .mmu_version = 0x05 << 24, /* Impl 0, ver 5 */
         .mmu_bm = 0x00004000,
         .mmu_ctpr_mask = 0xffffffc0,
@@ -394,7 +394,7 @@ static const sparc_def_t sparc_defs[] = {
     {
         .name = "TI MicroSparc I",
         .iu_version = 0x41000000,
-        .fpu_version = 4 << 17,
+        .fpu_version = 4 << FSR_VER_SHIFT,
         .mmu_version = 0x41000000,
         .mmu_bm = 0x00004000,
         .mmu_ctpr_mask = 0x007ffff0,
@@ -407,7 +407,7 @@ static const sparc_def_t sparc_defs[] = {
     {
         .name = "TI MicroSparc II",
         .iu_version = 0x42000000,
-        .fpu_version = 4 << 17,
+        .fpu_version = 4 << FSR_VER_SHIFT,
         .mmu_version = 0x02000000,
         .mmu_bm = 0x00004000,
         .mmu_ctpr_mask = 0x00ffffc0,
@@ -420,7 +420,7 @@ static const sparc_def_t sparc_defs[] = {
     {
         .name = "TI MicroSparc IIep",
         .iu_version = 0x42000000,
-        .fpu_version = 4 << 17,
+        .fpu_version = 4 << FSR_VER_SHIFT,
         .mmu_version = 0x04000000,
         .mmu_bm = 0x00004000,
         .mmu_ctpr_mask = 0x00ffffc0,
@@ -433,7 +433,7 @@ static const sparc_def_t sparc_defs[] = {
     {
         .name = "TI SuperSparc 40", /* STP1020NPGA */
         .iu_version = 0x41000000, /* SuperSPARC 2.x */
-        .fpu_version = 0 << 17,
+        .fpu_version = 0 << FSR_VER_SHIFT,
         .mmu_version = 0x00000800, /* SuperSPARC 2.x, no MXCC */
         .mmu_bm = 0x00002000,
         .mmu_ctpr_mask = 0xffffffc0,
@@ -446,7 +446,7 @@ static const sparc_def_t sparc_defs[] = {
     {
         .name = "TI SuperSparc 50", /* STP1020PGA */
         .iu_version = 0x40000000, /* SuperSPARC 3.x */
-        .fpu_version = 0 << 17,
+        .fpu_version = 0 << FSR_VER_SHIFT,
         .mmu_version = 0x01000800, /* SuperSPARC 3.x, no MXCC */
         .mmu_bm = 0x00002000,
         .mmu_ctpr_mask = 0xffffffc0,
@@ -459,7 +459,7 @@ static const sparc_def_t sparc_defs[] = {
     {
         .name = "TI SuperSparc 51",
         .iu_version = 0x40000000, /* SuperSPARC 3.x */
-        .fpu_version = 0 << 17,
+        .fpu_version = 0 << FSR_VER_SHIFT,
         .mmu_version = 0x01000000, /* SuperSPARC 3.x, MXCC */
         .mmu_bm = 0x00002000,
         .mmu_ctpr_mask = 0xffffffc0,
@@ -473,7 +473,7 @@ static const sparc_def_t sparc_defs[] = {
     {
         .name = "TI SuperSparc 60", /* STP1020APGA */
         .iu_version = 0x40000000, /* SuperSPARC 3.x */
-        .fpu_version = 0 << 17,
+        .fpu_version = 0 << FSR_VER_SHIFT,
         .mmu_version = 0x01000800, /* SuperSPARC 3.x, no MXCC */
         .mmu_bm = 0x00002000,
         .mmu_ctpr_mask = 0xffffffc0,
@@ -486,7 +486,7 @@ static const sparc_def_t sparc_defs[] = {
     {
         .name = "TI SuperSparc 61",
         .iu_version = 0x44000000, /* SuperSPARC 3.x */
-        .fpu_version = 0 << 17,
+        .fpu_version = 0 << FSR_VER_SHIFT,
         .mmu_version = 0x01000000, /* SuperSPARC 3.x, MXCC */
         .mmu_bm = 0x00002000,
         .mmu_ctpr_mask = 0xffffffc0,
@@ -500,7 +500,7 @@ static const sparc_def_t sparc_defs[] = {
     {
         .name = "TI SuperSparc II",
         .iu_version = 0x40000000, /* SuperSPARC II 1.x */
-        .fpu_version = 0 << 17,
+        .fpu_version = 0 << FSR_VER_SHIFT,
         .mmu_version = 0x08000000, /* SuperSPARC II 1.x, MXCC */
         .mmu_bm = 0x00002000,
         .mmu_ctpr_mask = 0xffffffc0,
@@ -514,7 +514,7 @@ static const sparc_def_t sparc_defs[] = {
     {
         .name = "LEON2",
         .iu_version = 0xf2000000,
-        .fpu_version = 4 << 17, /* FPU version 4 (Meiko) */
+        .fpu_version = 4 << FSR_VER_SHIFT, /* FPU version 4 (Meiko) */
         .mmu_version = 0xf2000000,
         .mmu_bm = 0x00004000,
         .mmu_ctpr_mask = 0x007ffff0,
@@ -527,7 +527,7 @@ static const sparc_def_t sparc_defs[] = {
     {
         .name = "LEON3",
         .iu_version = 0xf3000000,
-        .fpu_version = 4 << 17, /* FPU version 4 (Meiko) */
+        .fpu_version = 4 << FSR_VER_SHIFT, /* FPU version 4 (Meiko) */
         .mmu_version = 0xf3000000,
         .mmu_bm = 0x00000000,
         .mmu_ctpr_mask = 0xfffffffc,
@@ -758,7 +758,6 @@ static void sparc_cpu_realizefn(DeviceState *dev, Error **errp)
 #endif
 
     env->version = env->def.iu_version;
-    env->fsr = env->def.fpu_version;
     env->nwindows = env->def.nwindows;
 #if !defined(TARGET_SPARC64)
     env->mmuregs[0] |= env->def.mmu_version;
diff --git a/target/sparc/fop_helper.c b/target/sparc/fop_helper.c
index 70b38011d2..22b412adb5 100644
--- a/target/sparc/fop_helper.c
+++ b/target/sparc/fop_helper.c
@@ -349,7 +349,12 @@ GEN_FCMP(fcmpeq_fcc3, float128, 26, 1);
 
 target_ulong cpu_get_fsr(CPUSPARCState *env)
 {
-    return env->fsr;
+    target_ulong fsr = env->fsr;
+
+    /* VER is kept completely separate until re-assembly. */
+    fsr |= env->def.fpu_version;
+
+    return fsr;
 }
 
 target_ulong helper_get_fsr(CPUSPARCState *env)
@@ -361,7 +366,7 @@ static void set_fsr_nonsplit(CPUSPARCState *env, target_ulong fsr)
 {
     int rnd_mode;
 
-    env->fsr = fsr;
+    env->fsr = fsr & ~FSR_VER_MASK;
 
     switch (fsr & FSR_RD_MASK) {
     case FSR_RD_NEAREST:
-- 
2.34.1

Re: [PATCH 16/22] target/split: Split ver from env->fsr

[Philippe Mathieu-Daudé]

On 3/11/23 18:38, Richard Henderson wrote:
> This field is read-only.  It is easier to store it separately
> and merge it only upon read.
> 
> While we're at it, use FSR_VER_SHIFT to initialize fpu_version.
> 
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
>   target/sparc/cpu.h        |  3 +++
>   target/sparc/cpu.c        | 27 +++++++++++++--------------
>   target/sparc/fop_helper.c |  9 +++++++--
>   3 files changed, 23 insertions(+), 16 deletions(-)

Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>

Re: [PATCH 16/22] target/split: Split ver from env->fsr

[BALATON Zoltan]

[-- Attachment #1: Type: text/plain, Size: 693 bytes --]

On Tue, 30 Jan 2024, Philippe Mathieu-Daudé wrote:
> On 3/11/23 18:38, Richard Henderson wrote:
>> This field is read-only.  It is easier to store it separately
>> and merge it only upon read.
>> 
>> While we're at it, use FSR_VER_SHIFT to initialize fpu_version.
>> 
>> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
>> ---
>>   target/sparc/cpu.h        |  3 +++
>>   target/sparc/cpu.c        | 27 +++++++++++++--------------
>>   target/sparc/fop_helper.c |  9 +++++++--
>>   3 files changed, 23 insertions(+), 16 deletions(-)
>
> Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>

Typo in subject: should be target/sparc not target/split.

Regards,
BALATON Zoltan

[PATCH 17/22] target/sparc: Clear cexc and ftt in do_check_ieee_exceptions

[Richard Henderson]

Don't do the clearing explicitly before each FPop,
rather do it as part of the rest of exception handling.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/sparc/fop_helper.c |  2 ++
 target/sparc/translate.c  | 16 ----------------
 2 files changed, 2 insertions(+), 16 deletions(-)

diff --git a/target/sparc/fop_helper.c b/target/sparc/fop_helper.c
index 22b412adb5..64f20e78f1 100644
--- a/target/sparc/fop_helper.c
+++ b/target/sparc/fop_helper.c
@@ -50,6 +50,8 @@ static target_ulong do_check_ieee_exceptions(CPUSPARCState *env, uintptr_t ra)
     target_ulong status = get_float_exception_flags(&env->fp_status);
     target_ulong fsr = env->fsr;
 
+    fsr &= FSR_FTT_CEXC_NMASK;
+
     if (unlikely(status)) {
         /* Keep exception flags clear for next time.  */
         set_float_exception_flags(0, &env->fp_status);
diff --git a/target/sparc/translate.c b/target/sparc/translate.c
index 1d5f36dafc..11d025f4ea 100644
--- a/target/sparc/translate.c
+++ b/target/sparc/translate.c
@@ -4524,7 +4524,6 @@ static bool do_env_ff(DisasContext *dc, arg_r_r *a,
         return true;
     }
 
-    gen_op_clear_ieee_excp_and_FTT();
     tmp = gen_load_fpr_F(dc, a->rs);
     func(tmp, tcg_env, tmp);
     gen_helper_check_ieee_exceptions(cpu_fsr, tcg_env);
@@ -4546,7 +4545,6 @@ static bool do_env_fd(DisasContext *dc, arg_r_r *a,
         return true;
     }
 
-    gen_op_clear_ieee_excp_and_FTT();
     dst = tcg_temp_new_i32();
     src = gen_load_fpr_D(dc, a->rs);
     func(dst, tcg_env, src);
@@ -4590,7 +4588,6 @@ static bool do_env_dd(DisasContext *dc, arg_r_r *a,
         return true;
     }
 
-    gen_op_clear_ieee_excp_and_FTT();
     dst = gen_dest_fpr_D(dc, a->rd);
     src = gen_load_fpr_D(dc, a->rs);
     func(dst, tcg_env, src);
@@ -4613,7 +4610,6 @@ static bool do_env_df(DisasContext *dc, arg_r_r *a,
         return true;
     }
 
-    gen_op_clear_ieee_excp_and_FTT();
     dst = gen_dest_fpr_D(dc, a->rd);
     src = gen_load_fpr_F(dc, a->rs);
     func(dst, tcg_env, src);
@@ -4661,8 +4657,6 @@ static bool do_env_qq(DisasContext *dc, arg_r_r *a,
         return true;
     }
 
-    gen_op_clear_ieee_excp_and_FTT();
-
     t = gen_load_fpr_Q(dc, a->rs);
     func(t, tcg_env, t);
     gen_helper_check_ieee_exceptions(cpu_fsr, tcg_env);
@@ -4685,7 +4679,6 @@ static bool do_env_fq(DisasContext *dc, arg_r_r *a,
         return true;
     }
 
-    gen_op_clear_ieee_excp_and_FTT();
     src = gen_load_fpr_Q(dc, a->rs);
     dst = tcg_temp_new_i32();
     func(dst, tcg_env, src);
@@ -4710,7 +4703,6 @@ static bool do_env_dq(DisasContext *dc, arg_r_r *a,
         return true;
     }
 
-    gen_op_clear_ieee_excp_and_FTT();
     src = gen_load_fpr_Q(dc, a->rs);
     dst = gen_dest_fpr_D(dc, a->rd);
     func(dst, tcg_env, src);
@@ -4808,7 +4800,6 @@ static bool do_env_fff(DisasContext *dc, arg_r_r_r *a,
         return true;
     }
 
-    gen_op_clear_ieee_excp_and_FTT();
     src1 = gen_load_fpr_F(dc, a->rs1);
     src2 = gen_load_fpr_F(dc, a->rs2);
     func(src1, tcg_env, src1, src2);
@@ -4903,7 +4894,6 @@ static bool do_env_ddd(DisasContext *dc, arg_r_r_r *a,
         return true;
     }
 
-    gen_op_clear_ieee_excp_and_FTT();
     dst = gen_dest_fpr_D(dc, a->rd);
     src1 = gen_load_fpr_D(dc, a->rs1);
     src2 = gen_load_fpr_D(dc, a->rs2);
@@ -4930,7 +4920,6 @@ static bool trans_FsMULd(DisasContext *dc, arg_r_r_r *a)
         return raise_unimpfpop(dc);
     }
 
-    gen_op_clear_ieee_excp_and_FTT();
     dst = gen_dest_fpr_D(dc, a->rd);
     src1 = gen_load_fpr_F(dc, a->rs1);
     src2 = gen_load_fpr_F(dc, a->rs2);
@@ -4972,7 +4961,6 @@ static bool do_env_qqq(DisasContext *dc, arg_r_r_r *a,
         return true;
     }
 
-    gen_op_clear_ieee_excp_and_FTT();
     src1 = gen_load_fpr_Q(dc, a->rs1);
     src2 = gen_load_fpr_Q(dc, a->rs2);
     func(src1, tcg_env, src1, src2);
@@ -4998,7 +4986,6 @@ static bool trans_FdMULq(DisasContext *dc, arg_r_r_r *a)
         return true;
     }
 
-    gen_op_clear_ieee_excp_and_FTT();
     src1 = gen_load_fpr_D(dc, a->rs1);
     src2 = gen_load_fpr_D(dc, a->rs2);
     dst = tcg_temp_new_i128();
@@ -5087,7 +5074,6 @@ static bool do_fcmps(DisasContext *dc, arg_FCMPs *a, bool e)
         return true;
     }
 
-    gen_op_clear_ieee_excp_and_FTT();
     src1 = gen_load_fpr_F(dc, a->rs1);
     src2 = gen_load_fpr_F(dc, a->rs2);
     if (e) {
@@ -5112,7 +5098,6 @@ static bool do_fcmpd(DisasContext *dc, arg_FCMPd *a, bool e)
         return true;
     }
 
-    gen_op_clear_ieee_excp_and_FTT();
     src1 = gen_load_fpr_D(dc, a->rs1);
     src2 = gen_load_fpr_D(dc, a->rs2);
     if (e) {
@@ -5140,7 +5125,6 @@ static bool do_fcmpq(DisasContext *dc, arg_FCMPq *a, bool e)
         return true;
     }
 
-    gen_op_clear_ieee_excp_and_FTT();
     src1 = gen_load_fpr_Q(dc, a->rs1);
     src2 = gen_load_fpr_Q(dc, a->rs2);
     if (e) {
-- 
2.34.1

[PATCH 18/22] target/sparc: Merge check_ieee_exceptions with FPop helpers

[Richard Henderson]

If an exception is to be raised, the destination fp register
should be unmodified.  The current implementation is incorrect,
in that double results will be written back before calling
gen_helper_check_ieee_exceptions, despite the placement of
gen_store_fpr_D, since gen_dest_fpr_D returns cpu_fpr[].

We can simplify the entire implementation by having each
FPOp helper call check_ieee_exceptions.  For the moment this
requires that all FPop helpers write to the TCG global cpu_fsr,
so remove TCG_CALL_NO_WG from the DEF_HELPER_FLAGS_*.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/sparc/helper.h     | 119 +++++++++++----------
 target/sparc/fop_helper.c | 215 ++++++++++++++++++++++++++++----------
 target/sparc/translate.c  |  14 ---
 3 files changed, 219 insertions(+), 129 deletions(-)

diff --git a/target/sparc/helper.h b/target/sparc/helper.h
index cc8db50d75..7c688edd62 100644
--- a/target/sparc/helper.h
+++ b/target/sparc/helper.h
@@ -35,81 +35,80 @@ DEF_HELPER_3(tsubcctv, tl, env, tl, tl)
 DEF_HELPER_FLAGS_4(ld_asi, TCG_CALL_NO_WG, i64, env, tl, int, i32)
 DEF_HELPER_FLAGS_5(st_asi, TCG_CALL_NO_WG, void, env, tl, i64, int, i32)
 #endif
-DEF_HELPER_FLAGS_1(check_ieee_exceptions, TCG_CALL_NO_WG, tl, env)
 DEF_HELPER_FLAGS_1(get_fsr, TCG_CALL_NO_WG_SE, tl, env)
 DEF_HELPER_FLAGS_2(set_fsr, TCG_CALL_NO_RWG, void, env, tl)
-DEF_HELPER_FLAGS_2(fsqrts, TCG_CALL_NO_RWG, f32, env, f32)
-DEF_HELPER_FLAGS_2(fsqrtd, TCG_CALL_NO_RWG, f64, env, f64)
-DEF_HELPER_FLAGS_3(fcmps, TCG_CALL_NO_WG, tl, env, f32, f32)
-DEF_HELPER_FLAGS_3(fcmpd, TCG_CALL_NO_WG, tl, env, f64, f64)
-DEF_HELPER_FLAGS_3(fcmpes, TCG_CALL_NO_WG, tl, env, f32, f32)
-DEF_HELPER_FLAGS_3(fcmped, TCG_CALL_NO_WG, tl, env, f64, f64)
-DEF_HELPER_FLAGS_2(fsqrtq, TCG_CALL_NO_RWG, i128, env, i128)
-DEF_HELPER_FLAGS_3(fcmpq, TCG_CALL_NO_WG, tl, env, i128, i128)
-DEF_HELPER_FLAGS_3(fcmpeq, TCG_CALL_NO_WG, tl, env, i128, i128)
+DEF_HELPER_FLAGS_2(fsqrts, 0, f32, env, f32)
+DEF_HELPER_FLAGS_2(fsqrtd, 0, f64, env, f64)
+DEF_HELPER_FLAGS_2(fsqrtq, 0, i128, env, i128)
+DEF_HELPER_FLAGS_3(fcmps, 0, tl, env, f32, f32)
+DEF_HELPER_FLAGS_3(fcmpd, 0, tl, env, f64, f64)
+DEF_HELPER_FLAGS_3(fcmpes, 0, tl, env, f32, f32)
+DEF_HELPER_FLAGS_3(fcmped, 0, tl, env, f64, f64)
+DEF_HELPER_FLAGS_3(fcmpq, 0, tl, env, i128, i128)
+DEF_HELPER_FLAGS_3(fcmpeq, 0, tl, env, i128, i128)
 #ifdef TARGET_SPARC64
-DEF_HELPER_FLAGS_3(fcmps_fcc1, TCG_CALL_NO_WG, tl, env, f32, f32)
-DEF_HELPER_FLAGS_3(fcmps_fcc2, TCG_CALL_NO_WG, tl, env, f32, f32)
-DEF_HELPER_FLAGS_3(fcmps_fcc3, TCG_CALL_NO_WG, tl, env, f32, f32)
-DEF_HELPER_FLAGS_3(fcmpd_fcc1, TCG_CALL_NO_WG, tl, env, f64, f64)
-DEF_HELPER_FLAGS_3(fcmpd_fcc2, TCG_CALL_NO_WG, tl, env, f64, f64)
-DEF_HELPER_FLAGS_3(fcmpd_fcc3, TCG_CALL_NO_WG, tl, env, f64, f64)
-DEF_HELPER_FLAGS_3(fcmpes_fcc1, TCG_CALL_NO_WG, tl, env, f32, f32)
-DEF_HELPER_FLAGS_3(fcmpes_fcc2, TCG_CALL_NO_WG, tl, env, f32, f32)
-DEF_HELPER_FLAGS_3(fcmpes_fcc3, TCG_CALL_NO_WG, tl, env, f32, f32)
-DEF_HELPER_FLAGS_3(fcmped_fcc1, TCG_CALL_NO_WG, tl, env, f64, f64)
-DEF_HELPER_FLAGS_3(fcmped_fcc2, TCG_CALL_NO_WG, tl, env, f64, f64)
-DEF_HELPER_FLAGS_3(fcmped_fcc3, TCG_CALL_NO_WG, tl, env, f64, f64)
-DEF_HELPER_FLAGS_3(fcmpq_fcc1, TCG_CALL_NO_WG, tl, env, i128, i128)
-DEF_HELPER_FLAGS_3(fcmpq_fcc2, TCG_CALL_NO_WG, tl, env, i128, i128)
-DEF_HELPER_FLAGS_3(fcmpq_fcc3, TCG_CALL_NO_WG, tl, env, i128, i128)
-DEF_HELPER_FLAGS_3(fcmpeq_fcc1, TCG_CALL_NO_WG, tl, env, i128, i128)
-DEF_HELPER_FLAGS_3(fcmpeq_fcc2, TCG_CALL_NO_WG, tl, env, i128, i128)
-DEF_HELPER_FLAGS_3(fcmpeq_fcc3, TCG_CALL_NO_WG, tl, env, i128, i128)
+DEF_HELPER_FLAGS_3(fcmps_fcc1, 0, tl, env, f32, f32)
+DEF_HELPER_FLAGS_3(fcmps_fcc2, 0, tl, env, f32, f32)
+DEF_HELPER_FLAGS_3(fcmps_fcc3, 0, tl, env, f32, f32)
+DEF_HELPER_FLAGS_3(fcmpd_fcc1, 0, tl, env, f64, f64)
+DEF_HELPER_FLAGS_3(fcmpd_fcc2, 0, tl, env, f64, f64)
+DEF_HELPER_FLAGS_3(fcmpd_fcc3, 0, tl, env, f64, f64)
+DEF_HELPER_FLAGS_3(fcmpes_fcc1, 0, tl, env, f32, f32)
+DEF_HELPER_FLAGS_3(fcmpes_fcc2, 0, tl, env, f32, f32)
+DEF_HELPER_FLAGS_3(fcmpes_fcc3, 0, tl, env, f32, f32)
+DEF_HELPER_FLAGS_3(fcmped_fcc1, 0, tl, env, f64, f64)
+DEF_HELPER_FLAGS_3(fcmped_fcc2, 0, tl, env, f64, f64)
+DEF_HELPER_FLAGS_3(fcmped_fcc3, 0, tl, env, f64, f64)
+DEF_HELPER_FLAGS_3(fcmpq_fcc1, 0, tl, env, i128, i128)
+DEF_HELPER_FLAGS_3(fcmpq_fcc2, 0, tl, env, i128, i128)
+DEF_HELPER_FLAGS_3(fcmpq_fcc3, 0, tl, env, i128, i128)
+DEF_HELPER_FLAGS_3(fcmpeq_fcc1, 0, tl, env, i128, i128)
+DEF_HELPER_FLAGS_3(fcmpeq_fcc2, 0, tl, env, i128, i128)
+DEF_HELPER_FLAGS_3(fcmpeq_fcc3, 0, tl, env, i128, i128)
 #endif
 DEF_HELPER_2(raise_exception, noreturn, env, int)
 
-DEF_HELPER_FLAGS_3(faddd, TCG_CALL_NO_RWG, f64, env, f64, f64)
-DEF_HELPER_FLAGS_3(fsubd, TCG_CALL_NO_RWG, f64, env, f64, f64)
-DEF_HELPER_FLAGS_3(fmuld, TCG_CALL_NO_RWG, f64, env, f64, f64)
-DEF_HELPER_FLAGS_3(fdivd, TCG_CALL_NO_RWG, f64, env, f64, f64)
+DEF_HELPER_FLAGS_3(faddd, 0, f64, env, f64, f64)
+DEF_HELPER_FLAGS_3(fsubd, 0, f64, env, f64, f64)
+DEF_HELPER_FLAGS_3(fmuld, 0, f64, env, f64, f64)
+DEF_HELPER_FLAGS_3(fdivd, 0, f64, env, f64, f64)
 
-DEF_HELPER_FLAGS_3(faddq, TCG_CALL_NO_RWG, i128, env, i128, i128)
-DEF_HELPER_FLAGS_3(fsubq, TCG_CALL_NO_RWG, i128, env, i128, i128)
-DEF_HELPER_FLAGS_3(fmulq, TCG_CALL_NO_RWG, i128, env, i128, i128)
-DEF_HELPER_FLAGS_3(fdivq, TCG_CALL_NO_RWG, i128, env, i128, i128)
+DEF_HELPER_FLAGS_3(faddq, 0, i128, env, i128, i128)
+DEF_HELPER_FLAGS_3(fsubq, 0, i128, env, i128, i128)
+DEF_HELPER_FLAGS_3(fmulq, 0, i128, env, i128, i128)
+DEF_HELPER_FLAGS_3(fdivq, 0, i128, env, i128, i128)
 
-DEF_HELPER_FLAGS_3(fadds, TCG_CALL_NO_RWG, f32, env, f32, f32)
-DEF_HELPER_FLAGS_3(fsubs, TCG_CALL_NO_RWG, f32, env, f32, f32)
-DEF_HELPER_FLAGS_3(fmuls, TCG_CALL_NO_RWG, f32, env, f32, f32)
-DEF_HELPER_FLAGS_3(fdivs, TCG_CALL_NO_RWG, f32, env, f32, f32)
+DEF_HELPER_FLAGS_3(fadds, 0, f32, env, f32, f32)
+DEF_HELPER_FLAGS_3(fsubs, 0, f32, env, f32, f32)
+DEF_HELPER_FLAGS_3(fmuls, 0, f32, env, f32, f32)
+DEF_HELPER_FLAGS_3(fdivs, 0, f32, env, f32, f32)
 
-DEF_HELPER_FLAGS_3(fsmuld, TCG_CALL_NO_RWG, f64, env, f32, f32)
-DEF_HELPER_FLAGS_3(fdmulq, TCG_CALL_NO_RWG, i128, env, f64, f64)
+DEF_HELPER_FLAGS_3(fsmuld, 0, f64, env, f32, f32)
+DEF_HELPER_FLAGS_3(fdmulq, 0, i128, env, f64, f64)
 
-DEF_HELPER_FLAGS_2(fitod, TCG_CALL_NO_RWG_SE, f64, env, s32)
-DEF_HELPER_FLAGS_2(fitoq, TCG_CALL_NO_RWG, i128, env, s32)
+DEF_HELPER_FLAGS_2(fitod, 0, f64, env, s32)
+DEF_HELPER_FLAGS_2(fitoq, 0, i128, env, s32)
 
-DEF_HELPER_FLAGS_2(fitos, TCG_CALL_NO_RWG, f32, env, s32)
+DEF_HELPER_FLAGS_2(fitos, 0, f32, env, s32)
 
 #ifdef TARGET_SPARC64
-DEF_HELPER_FLAGS_2(fxtos, TCG_CALL_NO_RWG, f32, env, s64)
-DEF_HELPER_FLAGS_2(fxtod, TCG_CALL_NO_RWG, f64, env, s64)
-DEF_HELPER_FLAGS_2(fxtoq, TCG_CALL_NO_RWG, i128, env, s64)
+DEF_HELPER_FLAGS_2(fxtos, 0, f32, env, s64)
+DEF_HELPER_FLAGS_2(fxtod, 0, f64, env, s64)
+DEF_HELPER_FLAGS_2(fxtoq, 0, i128, env, s64)
 #endif
-DEF_HELPER_FLAGS_2(fdtos, TCG_CALL_NO_RWG, f32, env, f64)
-DEF_HELPER_FLAGS_2(fstod, TCG_CALL_NO_RWG, f64, env, f32)
-DEF_HELPER_FLAGS_2(fqtos, TCG_CALL_NO_RWG, f32, env, i128)
-DEF_HELPER_FLAGS_2(fstoq, TCG_CALL_NO_RWG, i128, env, f32)
-DEF_HELPER_FLAGS_2(fqtod, TCG_CALL_NO_RWG, f64, env, i128)
-DEF_HELPER_FLAGS_2(fdtoq, TCG_CALL_NO_RWG, i128, env, f64)
-DEF_HELPER_FLAGS_2(fstoi, TCG_CALL_NO_RWG, s32, env, f32)
-DEF_HELPER_FLAGS_2(fdtoi, TCG_CALL_NO_RWG, s32, env, f64)
-DEF_HELPER_FLAGS_2(fqtoi, TCG_CALL_NO_RWG, s32, env, i128)
+DEF_HELPER_FLAGS_2(fdtos, 0, f32, env, f64)
+DEF_HELPER_FLAGS_2(fstod, 0, f64, env, f32)
+DEF_HELPER_FLAGS_2(fqtos, 0, f32, env, i128)
+DEF_HELPER_FLAGS_2(fstoq, 0, i128, env, f32)
+DEF_HELPER_FLAGS_2(fqtod, 0, f64, env, i128)
+DEF_HELPER_FLAGS_2(fdtoq, 0, i128, env, f64)
+DEF_HELPER_FLAGS_2(fstoi, 0, s32, env, f32)
+DEF_HELPER_FLAGS_2(fdtoi, 0, s32, env, f64)
+DEF_HELPER_FLAGS_2(fqtoi, 0, s32, env, i128)
 #ifdef TARGET_SPARC64
-DEF_HELPER_FLAGS_2(fstox, TCG_CALL_NO_RWG, s64, env, f32)
-DEF_HELPER_FLAGS_2(fdtox, TCG_CALL_NO_RWG, s64, env, f64)
-DEF_HELPER_FLAGS_2(fqtox, TCG_CALL_NO_RWG, s64, env, i128)
+DEF_HELPER_FLAGS_2(fstox, 0, s64, env, f32)
+DEF_HELPER_FLAGS_2(fdtox, 0, s64, env, f64)
+DEF_HELPER_FLAGS_2(fqtox, 0, s64, env, i128)
 
 DEF_HELPER_FLAGS_2(fpmerge, TCG_CALL_NO_RWG_SE, i64, i64, i64)
 DEF_HELPER_FLAGS_2(fmul8x16, TCG_CALL_NO_RWG_SE, i64, i64, i64)
diff --git a/target/sparc/fop_helper.c b/target/sparc/fop_helper.c
index 64f20e78f1..755117ea08 100644
--- a/target/sparc/fop_helper.c
+++ b/target/sparc/fop_helper.c
@@ -45,7 +45,7 @@ static inline Int128 f128_ret(float128 f)
     return u.i;
 }
 
-static target_ulong do_check_ieee_exceptions(CPUSPARCState *env, uintptr_t ra)
+static void check_ieee_exceptions(CPUSPARCState *env, uintptr_t ra)
 {
     target_ulong status = get_float_exception_flags(&env->fp_status);
     target_ulong fsr = env->fsr;
@@ -89,162 +89,265 @@ static target_ulong do_check_ieee_exceptions(CPUSPARCState *env, uintptr_t ra)
         }
     }
 
-    return fsr;
+    env->fsr = fsr;
 }
 
-target_ulong helper_check_ieee_exceptions(CPUSPARCState *env)
+float32 helper_fadds(CPUSPARCState *env, float32 src1, float32 src2)
 {
-    return do_check_ieee_exceptions(env, GETPC());
+    float32 ret = float32_add(src1, src2, &env->fp_status);
+    check_ieee_exceptions(env, GETPC());
+    return ret;
 }
 
-#define F_BINOP(name)                                                \
-    float32 helper_f ## name ## s (CPUSPARCState *env, float32 src1, \
-                                   float32 src2)                     \
-    {                                                                \
-        return float32_ ## name (src1, src2, &env->fp_status);       \
-    }                                                                \
-    float64 helper_f ## name ## d (CPUSPARCState * env, float64 src1,\
-                                   float64 src2)                     \
-    {                                                                \
-        return float64_ ## name (src1, src2, &env->fp_status);       \
-    }                                                                \
-    Int128 helper_f ## name ## q(CPUSPARCState * env, Int128 src1,   \
-                                 Int128 src2)                        \
-    {                                                                \
-        return f128_ret(float128_ ## name (f128_in(src1), f128_in(src2), \
-                                           &env->fp_status));        \
-    }
+float32 helper_fsubs(CPUSPARCState *env, float32 src1, float32 src2)
+{
+    float32 ret = float32_sub(src1, src2, &env->fp_status);
+    check_ieee_exceptions(env, GETPC());
+    return ret;
+}
 
-F_BINOP(add);
-F_BINOP(sub);
-F_BINOP(mul);
-F_BINOP(div);
-#undef F_BINOP
+float32 helper_fmuls(CPUSPARCState *env, float32 src1, float32 src2)
+{
+    float32 ret = float32_mul(src1, src2, &env->fp_status);
+    check_ieee_exceptions(env, GETPC());
+    return ret;
+}
+
+float32 helper_fdivs(CPUSPARCState *env, float32 src1, float32 src2)
+{
+    float32 ret = float32_div(src1, src2, &env->fp_status);
+    check_ieee_exceptions(env, GETPC());
+    return ret;
+}
+
+float64 helper_faddd(CPUSPARCState *env, float64 src1, float64 src2)
+{
+    float64 ret = float64_add(src1, src2, &env->fp_status);
+    check_ieee_exceptions(env, GETPC());
+    return ret;
+}
+
+float64 helper_fsubd(CPUSPARCState *env, float64 src1, float64 src2)
+{
+    float64 ret = float64_sub(src1, src2, &env->fp_status);
+    check_ieee_exceptions(env, GETPC());
+    return ret;
+}
+
+float64 helper_fmuld(CPUSPARCState *env, float64 src1, float64 src2)
+{
+    float64 ret = float64_mul(src1, src2, &env->fp_status);
+    check_ieee_exceptions(env, GETPC());
+    return ret;
+}
+
+float64 helper_fdivd(CPUSPARCState *env, float64 src1, float64 src2)
+{
+    float64 ret = float64_div(src1, src2, &env->fp_status);
+    check_ieee_exceptions(env, GETPC());
+    return ret;
+}
+
+Int128 helper_faddq(CPUSPARCState *env, Int128 src1, Int128 src2)
+{
+    float128 ret = float128_add(f128_in(src1), f128_in(src2), &env->fp_status);
+    check_ieee_exceptions(env, GETPC());
+    return f128_ret(ret);
+}
+
+Int128 helper_fsubq(CPUSPARCState *env, Int128 src1, Int128 src2)
+{
+    float128 ret = float128_sub(f128_in(src1), f128_in(src2), &env->fp_status);
+    check_ieee_exceptions(env, GETPC());
+    return f128_ret(ret);
+}
+
+Int128 helper_fmulq(CPUSPARCState *env, Int128 src1, Int128 src2)
+{
+    float128 ret = float128_mul(f128_in(src1), f128_in(src2), &env->fp_status);
+    check_ieee_exceptions(env, GETPC());
+    return f128_ret(ret);
+}
+
+Int128 helper_fdivq(CPUSPARCState *env, Int128 src1, Int128 src2)
+{
+    float128 ret = float128_div(f128_in(src1), f128_in(src2), &env->fp_status);
+    check_ieee_exceptions(env, GETPC());
+    return f128_ret(ret);
+}
 
 float64 helper_fsmuld(CPUSPARCState *env, float32 src1, float32 src2)
 {
-    return float64_mul(float32_to_float64(src1, &env->fp_status),
-                       float32_to_float64(src2, &env->fp_status),
-                       &env->fp_status);
+    float64 ret = float64_mul(float32_to_float64(src1, &env->fp_status),
+                              float32_to_float64(src2, &env->fp_status),
+                              &env->fp_status);
+    check_ieee_exceptions(env, GETPC());
+    return ret;
 }
 
 Int128 helper_fdmulq(CPUSPARCState *env, float64 src1, float64 src2)
 {
-    return f128_ret(float128_mul(float64_to_float128(src1, &env->fp_status),
-                                 float64_to_float128(src2, &env->fp_status),
-                                 &env->fp_status));
+    float128 ret = float128_mul(float64_to_float128(src1, &env->fp_status),
+                                float64_to_float128(src2, &env->fp_status),
+                                &env->fp_status);
+    check_ieee_exceptions(env, GETPC());
+    return f128_ret(ret);
 }
 
 /* Integer to float conversion.  */
 float32 helper_fitos(CPUSPARCState *env, int32_t src)
 {
-    return int32_to_float32(src, &env->fp_status);
+    float32 ret = int32_to_float32(src, &env->fp_status);
+    check_ieee_exceptions(env, GETPC());
+    return ret;
 }
 
 float64 helper_fitod(CPUSPARCState *env, int32_t src)
 {
-    return int32_to_float64(src, &env->fp_status);
+    float64 ret = int32_to_float64(src, &env->fp_status);
+    check_ieee_exceptions(env, GETPC());
+    return ret;
 }
 
 Int128 helper_fitoq(CPUSPARCState *env, int32_t src)
 {
-    return f128_ret(int32_to_float128(src, &env->fp_status));
+    float128 ret = int32_to_float128(src, &env->fp_status);
+    check_ieee_exceptions(env, GETPC());
+    return f128_ret(ret);
 }
 
 #ifdef TARGET_SPARC64
 float32 helper_fxtos(CPUSPARCState *env, int64_t src)
 {
-    return int64_to_float32(src, &env->fp_status);
+    float32 ret = int64_to_float32(src, &env->fp_status);
+    check_ieee_exceptions(env, GETPC());
+    return ret;
 }
 
 float64 helper_fxtod(CPUSPARCState *env, int64_t src)
 {
-    return int64_to_float64(src, &env->fp_status);
+    float64 ret = int64_to_float64(src, &env->fp_status);
+    check_ieee_exceptions(env, GETPC());
+    return ret;
 }
 
 Int128 helper_fxtoq(CPUSPARCState *env, int64_t src)
 {
-    return f128_ret(int64_to_float128(src, &env->fp_status));
+    float128 ret = int64_to_float128(src, &env->fp_status);
+    check_ieee_exceptions(env, GETPC());
+    return f128_ret(ret);
 }
 #endif
 
 /* floating point conversion */
 float32 helper_fdtos(CPUSPARCState *env, float64 src)
 {
-    return float64_to_float32(src, &env->fp_status);
+    float32 ret = float64_to_float32(src, &env->fp_status);
+    check_ieee_exceptions(env, GETPC());
+    return ret;
 }
 
 float64 helper_fstod(CPUSPARCState *env, float32 src)
 {
-    return float32_to_float64(src, &env->fp_status);
+    float64 ret = float32_to_float64(src, &env->fp_status);
+    check_ieee_exceptions(env, GETPC());
+    return ret;
 }
 
 float32 helper_fqtos(CPUSPARCState *env, Int128 src)
 {
-    return float128_to_float32(f128_in(src), &env->fp_status);
+    float32 ret = float128_to_float32(f128_in(src), &env->fp_status);
+    check_ieee_exceptions(env, GETPC());
+    return ret;
 }
 
 Int128 helper_fstoq(CPUSPARCState *env, float32 src)
 {
-    return f128_ret(float32_to_float128(src, &env->fp_status));
+    float128 ret = float32_to_float128(src, &env->fp_status);
+    check_ieee_exceptions(env, GETPC());
+    return f128_ret(ret);
 }
 
 float64 helper_fqtod(CPUSPARCState *env, Int128 src)
 {
-    return float128_to_float64(f128_in(src), &env->fp_status);
+    float64 ret = float128_to_float64(f128_in(src), &env->fp_status);
+    check_ieee_exceptions(env, GETPC());
+    return ret;
 }
 
 Int128 helper_fdtoq(CPUSPARCState *env, float64 src)
 {
-    return f128_ret(float64_to_float128(src, &env->fp_status));
+    float128 ret = float64_to_float128(src, &env->fp_status);
+    check_ieee_exceptions(env, GETPC());
+    return f128_ret(ret);
 }
 
 /* Float to integer conversion.  */
 int32_t helper_fstoi(CPUSPARCState *env, float32 src)
 {
-    return float32_to_int32_round_to_zero(src, &env->fp_status);
+    int32_t ret = float32_to_int32_round_to_zero(src, &env->fp_status);
+    check_ieee_exceptions(env, GETPC());
+    return ret;
 }
 
 int32_t helper_fdtoi(CPUSPARCState *env, float64 src)
 {
-    return float64_to_int32_round_to_zero(src, &env->fp_status);
+    int32_t ret = float64_to_int32_round_to_zero(src, &env->fp_status);
+    check_ieee_exceptions(env, GETPC());
+    return ret;
 }
 
 int32_t helper_fqtoi(CPUSPARCState *env, Int128 src)
 {
-    return float128_to_int32_round_to_zero(f128_in(src), &env->fp_status);
+    int32_t ret = float128_to_int32_round_to_zero(f128_in(src),
+                                                  &env->fp_status);
+    check_ieee_exceptions(env, GETPC());
+    return ret;
 }
 
 #ifdef TARGET_SPARC64
 int64_t helper_fstox(CPUSPARCState *env, float32 src)
 {
-    return float32_to_int64_round_to_zero(src, &env->fp_status);
+    int64_t ret = float32_to_int64_round_to_zero(src, &env->fp_status);
+    check_ieee_exceptions(env, GETPC());
+    return ret;
 }
 
 int64_t helper_fdtox(CPUSPARCState *env, float64 src)
 {
-    return float64_to_int64_round_to_zero(src, &env->fp_status);
+    int64_t ret = float64_to_int64_round_to_zero(src, &env->fp_status);
+    check_ieee_exceptions(env, GETPC());
+    return ret;
 }
 
 int64_t helper_fqtox(CPUSPARCState *env, Int128 src)
 {
-    return float128_to_int64_round_to_zero(f128_in(src), &env->fp_status);
+    int64_t ret = float128_to_int64_round_to_zero(f128_in(src),
+                                                  &env->fp_status);
+    check_ieee_exceptions(env, GETPC());
+    return ret;
 }
 #endif
 
 float32 helper_fsqrts(CPUSPARCState *env, float32 src)
 {
-    return float32_sqrt(src, &env->fp_status);
+    float32 ret = float32_sqrt(src, &env->fp_status);
+    check_ieee_exceptions(env, GETPC());
+    return ret;
 }
 
 float64 helper_fsqrtd(CPUSPARCState *env, float64 src)
 {
-    return float64_sqrt(src, &env->fp_status);
+    float64 ret = float64_sqrt(src, &env->fp_status);
+    check_ieee_exceptions(env, GETPC());
+    return ret;
 }
 
 Int128 helper_fsqrtq(CPUSPARCState *env, Int128 src)
 {
-    return f128_ret(float128_sqrt(f128_in(src), &env->fp_status));
+    float128 ret = float128_sqrt(f128_in(src), &env->fp_status);
+    check_ieee_exceptions(env, GETPC());
+    return f128_ret(ret);
 }
 
 #define GEN_FCMP(name, size, FS, E)                                     \
@@ -261,7 +364,8 @@ Int128 helper_fsqrtq(CPUSPARCState *env, Int128 src)
             ret = glue(size, _compare_quiet)(reg1, reg2,                \
                                              &env->fp_status);          \
         }                                                               \
-        fsr = do_check_ieee_exceptions(env, GETPC());                   \
+        check_ieee_exceptions(env, GETPC());                            \
+        fsr = env->fsr;                                                 \
         switch (ret) {                                                  \
         case float_relation_unordered:                                  \
             fsr |= (FSR_FCC1 | FSR_FCC0) << FS;                         \
@@ -292,7 +396,8 @@ Int128 helper_fsqrtq(CPUSPARCState *env, Int128 src)
             ret = glue(size, _compare_quiet)(src1, src2,                \
                                              &env->fp_status);          \
         }                                                               \
-        fsr = do_check_ieee_exceptions(env, GETPC());                   \
+        check_ieee_exceptions(env, GETPC());                            \
+        fsr = env->fsr;                                                 \
         switch (ret) {                                                  \
         case float_relation_unordered:                                  \
             fsr |= (FSR_FCC1 | FSR_FCC0) << FS;                         \
diff --git a/target/sparc/translate.c b/target/sparc/translate.c
index 11d025f4ea..8787cb3bfe 100644
--- a/target/sparc/translate.c
+++ b/target/sparc/translate.c
@@ -4526,7 +4526,6 @@ static bool do_env_ff(DisasContext *dc, arg_r_r *a,
 
     tmp = gen_load_fpr_F(dc, a->rs);
     func(tmp, tcg_env, tmp);
-    gen_helper_check_ieee_exceptions(cpu_fsr, tcg_env);
     gen_store_fpr_F(dc, a->rd, tmp);
     return advance_pc(dc);
 }
@@ -4548,7 +4547,6 @@ static bool do_env_fd(DisasContext *dc, arg_r_r *a,
     dst = tcg_temp_new_i32();
     src = gen_load_fpr_D(dc, a->rs);
     func(dst, tcg_env, src);
-    gen_helper_check_ieee_exceptions(cpu_fsr, tcg_env);
     gen_store_fpr_F(dc, a->rd, dst);
     return advance_pc(dc);
 }
@@ -4591,7 +4589,6 @@ static bool do_env_dd(DisasContext *dc, arg_r_r *a,
     dst = gen_dest_fpr_D(dc, a->rd);
     src = gen_load_fpr_D(dc, a->rs);
     func(dst, tcg_env, src);
-    gen_helper_check_ieee_exceptions(cpu_fsr, tcg_env);
     gen_store_fpr_D(dc, a->rd, dst);
     return advance_pc(dc);
 }
@@ -4613,7 +4610,6 @@ static bool do_env_df(DisasContext *dc, arg_r_r *a,
     dst = gen_dest_fpr_D(dc, a->rd);
     src = gen_load_fpr_F(dc, a->rs);
     func(dst, tcg_env, src);
-    gen_helper_check_ieee_exceptions(cpu_fsr, tcg_env);
     gen_store_fpr_D(dc, a->rd, dst);
     return advance_pc(dc);
 }
@@ -4659,7 +4655,6 @@ static bool do_env_qq(DisasContext *dc, arg_r_r *a,
 
     t = gen_load_fpr_Q(dc, a->rs);
     func(t, tcg_env, t);
-    gen_helper_check_ieee_exceptions(cpu_fsr, tcg_env);
     gen_store_fpr_Q(dc, a->rd, t);
     return advance_pc(dc);
 }
@@ -4682,7 +4677,6 @@ static bool do_env_fq(DisasContext *dc, arg_r_r *a,
     src = gen_load_fpr_Q(dc, a->rs);
     dst = tcg_temp_new_i32();
     func(dst, tcg_env, src);
-    gen_helper_check_ieee_exceptions(cpu_fsr, tcg_env);
     gen_store_fpr_F(dc, a->rd, dst);
     return advance_pc(dc);
 }
@@ -4706,7 +4700,6 @@ static bool do_env_dq(DisasContext *dc, arg_r_r *a,
     src = gen_load_fpr_Q(dc, a->rs);
     dst = gen_dest_fpr_D(dc, a->rd);
     func(dst, tcg_env, src);
-    gen_helper_check_ieee_exceptions(cpu_fsr, tcg_env);
     gen_store_fpr_D(dc, a->rd, dst);
     return advance_pc(dc);
 }
@@ -4727,7 +4720,6 @@ static bool do_env_qf(DisasContext *dc, arg_r_r *a,
         return true;
     }
 
-    gen_op_clear_ieee_excp_and_FTT();
     src = gen_load_fpr_F(dc, a->rs);
     dst = tcg_temp_new_i128();
     func(dst, tcg_env, src);
@@ -4751,7 +4743,6 @@ static bool do_env_qd(DisasContext *dc, arg_r_r *a,
         return true;
     }
 
-    gen_op_clear_ieee_excp_and_FTT();
     src = gen_load_fpr_D(dc, a->rs);
     dst = tcg_temp_new_i128();
     func(dst, tcg_env, src);
@@ -4803,7 +4794,6 @@ static bool do_env_fff(DisasContext *dc, arg_r_r_r *a,
     src1 = gen_load_fpr_F(dc, a->rs1);
     src2 = gen_load_fpr_F(dc, a->rs2);
     func(src1, tcg_env, src1, src2);
-    gen_helper_check_ieee_exceptions(cpu_fsr, tcg_env);
     gen_store_fpr_F(dc, a->rd, src1);
     return advance_pc(dc);
 }
@@ -4898,7 +4888,6 @@ static bool do_env_ddd(DisasContext *dc, arg_r_r_r *a,
     src1 = gen_load_fpr_D(dc, a->rs1);
     src2 = gen_load_fpr_D(dc, a->rs2);
     func(dst, tcg_env, src1, src2);
-    gen_helper_check_ieee_exceptions(cpu_fsr, tcg_env);
     gen_store_fpr_D(dc, a->rd, dst);
     return advance_pc(dc);
 }
@@ -4924,7 +4913,6 @@ static bool trans_FsMULd(DisasContext *dc, arg_r_r_r *a)
     src1 = gen_load_fpr_F(dc, a->rs1);
     src2 = gen_load_fpr_F(dc, a->rs2);
     gen_helper_fsmuld(dst, tcg_env, src1, src2);
-    gen_helper_check_ieee_exceptions(cpu_fsr, tcg_env);
     gen_store_fpr_D(dc, a->rd, dst);
     return advance_pc(dc);
 }
@@ -4964,7 +4952,6 @@ static bool do_env_qqq(DisasContext *dc, arg_r_r_r *a,
     src1 = gen_load_fpr_Q(dc, a->rs1);
     src2 = gen_load_fpr_Q(dc, a->rs2);
     func(src1, tcg_env, src1, src2);
-    gen_helper_check_ieee_exceptions(cpu_fsr, tcg_env);
     gen_store_fpr_Q(dc, a->rd, src1);
     return advance_pc(dc);
 }
@@ -4990,7 +4977,6 @@ static bool trans_FdMULq(DisasContext *dc, arg_r_r_r *a)
     src2 = gen_load_fpr_D(dc, a->rs2);
     dst = tcg_temp_new_i128();
     gen_helper_fdmulq(dst, tcg_env, src1, src2);
-    gen_helper_check_ieee_exceptions(cpu_fsr, tcg_env);
     gen_store_fpr_Q(dc, a->rd, dst);
     return advance_pc(dc);
 }
-- 
2.34.1

[PATCH 19/22] target/sparc: Split cexc and ftt from env->fsr

[Richard Henderson]

These two fields are adjusted by all FPop insns.
Having them separate makes it easier to set without masking.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/sparc/cpu.h        |  7 +++++-
 target/sparc/helper.h     |  2 +-
 target/sparc/fop_helper.c | 46 ++++++++++++++++++---------------------
 target/sparc/translate.c  | 31 ++++++++++++++++----------
 4 files changed, 48 insertions(+), 38 deletions(-)

diff --git a/target/sparc/cpu.h b/target/sparc/cpu.h
index 8ff222595e..90a7dfb004 100644
--- a/target/sparc/cpu.h
+++ b/target/sparc/cpu.h
@@ -176,6 +176,7 @@ enum {
 #define FSR_DZM   (1ULL << 24)
 #define FSR_NXM   (1ULL << 23)
 #define FSR_TEM_MASK (FSR_NVM | FSR_OFM | FSR_UFM | FSR_DZM | FSR_NXM)
+#define FSR_TEM_SHIFT  23
 
 #define FSR_NVA   (1ULL << 9)
 #define FSR_OFA   (1ULL << 8)
@@ -183,6 +184,7 @@ enum {
 #define FSR_DZA   (1ULL << 6)
 #define FSR_NXA   (1ULL << 5)
 #define FSR_AEXC_MASK (FSR_NVA | FSR_OFA | FSR_UFA | FSR_DZA | FSR_NXA)
+#define FSR_AEXC_SHIFT 5
 
 #define FSR_NVC   (1ULL << 4)
 #define FSR_OFC   (1ULL << 3)
@@ -464,7 +466,10 @@ struct CPUArchState {
     target_ulong cond; /* conditional branch result (XXX: save it in a
                           temporary register when possible) */
 
-    target_ulong fsr;      /* FPU state register */
+    /* FPU State Register, in parts */
+    target_ulong fsr;       /* rm, tem, aexc, fcc* */
+    uint32_t fsr_cexc_ftt;  /* cexc, ftt */
+
     CPU_DoubleU fpr[TARGET_DPREGS];  /* floating point registers */
     uint32_t cwp;      /* index of current register window (extracted
                           from PSR) */
diff --git a/target/sparc/helper.h b/target/sparc/helper.h
index 7c688edd62..7466164468 100644
--- a/target/sparc/helper.h
+++ b/target/sparc/helper.h
@@ -36,7 +36,7 @@ DEF_HELPER_FLAGS_4(ld_asi, TCG_CALL_NO_WG, i64, env, tl, int, i32)
 DEF_HELPER_FLAGS_5(st_asi, TCG_CALL_NO_WG, void, env, tl, i64, int, i32)
 #endif
 DEF_HELPER_FLAGS_1(get_fsr, TCG_CALL_NO_WG_SE, tl, env)
-DEF_HELPER_FLAGS_2(set_fsr, TCG_CALL_NO_RWG, void, env, tl)
+DEF_HELPER_FLAGS_2(set_fsr_noftt, 0, void, env, tl)
 DEF_HELPER_FLAGS_2(fsqrts, 0, f32, env, f32)
 DEF_HELPER_FLAGS_2(fsqrtd, 0, f64, env, f64)
 DEF_HELPER_FLAGS_2(fsqrtq, 0, i128, env, i128)
diff --git a/target/sparc/fop_helper.c b/target/sparc/fop_helper.c
index 755117ea08..ac30f88810 100644
--- a/target/sparc/fop_helper.c
+++ b/target/sparc/fop_helper.c
@@ -48,9 +48,7 @@ static inline Int128 f128_ret(float128 f)
 static void check_ieee_exceptions(CPUSPARCState *env, uintptr_t ra)
 {
     target_ulong status = get_float_exception_flags(&env->fp_status);
-    target_ulong fsr = env->fsr;
-
-    fsr &= FSR_FTT_CEXC_NMASK;
+    uint32_t cexc = 0;
 
     if (unlikely(status)) {
         /* Keep exception flags clear for next time.  */
@@ -58,38 +56,33 @@ static void check_ieee_exceptions(CPUSPARCState *env, uintptr_t ra)
 
         /* Copy IEEE 754 flags into FSR */
         if (status & float_flag_invalid) {
-            fsr |= FSR_NVC;
+            cexc |= FSR_NVC;
         }
         if (status & float_flag_overflow) {
-            fsr |= FSR_OFC;
+            cexc |= FSR_OFC;
         }
         if (status & float_flag_underflow) {
-            fsr |= FSR_UFC;
+            cexc |= FSR_UFC;
         }
         if (status & float_flag_divbyzero) {
-            fsr |= FSR_DZC;
+            cexc |= FSR_DZC;
         }
         if (status & float_flag_inexact) {
-            fsr |= FSR_NXC;
+            cexc |= FSR_NXC;
         }
 
-        if ((fsr & FSR_CEXC_MASK) & ((fsr & FSR_TEM_MASK) >> 23)) {
-            CPUState *cs = env_cpu(env);
-
-            /* Unmasked exception, generate a trap.  Note that while
-               the helper is marked as NO_WG, we can get away with
-               writing to cpu state along the exception path, since
-               TCG generated code will never see the write.  */
-            env->fsr = fsr | FSR_FTT_IEEE_EXCP;
-            cs->exception_index = TT_FP_EXCP;
-            cpu_loop_exit_restore(cs, ra);
-        } else {
-            /* Accumulate exceptions */
-            fsr |= (fsr & FSR_CEXC_MASK) << 5;
+        if (cexc & (env->fsr >> FSR_TEM_SHIFT)) {
+            /* Unmasked exception, generate an IEEE trap. */
+            env->fsr_cexc_ftt = cexc | FSR_FTT_IEEE_EXCP;
+            cpu_raise_exception_ra(env, TT_FP_EXCP, ra);
         }
+
+        /* Accumulate exceptions */
+        env->fsr |= cexc << FSR_AEXC_SHIFT;
     }
 
-    env->fsr = fsr;
+    /* No trap, so FTT is cleared. */
+    env->fsr_cexc_ftt = cexc;
 }
 
 float32 helper_fadds(CPUSPARCState *env, float32 src1, float32 src2)
@@ -456,7 +449,7 @@ GEN_FCMP(fcmpeq_fcc3, float128, 26, 1);
 
 target_ulong cpu_get_fsr(CPUSPARCState *env)
 {
-    target_ulong fsr = env->fsr;
+    target_ulong fsr = env->fsr | env->fsr_cexc_ftt;
 
     /* VER is kept completely separate until re-assembly. */
     fsr |= env->def.fpu_version;
@@ -473,7 +466,7 @@ static void set_fsr_nonsplit(CPUSPARCState *env, target_ulong fsr)
 {
     int rnd_mode;
 
-    env->fsr = fsr & ~FSR_VER_MASK;
+    env->fsr = fsr & ~(FSR_VER_MASK | FSR_CEXC_MASK | FSR_FTT_MASK);
 
     switch (fsr & FSR_RD_MASK) {
     case FSR_RD_NEAREST:
@@ -495,10 +488,13 @@ static void set_fsr_nonsplit(CPUSPARCState *env, target_ulong fsr)
 
 void cpu_put_fsr(CPUSPARCState *env, target_ulong fsr)
 {
+    env->fsr_cexc_ftt = fsr & (FSR_CEXC_MASK | FSR_FTT_MASK);
     set_fsr_nonsplit(env, fsr);
 }
 
-void helper_set_fsr(CPUSPARCState *env, target_ulong fsr)
+void helper_set_fsr_noftt(CPUSPARCState *env, target_ulong fsr)
 {
+    env->fsr_cexc_ftt &= FSR_FTT_MASK;
+    env->fsr_cexc_ftt |= fsr & FSR_CEXC_MASK;
     set_fsr_nonsplit(env, fsr);
 }
diff --git a/target/sparc/translate.c b/target/sparc/translate.c
index 8787cb3bfe..d2145dcc0b 100644
--- a/target/sparc/translate.c
+++ b/target/sparc/translate.c
@@ -1199,7 +1199,8 @@ static bool gen_compare_reg(DisasCompare *cmp, int cond, TCGv r_src)
 
 static void gen_op_clear_ieee_excp_and_FTT(void)
 {
-    tcg_gen_andi_tl(cpu_fsr, cpu_fsr, FSR_FTT_CEXC_NMASK);
+    tcg_gen_st_i32(tcg_constant_i32(0), tcg_env,
+                   offsetof(CPUSPARCState, fsr_cexc_ftt));
 }
 
 static void gen_op_fmovs(TCGv_i32 dst, TCGv_i32 src)
@@ -1400,10 +1401,15 @@ static void gen_op_fcmpeq(int fccno, TCGv_i128 r_rs1, TCGv_i128 r_rs2)
 }
 #endif
 
-static void gen_op_fpexception_im(DisasContext *dc, int fsr_flags)
+static void gen_op_fpexception_im(DisasContext *dc, int ftt)
 {
-    tcg_gen_andi_tl(cpu_fsr, cpu_fsr, FSR_FTT_NMASK);
-    tcg_gen_ori_tl(cpu_fsr, cpu_fsr, fsr_flags);
+    /*
+     * CEXC is only set when succesfully completing an FPop,
+     * or when raising FSR_FTT_IEEE_EXCP, i.e. check_ieee_exception.
+     * Thus we can simply store FTT into this field.
+     */
+    tcg_gen_st_i32(tcg_constant_i32(ftt), tcg_env,
+                   offsetof(CPUSPARCState, fsr_cexc_ftt));
     gen_exception(dc, TT_FP_EXCP);
 }
 
@@ -4395,19 +4401,22 @@ static bool trans_STDFQ(DisasContext *dc, arg_STDFQ *a)
 static bool do_ldfsr(DisasContext *dc, arg_r_r_ri *a, MemOp mop,
                      target_ulong new_mask, target_ulong old_mask)
 {
-    TCGv tmp, addr = gen_ldst_addr(dc, a->rs1, a->imm, a->rs2_or_imm);
+    TCGv addr = gen_ldst_addr(dc, a->rs1, a->imm, a->rs2_or_imm);
+    TCGv tnew, told;
+
     if (addr == NULL) {
         return false;
     }
     if (gen_trap_ifnofpu(dc)) {
         return true;
     }
-    tmp = tcg_temp_new();
-    tcg_gen_qemu_ld_tl(tmp, addr, dc->mem_idx, mop | MO_ALIGN);
-    tcg_gen_andi_tl(tmp, tmp, new_mask);
-    tcg_gen_andi_tl(cpu_fsr, cpu_fsr, old_mask);
-    tcg_gen_or_tl(cpu_fsr, cpu_fsr, tmp);
-    gen_helper_set_fsr(tcg_env, cpu_fsr);
+    tnew = tcg_temp_new();
+    told = tcg_temp_new();
+    tcg_gen_qemu_ld_tl(tnew, addr, dc->mem_idx, mop | MO_ALIGN);
+    tcg_gen_andi_tl(tnew, tnew, new_mask);
+    tcg_gen_andi_tl(told, cpu_fsr, old_mask);
+    tcg_gen_or_tl(tnew, tnew, told);
+    gen_helper_set_fsr_noftt(tcg_env, tnew);
     return advance_pc(dc);
 }
 
-- 
2.34.1

[PATCH 20/22] target/sparc: Remove cpu_fsr

[Richard Henderson]

Drop this field as a tcg global, loading it explicitly in the
few places required.  This means that all FPop helpers may
once again be TCG_CALL_NO_WG.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/sparc/helper.h     | 120 +++++++++++++++++++-------------------
 target/sparc/fop_helper.c |   9 ++-
 target/sparc/translate.c  |  98 ++++++++++++++++---------------
 3 files changed, 114 insertions(+), 113 deletions(-)

diff --git a/target/sparc/helper.h b/target/sparc/helper.h
index 7466164468..c8e14fe371 100644
--- a/target/sparc/helper.h
+++ b/target/sparc/helper.h
@@ -36,79 +36,79 @@ DEF_HELPER_FLAGS_4(ld_asi, TCG_CALL_NO_WG, i64, env, tl, int, i32)
 DEF_HELPER_FLAGS_5(st_asi, TCG_CALL_NO_WG, void, env, tl, i64, int, i32)
 #endif
 DEF_HELPER_FLAGS_1(get_fsr, TCG_CALL_NO_WG_SE, tl, env)
-DEF_HELPER_FLAGS_2(set_fsr_noftt, 0, void, env, tl)
-DEF_HELPER_FLAGS_2(fsqrts, 0, f32, env, f32)
-DEF_HELPER_FLAGS_2(fsqrtd, 0, f64, env, f64)
-DEF_HELPER_FLAGS_2(fsqrtq, 0, i128, env, i128)
-DEF_HELPER_FLAGS_3(fcmps, 0, tl, env, f32, f32)
-DEF_HELPER_FLAGS_3(fcmpd, 0, tl, env, f64, f64)
-DEF_HELPER_FLAGS_3(fcmpes, 0, tl, env, f32, f32)
-DEF_HELPER_FLAGS_3(fcmped, 0, tl, env, f64, f64)
-DEF_HELPER_FLAGS_3(fcmpq, 0, tl, env, i128, i128)
-DEF_HELPER_FLAGS_3(fcmpeq, 0, tl, env, i128, i128)
+DEF_HELPER_FLAGS_2(set_fsr_noftt, TCG_CALL_NO_RWG, void, env, tl)
+DEF_HELPER_FLAGS_2(fsqrts, TCG_CALL_NO_WG, f32, env, f32)
+DEF_HELPER_FLAGS_2(fsqrtd, TCG_CALL_NO_WG, f64, env, f64)
+DEF_HELPER_FLAGS_2(fsqrtq, TCG_CALL_NO_WG, i128, env, i128)
+DEF_HELPER_FLAGS_3(fcmps, TCG_CALL_NO_WG, void, env, f32, f32)
+DEF_HELPER_FLAGS_3(fcmpd, TCG_CALL_NO_WG, void, env, f64, f64)
+DEF_HELPER_FLAGS_3(fcmpes, TCG_CALL_NO_WG, void, env, f32, f32)
+DEF_HELPER_FLAGS_3(fcmped, TCG_CALL_NO_WG, void, env, f64, f64)
+DEF_HELPER_FLAGS_3(fcmpq, TCG_CALL_NO_WG, void, env, i128, i128)
+DEF_HELPER_FLAGS_3(fcmpeq, TCG_CALL_NO_WG, void, env, i128, i128)
 #ifdef TARGET_SPARC64
-DEF_HELPER_FLAGS_3(fcmps_fcc1, 0, tl, env, f32, f32)
-DEF_HELPER_FLAGS_3(fcmps_fcc2, 0, tl, env, f32, f32)
-DEF_HELPER_FLAGS_3(fcmps_fcc3, 0, tl, env, f32, f32)
-DEF_HELPER_FLAGS_3(fcmpd_fcc1, 0, tl, env, f64, f64)
-DEF_HELPER_FLAGS_3(fcmpd_fcc2, 0, tl, env, f64, f64)
-DEF_HELPER_FLAGS_3(fcmpd_fcc3, 0, tl, env, f64, f64)
-DEF_HELPER_FLAGS_3(fcmpes_fcc1, 0, tl, env, f32, f32)
-DEF_HELPER_FLAGS_3(fcmpes_fcc2, 0, tl, env, f32, f32)
-DEF_HELPER_FLAGS_3(fcmpes_fcc3, 0, tl, env, f32, f32)
-DEF_HELPER_FLAGS_3(fcmped_fcc1, 0, tl, env, f64, f64)
-DEF_HELPER_FLAGS_3(fcmped_fcc2, 0, tl, env, f64, f64)
-DEF_HELPER_FLAGS_3(fcmped_fcc3, 0, tl, env, f64, f64)
-DEF_HELPER_FLAGS_3(fcmpq_fcc1, 0, tl, env, i128, i128)
-DEF_HELPER_FLAGS_3(fcmpq_fcc2, 0, tl, env, i128, i128)
-DEF_HELPER_FLAGS_3(fcmpq_fcc3, 0, tl, env, i128, i128)
-DEF_HELPER_FLAGS_3(fcmpeq_fcc1, 0, tl, env, i128, i128)
-DEF_HELPER_FLAGS_3(fcmpeq_fcc2, 0, tl, env, i128, i128)
-DEF_HELPER_FLAGS_3(fcmpeq_fcc3, 0, tl, env, i128, i128)
+DEF_HELPER_FLAGS_3(fcmps_fcc1, TCG_CALL_NO_WG, void, env, f32, f32)
+DEF_HELPER_FLAGS_3(fcmps_fcc2, TCG_CALL_NO_WG, void, env, f32, f32)
+DEF_HELPER_FLAGS_3(fcmps_fcc3, TCG_CALL_NO_WG, void, env, f32, f32)
+DEF_HELPER_FLAGS_3(fcmpd_fcc1, TCG_CALL_NO_WG, void, env, f64, f64)
+DEF_HELPER_FLAGS_3(fcmpd_fcc2, TCG_CALL_NO_WG, void, env, f64, f64)
+DEF_HELPER_FLAGS_3(fcmpd_fcc3, TCG_CALL_NO_WG, void, env, f64, f64)
+DEF_HELPER_FLAGS_3(fcmpes_fcc1, TCG_CALL_NO_WG, void, env, f32, f32)
+DEF_HELPER_FLAGS_3(fcmpes_fcc2, TCG_CALL_NO_WG, void, env, f32, f32)
+DEF_HELPER_FLAGS_3(fcmpes_fcc3, TCG_CALL_NO_WG, void, env, f32, f32)
+DEF_HELPER_FLAGS_3(fcmped_fcc1, TCG_CALL_NO_WG, void, env, f64, f64)
+DEF_HELPER_FLAGS_3(fcmped_fcc2, TCG_CALL_NO_WG, void, env, f64, f64)
+DEF_HELPER_FLAGS_3(fcmped_fcc3, TCG_CALL_NO_WG, void, env, f64, f64)
+DEF_HELPER_FLAGS_3(fcmpq_fcc1, TCG_CALL_NO_WG, void, env, i128, i128)
+DEF_HELPER_FLAGS_3(fcmpq_fcc2, TCG_CALL_NO_WG, void, env, i128, i128)
+DEF_HELPER_FLAGS_3(fcmpq_fcc3, TCG_CALL_NO_WG, void, env, i128, i128)
+DEF_HELPER_FLAGS_3(fcmpeq_fcc1, TCG_CALL_NO_WG, void, env, i128, i128)
+DEF_HELPER_FLAGS_3(fcmpeq_fcc2, TCG_CALL_NO_WG, void, env, i128, i128)
+DEF_HELPER_FLAGS_3(fcmpeq_fcc3, TCG_CALL_NO_WG, void, env, i128, i128)
 #endif
 DEF_HELPER_2(raise_exception, noreturn, env, int)
 
-DEF_HELPER_FLAGS_3(faddd, 0, f64, env, f64, f64)
-DEF_HELPER_FLAGS_3(fsubd, 0, f64, env, f64, f64)
-DEF_HELPER_FLAGS_3(fmuld, 0, f64, env, f64, f64)
-DEF_HELPER_FLAGS_3(fdivd, 0, f64, env, f64, f64)
+DEF_HELPER_FLAGS_3(faddd, TCG_CALL_NO_WG, f64, env, f64, f64)
+DEF_HELPER_FLAGS_3(fsubd, TCG_CALL_NO_WG, f64, env, f64, f64)
+DEF_HELPER_FLAGS_3(fmuld, TCG_CALL_NO_WG, f64, env, f64, f64)
+DEF_HELPER_FLAGS_3(fdivd, TCG_CALL_NO_WG, f64, env, f64, f64)
 
-DEF_HELPER_FLAGS_3(faddq, 0, i128, env, i128, i128)
-DEF_HELPER_FLAGS_3(fsubq, 0, i128, env, i128, i128)
-DEF_HELPER_FLAGS_3(fmulq, 0, i128, env, i128, i128)
-DEF_HELPER_FLAGS_3(fdivq, 0, i128, env, i128, i128)
+DEF_HELPER_FLAGS_3(faddq, TCG_CALL_NO_WG, i128, env, i128, i128)
+DEF_HELPER_FLAGS_3(fsubq, TCG_CALL_NO_WG, i128, env, i128, i128)
+DEF_HELPER_FLAGS_3(fmulq, TCG_CALL_NO_WG, i128, env, i128, i128)
+DEF_HELPER_FLAGS_3(fdivq, TCG_CALL_NO_WG, i128, env, i128, i128)
 
-DEF_HELPER_FLAGS_3(fadds, 0, f32, env, f32, f32)
-DEF_HELPER_FLAGS_3(fsubs, 0, f32, env, f32, f32)
-DEF_HELPER_FLAGS_3(fmuls, 0, f32, env, f32, f32)
-DEF_HELPER_FLAGS_3(fdivs, 0, f32, env, f32, f32)
+DEF_HELPER_FLAGS_3(fadds, TCG_CALL_NO_WG, f32, env, f32, f32)
+DEF_HELPER_FLAGS_3(fsubs, TCG_CALL_NO_WG, f32, env, f32, f32)
+DEF_HELPER_FLAGS_3(fmuls, TCG_CALL_NO_WG, f32, env, f32, f32)
+DEF_HELPER_FLAGS_3(fdivs, TCG_CALL_NO_WG, f32, env, f32, f32)
 
-DEF_HELPER_FLAGS_3(fsmuld, 0, f64, env, f32, f32)
-DEF_HELPER_FLAGS_3(fdmulq, 0, i128, env, f64, f64)
+DEF_HELPER_FLAGS_3(fsmuld, TCG_CALL_NO_WG, f64, env, f32, f32)
+DEF_HELPER_FLAGS_3(fdmulq, TCG_CALL_NO_WG, i128, env, f64, f64)
 
-DEF_HELPER_FLAGS_2(fitod, 0, f64, env, s32)
-DEF_HELPER_FLAGS_2(fitoq, 0, i128, env, s32)
+DEF_HELPER_FLAGS_2(fitod, TCG_CALL_NO_WG, f64, env, s32)
+DEF_HELPER_FLAGS_2(fitoq, TCG_CALL_NO_WG, i128, env, s32)
 
-DEF_HELPER_FLAGS_2(fitos, 0, f32, env, s32)
+DEF_HELPER_FLAGS_2(fitos, TCG_CALL_NO_WG, f32, env, s32)
 
 #ifdef TARGET_SPARC64
-DEF_HELPER_FLAGS_2(fxtos, 0, f32, env, s64)
-DEF_HELPER_FLAGS_2(fxtod, 0, f64, env, s64)
-DEF_HELPER_FLAGS_2(fxtoq, 0, i128, env, s64)
+DEF_HELPER_FLAGS_2(fxtos, TCG_CALL_NO_WG, f32, env, s64)
+DEF_HELPER_FLAGS_2(fxtod, TCG_CALL_NO_WG, f64, env, s64)
+DEF_HELPER_FLAGS_2(fxtoq, TCG_CALL_NO_WG, i128, env, s64)
 #endif
-DEF_HELPER_FLAGS_2(fdtos, 0, f32, env, f64)
-DEF_HELPER_FLAGS_2(fstod, 0, f64, env, f32)
-DEF_HELPER_FLAGS_2(fqtos, 0, f32, env, i128)
-DEF_HELPER_FLAGS_2(fstoq, 0, i128, env, f32)
-DEF_HELPER_FLAGS_2(fqtod, 0, f64, env, i128)
-DEF_HELPER_FLAGS_2(fdtoq, 0, i128, env, f64)
-DEF_HELPER_FLAGS_2(fstoi, 0, s32, env, f32)
-DEF_HELPER_FLAGS_2(fdtoi, 0, s32, env, f64)
-DEF_HELPER_FLAGS_2(fqtoi, 0, s32, env, i128)
+DEF_HELPER_FLAGS_2(fdtos, TCG_CALL_NO_WG, f32, env, f64)
+DEF_HELPER_FLAGS_2(fstod, TCG_CALL_NO_WG, f64, env, f32)
+DEF_HELPER_FLAGS_2(fqtos, TCG_CALL_NO_WG, f32, env, i128)
+DEF_HELPER_FLAGS_2(fstoq, TCG_CALL_NO_WG, i128, env, f32)
+DEF_HELPER_FLAGS_2(fqtod, TCG_CALL_NO_WG, f64, env, i128)
+DEF_HELPER_FLAGS_2(fdtoq, TCG_CALL_NO_WG, i128, env, f64)
+DEF_HELPER_FLAGS_2(fstoi, TCG_CALL_NO_WG, s32, env, f32)
+DEF_HELPER_FLAGS_2(fdtoi, TCG_CALL_NO_WG, s32, env, f64)
+DEF_HELPER_FLAGS_2(fqtoi, TCG_CALL_NO_WG, s32, env, i128)
 #ifdef TARGET_SPARC64
-DEF_HELPER_FLAGS_2(fstox, 0, s64, env, f32)
-DEF_HELPER_FLAGS_2(fdtox, 0, s64, env, f64)
-DEF_HELPER_FLAGS_2(fqtox, 0, s64, env, i128)
+DEF_HELPER_FLAGS_2(fstox, TCG_CALL_NO_WG, s64, env, f32)
+DEF_HELPER_FLAGS_2(fdtox, TCG_CALL_NO_WG, s64, env, f64)
+DEF_HELPER_FLAGS_2(fqtox, TCG_CALL_NO_WG, s64, env, i128)
 
 DEF_HELPER_FLAGS_2(fpmerge, TCG_CALL_NO_RWG_SE, i64, i64, i64)
 DEF_HELPER_FLAGS_2(fmul8x16, TCG_CALL_NO_RWG_SE, i64, i64, i64)
diff --git a/target/sparc/fop_helper.c b/target/sparc/fop_helper.c
index ac30f88810..796f448bfd 100644
--- a/target/sparc/fop_helper.c
+++ b/target/sparc/fop_helper.c
@@ -344,8 +344,7 @@ Int128 helper_fsqrtq(CPUSPARCState *env, Int128 src)
 }
 
 #define GEN_FCMP(name, size, FS, E)                                     \
-    target_ulong glue(helper_, name) (CPUSPARCState *env,               \
-                                      Int128 src1, Int128 src2)         \
+    void glue(helper_, name)(CPUSPARCState *env, Int128 src1, Int128 src2) \
     {                                                                   \
         float128 reg1 = f128_in(src1);                                  \
         float128 reg2 = f128_in(src2);                                  \
@@ -376,10 +375,10 @@ Int128 helper_fsqrtq(CPUSPARCState *env, Int128 src)
             fsr &= ~((FSR_FCC1 | FSR_FCC0) << FS);                      \
             break;                                                      \
         }                                                               \
-        return fsr;                                                     \
+        env->fsr = fsr;                                                 \
     }
 #define GEN_FCMP_T(name, size, FS, E)                                   \
-    target_ulong glue(helper_, name)(CPUSPARCState *env, size src1, size src2)\
+    void glue(helper_, name)(CPUSPARCState *env, size src1, size src2)  \
     {                                                                   \
         FloatRelation ret;                                              \
         target_ulong fsr;                                               \
@@ -407,7 +406,7 @@ Int128 helper_fsqrtq(CPUSPARCState *env, Int128 src)
             fsr &= ~((FSR_FCC1 | FSR_FCC0) << FS);                      \
             break;                                                      \
         }                                                               \
-        return fsr;                                                     \
+        env->fsr = fsr;                                                 \
     }
 
 GEN_FCMP_T(fcmps, float32, 0, 0);
diff --git a/target/sparc/translate.c b/target/sparc/translate.c
index d2145dcc0b..4eb6291b81 100644
--- a/target/sparc/translate.c
+++ b/target/sparc/translate.c
@@ -99,7 +99,7 @@
 
 /* global register indexes */
 static TCGv_ptr cpu_regwptr;
-static TCGv cpu_fsr, cpu_pc, cpu_npc;
+static TCGv cpu_pc, cpu_npc;
 static TCGv cpu_regs[32];
 static TCGv cpu_y;
 static TCGv cpu_tbr;
@@ -1097,7 +1097,7 @@ static void gen_compare(DisasCompare *cmp, bool xcc, unsigned int cond,
 static void gen_fcompare(DisasCompare *cmp, unsigned int cc, unsigned int cond)
 {
     unsigned int offset;
-    TCGv r_dst;
+    TCGv r_dst, fsr;
 
     /* For now we still generate a straight boolean result.  */
     cmp->cond = TCG_COND_NE;
@@ -1120,54 +1120,56 @@ static void gen_fcompare(DisasCompare *cmp, unsigned int cc, unsigned int cond)
         break;
     }
 
+    fsr = tcg_temp_new();
+    tcg_gen_ld_tl(fsr, tcg_env, offsetof(CPUSPARCState, fsr));
     switch (cond) {
     case 0x0:
         gen_op_eval_bn(r_dst);
         break;
     case 0x1:
-        gen_op_eval_fbne(r_dst, cpu_fsr, offset);
+        gen_op_eval_fbne(r_dst, fsr, offset);
         break;
     case 0x2:
-        gen_op_eval_fblg(r_dst, cpu_fsr, offset);
+        gen_op_eval_fblg(r_dst, fsr, offset);
         break;
     case 0x3:
-        gen_op_eval_fbul(r_dst, cpu_fsr, offset);
+        gen_op_eval_fbul(r_dst, fsr, offset);
         break;
     case 0x4:
-        gen_op_eval_fbl(r_dst, cpu_fsr, offset);
+        gen_op_eval_fbl(r_dst, fsr, offset);
         break;
     case 0x5:
-        gen_op_eval_fbug(r_dst, cpu_fsr, offset);
+        gen_op_eval_fbug(r_dst, fsr, offset);
         break;
     case 0x6:
-        gen_op_eval_fbg(r_dst, cpu_fsr, offset);
+        gen_op_eval_fbg(r_dst, fsr, offset);
         break;
     case 0x7:
-        gen_op_eval_fbu(r_dst, cpu_fsr, offset);
+        gen_op_eval_fbu(r_dst, fsr, offset);
         break;
     case 0x8:
         gen_op_eval_ba(r_dst);
         break;
     case 0x9:
-        gen_op_eval_fbe(r_dst, cpu_fsr, offset);
+        gen_op_eval_fbe(r_dst, fsr, offset);
         break;
     case 0xa:
-        gen_op_eval_fbue(r_dst, cpu_fsr, offset);
+        gen_op_eval_fbue(r_dst, fsr, offset);
         break;
     case 0xb:
-        gen_op_eval_fbge(r_dst, cpu_fsr, offset);
+        gen_op_eval_fbge(r_dst, fsr, offset);
         break;
     case 0xc:
-        gen_op_eval_fbuge(r_dst, cpu_fsr, offset);
+        gen_op_eval_fbuge(r_dst, fsr, offset);
         break;
     case 0xd:
-        gen_op_eval_fble(r_dst, cpu_fsr, offset);
+        gen_op_eval_fble(r_dst, fsr, offset);
         break;
     case 0xe:
-        gen_op_eval_fbule(r_dst, cpu_fsr, offset);
+        gen_op_eval_fbule(r_dst, fsr, offset);
         break;
     case 0xf:
-        gen_op_eval_fbo(r_dst, cpu_fsr, offset);
+        gen_op_eval_fbo(r_dst, fsr, offset);
         break;
     }
 }
@@ -1264,16 +1266,16 @@ static void gen_op_fcmps(int fccno, TCGv_i32 r_rs1, TCGv_i32 r_rs2)
 {
     switch (fccno) {
     case 0:
-        gen_helper_fcmps(cpu_fsr, tcg_env, r_rs1, r_rs2);
+        gen_helper_fcmps(tcg_env, r_rs1, r_rs2);
         break;
     case 1:
-        gen_helper_fcmps_fcc1(cpu_fsr, tcg_env, r_rs1, r_rs2);
+        gen_helper_fcmps_fcc1(tcg_env, r_rs1, r_rs2);
         break;
     case 2:
-        gen_helper_fcmps_fcc2(cpu_fsr, tcg_env, r_rs1, r_rs2);
+        gen_helper_fcmps_fcc2(tcg_env, r_rs1, r_rs2);
         break;
     case 3:
-        gen_helper_fcmps_fcc3(cpu_fsr, tcg_env, r_rs1, r_rs2);
+        gen_helper_fcmps_fcc3(tcg_env, r_rs1, r_rs2);
         break;
     }
 }
@@ -1282,16 +1284,16 @@ static void gen_op_fcmpd(int fccno, TCGv_i64 r_rs1, TCGv_i64 r_rs2)
 {
     switch (fccno) {
     case 0:
-        gen_helper_fcmpd(cpu_fsr, tcg_env, r_rs1, r_rs2);
+        gen_helper_fcmpd(tcg_env, r_rs1, r_rs2);
         break;
     case 1:
-        gen_helper_fcmpd_fcc1(cpu_fsr, tcg_env, r_rs1, r_rs2);
+        gen_helper_fcmpd_fcc1(tcg_env, r_rs1, r_rs2);
         break;
     case 2:
-        gen_helper_fcmpd_fcc2(cpu_fsr, tcg_env, r_rs1, r_rs2);
+        gen_helper_fcmpd_fcc2(tcg_env, r_rs1, r_rs2);
         break;
     case 3:
-        gen_helper_fcmpd_fcc3(cpu_fsr, tcg_env, r_rs1, r_rs2);
+        gen_helper_fcmpd_fcc3(tcg_env, r_rs1, r_rs2);
         break;
     }
 }
@@ -1300,16 +1302,16 @@ static void gen_op_fcmpq(int fccno, TCGv_i128 r_rs1, TCGv_i128 r_rs2)
 {
     switch (fccno) {
     case 0:
-        gen_helper_fcmpq(cpu_fsr, tcg_env, r_rs1, r_rs2);
+        gen_helper_fcmpq(tcg_env, r_rs1, r_rs2);
         break;
     case 1:
-        gen_helper_fcmpq_fcc1(cpu_fsr, tcg_env, r_rs1, r_rs2);
+        gen_helper_fcmpq_fcc1(tcg_env, r_rs1, r_rs2);
         break;
     case 2:
-        gen_helper_fcmpq_fcc2(cpu_fsr, tcg_env, r_rs1, r_rs2);
+        gen_helper_fcmpq_fcc2(tcg_env, r_rs1, r_rs2);
         break;
     case 3:
-        gen_helper_fcmpq_fcc3(cpu_fsr, tcg_env, r_rs1, r_rs2);
+        gen_helper_fcmpq_fcc3(tcg_env, r_rs1, r_rs2);
         break;
     }
 }
@@ -1318,16 +1320,16 @@ static void gen_op_fcmpes(int fccno, TCGv_i32 r_rs1, TCGv_i32 r_rs2)
 {
     switch (fccno) {
     case 0:
-        gen_helper_fcmpes(cpu_fsr, tcg_env, r_rs1, r_rs2);
+        gen_helper_fcmpes(tcg_env, r_rs1, r_rs2);
         break;
     case 1:
-        gen_helper_fcmpes_fcc1(cpu_fsr, tcg_env, r_rs1, r_rs2);
+        gen_helper_fcmpes_fcc1(tcg_env, r_rs1, r_rs2);
         break;
     case 2:
-        gen_helper_fcmpes_fcc2(cpu_fsr, tcg_env, r_rs1, r_rs2);
+        gen_helper_fcmpes_fcc2(tcg_env, r_rs1, r_rs2);
         break;
     case 3:
-        gen_helper_fcmpes_fcc3(cpu_fsr, tcg_env, r_rs1, r_rs2);
+        gen_helper_fcmpes_fcc3(tcg_env, r_rs1, r_rs2);
         break;
     }
 }
@@ -1336,16 +1338,16 @@ static void gen_op_fcmped(int fccno, TCGv_i64 r_rs1, TCGv_i64 r_rs2)
 {
     switch (fccno) {
     case 0:
-        gen_helper_fcmped(cpu_fsr, tcg_env, r_rs1, r_rs2);
+        gen_helper_fcmped(tcg_env, r_rs1, r_rs2);
         break;
     case 1:
-        gen_helper_fcmped_fcc1(cpu_fsr, tcg_env, r_rs1, r_rs2);
+        gen_helper_fcmped_fcc1(tcg_env, r_rs1, r_rs2);
         break;
     case 2:
-        gen_helper_fcmped_fcc2(cpu_fsr, tcg_env, r_rs1, r_rs2);
+        gen_helper_fcmped_fcc2(tcg_env, r_rs1, r_rs2);
         break;
     case 3:
-        gen_helper_fcmped_fcc3(cpu_fsr, tcg_env, r_rs1, r_rs2);
+        gen_helper_fcmped_fcc3(tcg_env, r_rs1, r_rs2);
         break;
     }
 }
@@ -1354,16 +1356,16 @@ static void gen_op_fcmpeq(int fccno, TCGv_i128 r_rs1, TCGv_i128 r_rs2)
 {
     switch (fccno) {
     case 0:
-        gen_helper_fcmpeq(cpu_fsr, tcg_env, r_rs1, r_rs2);
+        gen_helper_fcmpeq(tcg_env, r_rs1, r_rs2);
         break;
     case 1:
-        gen_helper_fcmpeq_fcc1(cpu_fsr, tcg_env, r_rs1, r_rs2);
+        gen_helper_fcmpeq_fcc1(tcg_env, r_rs1, r_rs2);
         break;
     case 2:
-        gen_helper_fcmpeq_fcc2(cpu_fsr, tcg_env, r_rs1, r_rs2);
+        gen_helper_fcmpeq_fcc2(tcg_env, r_rs1, r_rs2);
         break;
     case 3:
-        gen_helper_fcmpeq_fcc3(cpu_fsr, tcg_env, r_rs1, r_rs2);
+        gen_helper_fcmpeq_fcc3(tcg_env, r_rs1, r_rs2);
         break;
     }
 }
@@ -1372,32 +1374,32 @@ static void gen_op_fcmpeq(int fccno, TCGv_i128 r_rs1, TCGv_i128 r_rs2)
 
 static void gen_op_fcmps(int fccno, TCGv r_rs1, TCGv r_rs2)
 {
-    gen_helper_fcmps(cpu_fsr, tcg_env, r_rs1, r_rs2);
+    gen_helper_fcmps(tcg_env, r_rs1, r_rs2);
 }
 
 static void gen_op_fcmpd(int fccno, TCGv_i64 r_rs1, TCGv_i64 r_rs2)
 {
-    gen_helper_fcmpd(cpu_fsr, tcg_env, r_rs1, r_rs2);
+    gen_helper_fcmpd(tcg_env, r_rs1, r_rs2);
 }
 
 static void gen_op_fcmpq(int fccno, TCGv_i128 r_rs1, TCGv_i128 r_rs2)
 {
-    gen_helper_fcmpq(cpu_fsr, tcg_env, r_rs1, r_rs2);
+    gen_helper_fcmpq(tcg_env, r_rs1, r_rs2);
 }
 
 static void gen_op_fcmpes(int fccno, TCGv r_rs1, TCGv r_rs2)
 {
-    gen_helper_fcmpes(cpu_fsr, tcg_env, r_rs1, r_rs2);
+    gen_helper_fcmpes(tcg_env, r_rs1, r_rs2);
 }
 
 static void gen_op_fcmped(int fccno, TCGv_i64 r_rs1, TCGv_i64 r_rs2)
 {
-    gen_helper_fcmped(cpu_fsr, tcg_env, r_rs1, r_rs2);
+    gen_helper_fcmped(tcg_env, r_rs1, r_rs2);
 }
 
 static void gen_op_fcmpeq(int fccno, TCGv_i128 r_rs1, TCGv_i128 r_rs2)
 {
-    gen_helper_fcmpeq(cpu_fsr, tcg_env, r_rs1, r_rs2);
+    gen_helper_fcmpeq(tcg_env, r_rs1, r_rs2);
 }
 #endif
 
@@ -4413,8 +4415,9 @@ static bool do_ldfsr(DisasContext *dc, arg_r_r_ri *a, MemOp mop,
     tnew = tcg_temp_new();
     told = tcg_temp_new();
     tcg_gen_qemu_ld_tl(tnew, addr, dc->mem_idx, mop | MO_ALIGN);
+    tcg_gen_ld_tl(told, tcg_env, offsetof(CPUSPARCState, fsr));
     tcg_gen_andi_tl(tnew, tnew, new_mask);
-    tcg_gen_andi_tl(told, cpu_fsr, old_mask);
+    tcg_gen_andi_tl(told, told, old_mask);
     tcg_gen_or_tl(tnew, tnew, told);
     gen_helper_set_fsr_noftt(tcg_env, tnew);
     return advance_pc(dc);
@@ -5342,7 +5345,6 @@ void sparc_tcg_init(void)
         { &cpu_icc_Z, offsetof(CPUSPARCState, icc_Z), "icc_Z" },
         { &cpu_icc_C, offsetof(CPUSPARCState, icc_C), "icc_C" },
         { &cpu_cond, offsetof(CPUSPARCState, cond), "cond" },
-        { &cpu_fsr, offsetof(CPUSPARCState, fsr), "fsr" },
         { &cpu_pc, offsetof(CPUSPARCState, pc), "pc" },
         { &cpu_npc, offsetof(CPUSPARCState, npc), "npc" },
         { &cpu_y, offsetof(CPUSPARCState, y), "y" },
-- 
2.34.1

[PATCH 21/22] target/sparc: Split fcc out of env->fsr

[Richard Henderson]

Represent each fcc field separately from the rest of fsr.
This vastly simplifies floating-point comparisons.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/sparc/cpu.h        |  20 +-
 target/sparc/helper.h     |  34 +--
 target/sparc/fop_helper.c | 169 ++++++-------
 target/sparc/translate.c  | 503 +++++++++-----------------------------
 4 files changed, 201 insertions(+), 525 deletions(-)

diff --git a/target/sparc/cpu.h b/target/sparc/cpu.h
index 90a7dfb004..30bab9a7b3 100644
--- a/target/sparc/cpu.h
+++ b/target/sparc/cpu.h
@@ -31,8 +31,10 @@
 
 #if !defined(TARGET_SPARC64)
 #define TARGET_DPREGS 16
+#define TARGET_FCCREGS 1
 #else
 #define TARGET_DPREGS 32
+#define TARGET_FCCREGS 4
 #endif
 
 /*#define EXCP_INTERRUPT 0x100*/
@@ -203,24 +205,19 @@ enum {
 #ifdef TARGET_SPARC64
 #define FSR_FTT_NMASK      0xfffffffffffe3fffULL
 #define FSR_FTT_CEXC_NMASK 0xfffffffffffe3fe0ULL
-#define FSR_LDFSR_OLDMASK  0x0000003f000fc000ULL
-#define FSR_LDXFSR_MASK    0x0000003fcfc00fffULL
-#define FSR_LDXFSR_OLDMASK 0x00000000000fc000ULL
 #else
 #define FSR_FTT_NMASK      0xfffe3fffULL
 #define FSR_FTT_CEXC_NMASK 0xfffe3fe0ULL
-#define FSR_LDFSR_OLDMASK  0x000fc000ULL
 #endif
-#define FSR_LDFSR_MASK     0xcfc00fffULL
 #define FSR_FTT_IEEE_EXCP (1ULL << 14)
 #define FSR_FTT_UNIMPFPOP (3ULL << 14)
 #define FSR_FTT_SEQ_ERROR (4ULL << 14)
 #define FSR_FTT_INVAL_FPR (6ULL << 14)
 
-#define FSR_FCC1_SHIFT 11
-#define FSR_FCC1  (1ULL << FSR_FCC1_SHIFT)
-#define FSR_FCC0_SHIFT 10
-#define FSR_FCC0  (1ULL << FSR_FCC0_SHIFT)
+#define FSR_FCC0_SHIFT    10
+#define FSR_FCC1_SHIFT    32
+#define FSR_FCC2_SHIFT    34
+#define FSR_FCC3_SHIFT    36
 
 /* MMU */
 #define MMU_E     (1<<0)
@@ -467,8 +464,9 @@ struct CPUArchState {
                           temporary register when possible) */
 
     /* FPU State Register, in parts */
-    target_ulong fsr;       /* rm, tem, aexc, fcc* */
-    uint32_t fsr_cexc_ftt;  /* cexc, ftt */
+    uint32_t fsr;                    /* rm, tem, aexc */
+    uint32_t fsr_cexc_ftt;           /* cexc, ftt */
+    uint32_t fcc[TARGET_FCCREGS];    /* fcc* */
 
     CPU_DoubleU fpr[TARGET_DPREGS];  /* floating point registers */
     uint32_t cwp;      /* index of current register window (extracted
diff --git a/target/sparc/helper.h b/target/sparc/helper.h
index c8e14fe371..6a42ba4e9e 100644
--- a/target/sparc/helper.h
+++ b/target/sparc/helper.h
@@ -36,36 +36,16 @@ DEF_HELPER_FLAGS_4(ld_asi, TCG_CALL_NO_WG, i64, env, tl, int, i32)
 DEF_HELPER_FLAGS_5(st_asi, TCG_CALL_NO_WG, void, env, tl, i64, int, i32)
 #endif
 DEF_HELPER_FLAGS_1(get_fsr, TCG_CALL_NO_WG_SE, tl, env)
-DEF_HELPER_FLAGS_2(set_fsr_noftt, TCG_CALL_NO_RWG, void, env, tl)
+DEF_HELPER_FLAGS_2(set_fsr_nofcc_noftt, TCG_CALL_NO_RWG, void, env, i32)
 DEF_HELPER_FLAGS_2(fsqrts, TCG_CALL_NO_WG, f32, env, f32)
 DEF_HELPER_FLAGS_2(fsqrtd, TCG_CALL_NO_WG, f64, env, f64)
 DEF_HELPER_FLAGS_2(fsqrtq, TCG_CALL_NO_WG, i128, env, i128)
-DEF_HELPER_FLAGS_3(fcmps, TCG_CALL_NO_WG, void, env, f32, f32)
-DEF_HELPER_FLAGS_3(fcmpd, TCG_CALL_NO_WG, void, env, f64, f64)
-DEF_HELPER_FLAGS_3(fcmpes, TCG_CALL_NO_WG, void, env, f32, f32)
-DEF_HELPER_FLAGS_3(fcmped, TCG_CALL_NO_WG, void, env, f64, f64)
-DEF_HELPER_FLAGS_3(fcmpq, TCG_CALL_NO_WG, void, env, i128, i128)
-DEF_HELPER_FLAGS_3(fcmpeq, TCG_CALL_NO_WG, void, env, i128, i128)
-#ifdef TARGET_SPARC64
-DEF_HELPER_FLAGS_3(fcmps_fcc1, TCG_CALL_NO_WG, void, env, f32, f32)
-DEF_HELPER_FLAGS_3(fcmps_fcc2, TCG_CALL_NO_WG, void, env, f32, f32)
-DEF_HELPER_FLAGS_3(fcmps_fcc3, TCG_CALL_NO_WG, void, env, f32, f32)
-DEF_HELPER_FLAGS_3(fcmpd_fcc1, TCG_CALL_NO_WG, void, env, f64, f64)
-DEF_HELPER_FLAGS_3(fcmpd_fcc2, TCG_CALL_NO_WG, void, env, f64, f64)
-DEF_HELPER_FLAGS_3(fcmpd_fcc3, TCG_CALL_NO_WG, void, env, f64, f64)
-DEF_HELPER_FLAGS_3(fcmpes_fcc1, TCG_CALL_NO_WG, void, env, f32, f32)
-DEF_HELPER_FLAGS_3(fcmpes_fcc2, TCG_CALL_NO_WG, void, env, f32, f32)
-DEF_HELPER_FLAGS_3(fcmpes_fcc3, TCG_CALL_NO_WG, void, env, f32, f32)
-DEF_HELPER_FLAGS_3(fcmped_fcc1, TCG_CALL_NO_WG, void, env, f64, f64)
-DEF_HELPER_FLAGS_3(fcmped_fcc2, TCG_CALL_NO_WG, void, env, f64, f64)
-DEF_HELPER_FLAGS_3(fcmped_fcc3, TCG_CALL_NO_WG, void, env, f64, f64)
-DEF_HELPER_FLAGS_3(fcmpq_fcc1, TCG_CALL_NO_WG, void, env, i128, i128)
-DEF_HELPER_FLAGS_3(fcmpq_fcc2, TCG_CALL_NO_WG, void, env, i128, i128)
-DEF_HELPER_FLAGS_3(fcmpq_fcc3, TCG_CALL_NO_WG, void, env, i128, i128)
-DEF_HELPER_FLAGS_3(fcmpeq_fcc1, TCG_CALL_NO_WG, void, env, i128, i128)
-DEF_HELPER_FLAGS_3(fcmpeq_fcc2, TCG_CALL_NO_WG, void, env, i128, i128)
-DEF_HELPER_FLAGS_3(fcmpeq_fcc3, TCG_CALL_NO_WG, void, env, i128, i128)
-#endif
+DEF_HELPER_FLAGS_3(fcmps, TCG_CALL_NO_WG, i32, env, f32, f32)
+DEF_HELPER_FLAGS_3(fcmpes, TCG_CALL_NO_WG, i32, env, f32, f32)
+DEF_HELPER_FLAGS_3(fcmpd, TCG_CALL_NO_WG, i32, env, f64, f64)
+DEF_HELPER_FLAGS_3(fcmped, TCG_CALL_NO_WG, i32, env, f64, f64)
+DEF_HELPER_FLAGS_3(fcmpq, TCG_CALL_NO_WG, i32, env, i128, i128)
+DEF_HELPER_FLAGS_3(fcmpeq, TCG_CALL_NO_WG, i32, env, i128, i128)
 DEF_HELPER_2(raise_exception, noreturn, env, int)
 
 DEF_HELPER_FLAGS_3(faddd, TCG_CALL_NO_WG, f64, env, f64, f64)
diff --git a/target/sparc/fop_helper.c b/target/sparc/fop_helper.c
index 796f448bfd..1205a599ef 100644
--- a/target/sparc/fop_helper.c
+++ b/target/sparc/fop_helper.c
@@ -343,113 +343,80 @@ Int128 helper_fsqrtq(CPUSPARCState *env, Int128 src)
     return f128_ret(ret);
 }
 
-#define GEN_FCMP(name, size, FS, E)                                     \
-    void glue(helper_, name)(CPUSPARCState *env, Int128 src1, Int128 src2) \
-    {                                                                   \
-        float128 reg1 = f128_in(src1);                                  \
-        float128 reg2 = f128_in(src2);                                  \
-        FloatRelation ret;                                              \
-        target_ulong fsr;                                               \
-        if (E) {                                                        \
-            ret = glue(size, _compare)(reg1, reg2, &env->fp_status);    \
-        } else {                                                        \
-            ret = glue(size, _compare_quiet)(reg1, reg2,                \
-                                             &env->fp_status);          \
-        }                                                               \
-        check_ieee_exceptions(env, GETPC());                            \
-        fsr = env->fsr;                                                 \
-        switch (ret) {                                                  \
-        case float_relation_unordered:                                  \
-            fsr |= (FSR_FCC1 | FSR_FCC0) << FS;                         \
-            fsr |= FSR_NVA;                                             \
-            break;                                                      \
-        case float_relation_less:                                       \
-            fsr &= ~(FSR_FCC1) << FS;                                   \
-            fsr |= FSR_FCC0 << FS;                                      \
-            break;                                                      \
-        case float_relation_greater:                                    \
-            fsr &= ~(FSR_FCC0) << FS;                                   \
-            fsr |= FSR_FCC1 << FS;                                      \
-            break;                                                      \
-        default:                                                        \
-            fsr &= ~((FSR_FCC1 | FSR_FCC0) << FS);                      \
-            break;                                                      \
-        }                                                               \
-        env->fsr = fsr;                                                 \
-    }
-#define GEN_FCMP_T(name, size, FS, E)                                   \
-    void glue(helper_, name)(CPUSPARCState *env, size src1, size src2)  \
-    {                                                                   \
-        FloatRelation ret;                                              \
-        target_ulong fsr;                                               \
-        if (E) {                                                        \
-            ret = glue(size, _compare)(src1, src2, &env->fp_status);    \
-        } else {                                                        \
-            ret = glue(size, _compare_quiet)(src1, src2,                \
-                                             &env->fp_status);          \
-        }                                                               \
-        check_ieee_exceptions(env, GETPC());                            \
-        fsr = env->fsr;                                                 \
-        switch (ret) {                                                  \
-        case float_relation_unordered:                                  \
-            fsr |= (FSR_FCC1 | FSR_FCC0) << FS;                         \
-            break;                                                      \
-        case float_relation_less:                                       \
-            fsr &= ~(FSR_FCC1 << FS);                                   \
-            fsr |= FSR_FCC0 << FS;                                      \
-            break;                                                      \
-        case float_relation_greater:                                    \
-            fsr &= ~(FSR_FCC0 << FS);                                   \
-            fsr |= FSR_FCC1 << FS;                                      \
-            break;                                                      \
-        default:                                                        \
-            fsr &= ~((FSR_FCC1 | FSR_FCC0) << FS);                      \
-            break;                                                      \
-        }                                                               \
-        env->fsr = fsr;                                                 \
+static uint32_t finish_fcmp(CPUSPARCState *env, FloatRelation r, uintptr_t ra)
+{
+    check_ieee_exceptions(env, ra);
+
+    /*
+     * FCC values:
+     * 0 =
+     * 1 <
+     * 2 >
+     * 3 unordered
+     */
+    switch (r) {
+    case float_relation_equal:
+        return 0;
+    case float_relation_less:
+        return 1;
+    case float_relation_greater:
+        return 2;
+    case float_relation_unordered:
+        env->fsr |= FSR_NVA;
+        return 3;
     }
+    g_assert_not_reached();
+}
 
-GEN_FCMP_T(fcmps, float32, 0, 0);
-GEN_FCMP_T(fcmpd, float64, 0, 0);
+uint32_t helper_fcmps(CPUSPARCState *env, float32 src1, float32 src2)
+{
+    FloatRelation r = float32_compare_quiet(src1, src2, &env->fp_status);
+    return finish_fcmp(env, r, GETPC());
+}
 
-GEN_FCMP_T(fcmpes, float32, 0, 1);
-GEN_FCMP_T(fcmped, float64, 0, 1);
+uint32_t helper_fcmpes(CPUSPARCState *env, float32 src1, float32 src2)
+{
+    FloatRelation r = float32_compare(src1, src2, &env->fp_status);
+    return finish_fcmp(env, r, GETPC());
+}
 
-GEN_FCMP(fcmpq, float128, 0, 0);
-GEN_FCMP(fcmpeq, float128, 0, 1);
+uint32_t helper_fcmpd(CPUSPARCState *env, float64 src1, float64 src2)
+{
+    FloatRelation r = float64_compare_quiet(src1, src2, &env->fp_status);
+    return finish_fcmp(env, r, GETPC());
+}
 
-#ifdef TARGET_SPARC64
-GEN_FCMP_T(fcmps_fcc1, float32, 22, 0);
-GEN_FCMP_T(fcmpd_fcc1, float64, 22, 0);
-GEN_FCMP(fcmpq_fcc1, float128, 22, 0);
+uint32_t helper_fcmped(CPUSPARCState *env, float64 src1, float64 src2)
+{
+    FloatRelation r = float64_compare(src1, src2, &env->fp_status);
+    return finish_fcmp(env, r, GETPC());
+}
 
-GEN_FCMP_T(fcmps_fcc2, float32, 24, 0);
-GEN_FCMP_T(fcmpd_fcc2, float64, 24, 0);
-GEN_FCMP(fcmpq_fcc2, float128, 24, 0);
+uint32_t helper_fcmpq(CPUSPARCState *env, Int128 src1, Int128 src2)
+{
+    FloatRelation r = float128_compare_quiet(f128_in(src1), f128_in(src2),
+                                             &env->fp_status);
+    return finish_fcmp(env, r, GETPC());
+}
 
-GEN_FCMP_T(fcmps_fcc3, float32, 26, 0);
-GEN_FCMP_T(fcmpd_fcc3, float64, 26, 0);
-GEN_FCMP(fcmpq_fcc3, float128, 26, 0);
-
-GEN_FCMP_T(fcmpes_fcc1, float32, 22, 1);
-GEN_FCMP_T(fcmped_fcc1, float64, 22, 1);
-GEN_FCMP(fcmpeq_fcc1, float128, 22, 1);
-
-GEN_FCMP_T(fcmpes_fcc2, float32, 24, 1);
-GEN_FCMP_T(fcmped_fcc2, float64, 24, 1);
-GEN_FCMP(fcmpeq_fcc2, float128, 24, 1);
-
-GEN_FCMP_T(fcmpes_fcc3, float32, 26, 1);
-GEN_FCMP_T(fcmped_fcc3, float64, 26, 1);
-GEN_FCMP(fcmpeq_fcc3, float128, 26, 1);
-#endif
-#undef GEN_FCMP_T
-#undef GEN_FCMP
+uint32_t helper_fcmpeq(CPUSPARCState *env, Int128 src1, Int128 src2)
+{
+    FloatRelation r = float128_compare(f128_in(src1), f128_in(src2),
+                                       &env->fp_status);
+    return finish_fcmp(env, r, GETPC());
+}
 
 target_ulong cpu_get_fsr(CPUSPARCState *env)
 {
     target_ulong fsr = env->fsr | env->fsr_cexc_ftt;
 
+    fsr |= env->fcc[0] << FSR_FCC0_SHIFT;
+#ifdef TARGET_SPARC64
+    fsr |= (uint64_t)env->fcc[1] << FSR_FCC1_SHIFT;
+    fsr |= (uint64_t)env->fcc[2] << FSR_FCC2_SHIFT;
+    fsr |= (uint64_t)env->fcc[3] << FSR_FCC3_SHIFT;
+#endif
+
     /* VER is kept completely separate until re-assembly. */
     fsr |= env->def.fpu_version;
 
@@ -465,7 +432,7 @@ static void set_fsr_nonsplit(CPUSPARCState *env, target_ulong fsr)
 {
     int rnd_mode;
 
-    env->fsr = fsr & ~(FSR_VER_MASK | FSR_CEXC_MASK | FSR_FTT_MASK);
+    env->fsr = fsr & (FSR_RD_MASK | FSR_TEM_MASK | FSR_AEXC_MASK);
 
     switch (fsr & FSR_RD_MASK) {
     case FSR_RD_NEAREST:
@@ -488,10 +455,18 @@ static void set_fsr_nonsplit(CPUSPARCState *env, target_ulong fsr)
 void cpu_put_fsr(CPUSPARCState *env, target_ulong fsr)
 {
     env->fsr_cexc_ftt = fsr & (FSR_CEXC_MASK | FSR_FTT_MASK);
+
+    env->fcc[0] = extract32(fsr, FSR_FCC0_SHIFT, 2);
+#ifdef TARGET_SPARC64
+    env->fcc[1] = extract64(fsr, FSR_FCC1_SHIFT, 2);
+    env->fcc[2] = extract64(fsr, FSR_FCC2_SHIFT, 2);
+    env->fcc[3] = extract64(fsr, FSR_FCC3_SHIFT, 2);
+#endif
+
     set_fsr_nonsplit(env, fsr);
 }
 
-void helper_set_fsr_noftt(CPUSPARCState *env, target_ulong fsr)
+void helper_set_fsr_nofcc_noftt(CPUSPARCState *env, uint32_t fsr)
 {
     env->fsr_cexc_ftt &= FSR_FTT_MASK;
     env->fsr_cexc_ftt |= fsr & FSR_CEXC_MASK;
diff --git a/target/sparc/translate.c b/target/sparc/translate.c
index 4eb6291b81..b83b4369fd 100644
--- a/target/sparc/translate.c
+++ b/target/sparc/translate.c
@@ -83,8 +83,6 @@
 # define gen_helper_fxtoq                ({ qemu_build_not_reached(); NULL; })
 # define gen_helper_fxtos                ({ qemu_build_not_reached(); NULL; })
 # define gen_helper_pdist                ({ qemu_build_not_reached(); NULL; })
-# define FSR_LDXFSR_MASK                        0
-# define FSR_LDXFSR_OLDMASK                     0
 # define MAXTL_MASK                             0
 #endif
 
@@ -130,6 +128,7 @@ static TCGv cpu_gsr;
 
 /* Floating point registers */
 static TCGv_i64 cpu_fpr[TARGET_DPREGS];
+static TCGv_i32 cpu_fcc[TARGET_FCCREGS];
 
 #define env_field_offsetof(X)     offsetof(CPUSPARCState, X)
 #ifdef TARGET_SPARC64
@@ -719,159 +718,6 @@ static void gen_op_bshuffle(TCGv_i64 dst, TCGv_i64 src1, TCGv_i64 src2)
 #endif
 }
 
-// 1
-static void gen_op_eval_ba(TCGv dst)
-{
-    tcg_gen_movi_tl(dst, 1);
-}
-
-// 0
-static void gen_op_eval_bn(TCGv dst)
-{
-    tcg_gen_movi_tl(dst, 0);
-}
-
-/*
-  FPSR bit field FCC1 | FCC0:
-   0 =
-   1 <
-   2 >
-   3 unordered
-*/
-static void gen_mov_reg_FCC0(TCGv reg, TCGv src,
-                                    unsigned int fcc_offset)
-{
-    tcg_gen_shri_tl(reg, src, FSR_FCC0_SHIFT + fcc_offset);
-    tcg_gen_andi_tl(reg, reg, 0x1);
-}
-
-static void gen_mov_reg_FCC1(TCGv reg, TCGv src, unsigned int fcc_offset)
-{
-    tcg_gen_shri_tl(reg, src, FSR_FCC1_SHIFT + fcc_offset);
-    tcg_gen_andi_tl(reg, reg, 0x1);
-}
-
-// !0: FCC0 | FCC1
-static void gen_op_eval_fbne(TCGv dst, TCGv src, unsigned int fcc_offset)
-{
-    TCGv t0 = tcg_temp_new();
-    gen_mov_reg_FCC0(dst, src, fcc_offset);
-    gen_mov_reg_FCC1(t0, src, fcc_offset);
-    tcg_gen_or_tl(dst, dst, t0);
-}
-
-// 1 or 2: FCC0 ^ FCC1
-static void gen_op_eval_fblg(TCGv dst, TCGv src, unsigned int fcc_offset)
-{
-    TCGv t0 = tcg_temp_new();
-    gen_mov_reg_FCC0(dst, src, fcc_offset);
-    gen_mov_reg_FCC1(t0, src, fcc_offset);
-    tcg_gen_xor_tl(dst, dst, t0);
-}
-
-// 1 or 3: FCC0
-static void gen_op_eval_fbul(TCGv dst, TCGv src, unsigned int fcc_offset)
-{
-    gen_mov_reg_FCC0(dst, src, fcc_offset);
-}
-
-// 1: FCC0 & !FCC1
-static void gen_op_eval_fbl(TCGv dst, TCGv src, unsigned int fcc_offset)
-{
-    TCGv t0 = tcg_temp_new();
-    gen_mov_reg_FCC0(dst, src, fcc_offset);
-    gen_mov_reg_FCC1(t0, src, fcc_offset);
-    tcg_gen_andc_tl(dst, dst, t0);
-}
-
-// 2 or 3: FCC1
-static void gen_op_eval_fbug(TCGv dst, TCGv src, unsigned int fcc_offset)
-{
-    gen_mov_reg_FCC1(dst, src, fcc_offset);
-}
-
-// 2: !FCC0 & FCC1
-static void gen_op_eval_fbg(TCGv dst, TCGv src, unsigned int fcc_offset)
-{
-    TCGv t0 = tcg_temp_new();
-    gen_mov_reg_FCC0(dst, src, fcc_offset);
-    gen_mov_reg_FCC1(t0, src, fcc_offset);
-    tcg_gen_andc_tl(dst, t0, dst);
-}
-
-// 3: FCC0 & FCC1
-static void gen_op_eval_fbu(TCGv dst, TCGv src, unsigned int fcc_offset)
-{
-    TCGv t0 = tcg_temp_new();
-    gen_mov_reg_FCC0(dst, src, fcc_offset);
-    gen_mov_reg_FCC1(t0, src, fcc_offset);
-    tcg_gen_and_tl(dst, dst, t0);
-}
-
-// 0: !(FCC0 | FCC1)
-static void gen_op_eval_fbe(TCGv dst, TCGv src, unsigned int fcc_offset)
-{
-    TCGv t0 = tcg_temp_new();
-    gen_mov_reg_FCC0(dst, src, fcc_offset);
-    gen_mov_reg_FCC1(t0, src, fcc_offset);
-    tcg_gen_or_tl(dst, dst, t0);
-    tcg_gen_xori_tl(dst, dst, 0x1);
-}
-
-// 0 or 3: !(FCC0 ^ FCC1)
-static void gen_op_eval_fbue(TCGv dst, TCGv src, unsigned int fcc_offset)
-{
-    TCGv t0 = tcg_temp_new();
-    gen_mov_reg_FCC0(dst, src, fcc_offset);
-    gen_mov_reg_FCC1(t0, src, fcc_offset);
-    tcg_gen_xor_tl(dst, dst, t0);
-    tcg_gen_xori_tl(dst, dst, 0x1);
-}
-
-// 0 or 2: !FCC0
-static void gen_op_eval_fbge(TCGv dst, TCGv src, unsigned int fcc_offset)
-{
-    gen_mov_reg_FCC0(dst, src, fcc_offset);
-    tcg_gen_xori_tl(dst, dst, 0x1);
-}
-
-// !1: !(FCC0 & !FCC1)
-static void gen_op_eval_fbuge(TCGv dst, TCGv src, unsigned int fcc_offset)
-{
-    TCGv t0 = tcg_temp_new();
-    gen_mov_reg_FCC0(dst, src, fcc_offset);
-    gen_mov_reg_FCC1(t0, src, fcc_offset);
-    tcg_gen_andc_tl(dst, dst, t0);
-    tcg_gen_xori_tl(dst, dst, 0x1);
-}
-
-// 0 or 1: !FCC1
-static void gen_op_eval_fble(TCGv dst, TCGv src, unsigned int fcc_offset)
-{
-    gen_mov_reg_FCC1(dst, src, fcc_offset);
-    tcg_gen_xori_tl(dst, dst, 0x1);
-}
-
-// !2: !(!FCC0 & FCC1)
-static void gen_op_eval_fbule(TCGv dst, TCGv src, unsigned int fcc_offset)
-{
-    TCGv t0 = tcg_temp_new();
-    gen_mov_reg_FCC0(dst, src, fcc_offset);
-    gen_mov_reg_FCC1(t0, src, fcc_offset);
-    tcg_gen_andc_tl(dst, t0, dst);
-    tcg_gen_xori_tl(dst, dst, 0x1);
-}
-
-// !3: !(FCC0 & FCC1)
-static void gen_op_eval_fbo(TCGv dst, TCGv src, unsigned int fcc_offset)
-{
-    TCGv t0 = tcg_temp_new();
-    gen_mov_reg_FCC0(dst, src, fcc_offset);
-    gen_mov_reg_FCC1(t0, src, fcc_offset);
-    tcg_gen_and_tl(dst, dst, t0);
-    tcg_gen_xori_tl(dst, dst, 0x1);
-}
-
 static void finishing_insn(DisasContext *dc)
 {
     /*
@@ -1096,82 +942,62 @@ static void gen_compare(DisasCompare *cmp, bool xcc, unsigned int cond,
 
 static void gen_fcompare(DisasCompare *cmp, unsigned int cc, unsigned int cond)
 {
-    unsigned int offset;
-    TCGv r_dst, fsr;
+    TCGv_i32 fcc = cpu_fcc[cc];
+    TCGv_i32 c1 = fcc;
+    int c2 = 0;
+    TCGCond tcond;
 
-    /* For now we still generate a straight boolean result.  */
-    cmp->cond = TCG_COND_NE;
-    cmp->c1 = r_dst = tcg_temp_new();
-    cmp->c2 = 0;
-
-    switch (cc) {
-    default:
-    case 0x0:
-        offset = 0;
+    /*
+     * FCC values:
+     * 0 =
+     * 1 <
+     * 2 >
+     * 3 unordered
+     */
+    switch (cond & 7) {
+    case 0x0: /* fbn */
+        tcond = TCG_COND_NEVER;
         break;
-    case 0x1:
-        offset = 32 - 10;
+    case 0x1: /* fbne : !0 */
+        tcond = TCG_COND_NE;
         break;
-    case 0x2:
-        offset = 34 - 10;
+    case 0x2: /* fblg : 1 or 2 */
+        /* fcc in {1,2} - 1 -> fcc in {0,1} */
+        c1 = tcg_temp_new_i32();
+        tcg_gen_addi_i32(c1, fcc, -1);
+        c2 = 1;
+        tcond = TCG_COND_LEU;
         break;
-    case 0x3:
-        offset = 36 - 10;
+    case 0x3: /* fbul : 1 or 3 */
+        c1 = tcg_temp_new_i32();
+        tcg_gen_andi_i32(c1, fcc, 1);
+        tcond = TCG_COND_NE;
+        break;
+    case 0x4: /* fbl  : 1 */
+        c2 = 1;
+        tcond = TCG_COND_EQ;
+        break;
+    case 0x5: /* fbug : 2 or 3 */
+        c2 = 2;
+        tcond = TCG_COND_GEU;
+        break;
+    case 0x6: /* fbg  : 2 */
+        c2 = 2;
+        tcond = TCG_COND_EQ;
+        break;
+    case 0x7: /* fbu  : 3 */
+        c2 = 3;
+        tcond = TCG_COND_EQ;
         break;
     }
-
-    fsr = tcg_temp_new();
-    tcg_gen_ld_tl(fsr, tcg_env, offsetof(CPUSPARCState, fsr));
-    switch (cond) {
-    case 0x0:
-        gen_op_eval_bn(r_dst);
-        break;
-    case 0x1:
-        gen_op_eval_fbne(r_dst, fsr, offset);
-        break;
-    case 0x2:
-        gen_op_eval_fblg(r_dst, fsr, offset);
-        break;
-    case 0x3:
-        gen_op_eval_fbul(r_dst, fsr, offset);
-        break;
-    case 0x4:
-        gen_op_eval_fbl(r_dst, fsr, offset);
-        break;
-    case 0x5:
-        gen_op_eval_fbug(r_dst, fsr, offset);
-        break;
-    case 0x6:
-        gen_op_eval_fbg(r_dst, fsr, offset);
-        break;
-    case 0x7:
-        gen_op_eval_fbu(r_dst, fsr, offset);
-        break;
-    case 0x8:
-        gen_op_eval_ba(r_dst);
-        break;
-    case 0x9:
-        gen_op_eval_fbe(r_dst, fsr, offset);
-        break;
-    case 0xa:
-        gen_op_eval_fbue(r_dst, fsr, offset);
-        break;
-    case 0xb:
-        gen_op_eval_fbge(r_dst, fsr, offset);
-        break;
-    case 0xc:
-        gen_op_eval_fbuge(r_dst, fsr, offset);
-        break;
-    case 0xd:
-        gen_op_eval_fble(r_dst, fsr, offset);
-        break;
-    case 0xe:
-        gen_op_eval_fbule(r_dst, fsr, offset);
-        break;
-    case 0xf:
-        gen_op_eval_fbo(r_dst, fsr, offset);
-        break;
+    if (cond & 8) {
+        tcond = tcg_invert_cond(tcond);
     }
+
+    cmp->cond = tcond;
+    cmp->c2 = c2;
+    cmp->c1 = tcg_temp_new();
+    tcg_gen_extu_i32_tl(cmp->c1, c1);
 }
 
 static bool gen_compare_reg(DisasCompare *cmp, int cond, TCGv r_src)
@@ -1261,148 +1087,6 @@ static void gen_op_fabsq(TCGv_i128 dst, TCGv_i128 src)
     tcg_gen_concat_i64_i128(dst, l, h);
 }
 
-#ifdef TARGET_SPARC64
-static void gen_op_fcmps(int fccno, TCGv_i32 r_rs1, TCGv_i32 r_rs2)
-{
-    switch (fccno) {
-    case 0:
-        gen_helper_fcmps(tcg_env, r_rs1, r_rs2);
-        break;
-    case 1:
-        gen_helper_fcmps_fcc1(tcg_env, r_rs1, r_rs2);
-        break;
-    case 2:
-        gen_helper_fcmps_fcc2(tcg_env, r_rs1, r_rs2);
-        break;
-    case 3:
-        gen_helper_fcmps_fcc3(tcg_env, r_rs1, r_rs2);
-        break;
-    }
-}
-
-static void gen_op_fcmpd(int fccno, TCGv_i64 r_rs1, TCGv_i64 r_rs2)
-{
-    switch (fccno) {
-    case 0:
-        gen_helper_fcmpd(tcg_env, r_rs1, r_rs2);
-        break;
-    case 1:
-        gen_helper_fcmpd_fcc1(tcg_env, r_rs1, r_rs2);
-        break;
-    case 2:
-        gen_helper_fcmpd_fcc2(tcg_env, r_rs1, r_rs2);
-        break;
-    case 3:
-        gen_helper_fcmpd_fcc3(tcg_env, r_rs1, r_rs2);
-        break;
-    }
-}
-
-static void gen_op_fcmpq(int fccno, TCGv_i128 r_rs1, TCGv_i128 r_rs2)
-{
-    switch (fccno) {
-    case 0:
-        gen_helper_fcmpq(tcg_env, r_rs1, r_rs2);
-        break;
-    case 1:
-        gen_helper_fcmpq_fcc1(tcg_env, r_rs1, r_rs2);
-        break;
-    case 2:
-        gen_helper_fcmpq_fcc2(tcg_env, r_rs1, r_rs2);
-        break;
-    case 3:
-        gen_helper_fcmpq_fcc3(tcg_env, r_rs1, r_rs2);
-        break;
-    }
-}
-
-static void gen_op_fcmpes(int fccno, TCGv_i32 r_rs1, TCGv_i32 r_rs2)
-{
-    switch (fccno) {
-    case 0:
-        gen_helper_fcmpes(tcg_env, r_rs1, r_rs2);
-        break;
-    case 1:
-        gen_helper_fcmpes_fcc1(tcg_env, r_rs1, r_rs2);
-        break;
-    case 2:
-        gen_helper_fcmpes_fcc2(tcg_env, r_rs1, r_rs2);
-        break;
-    case 3:
-        gen_helper_fcmpes_fcc3(tcg_env, r_rs1, r_rs2);
-        break;
-    }
-}
-
-static void gen_op_fcmped(int fccno, TCGv_i64 r_rs1, TCGv_i64 r_rs2)
-{
-    switch (fccno) {
-    case 0:
-        gen_helper_fcmped(tcg_env, r_rs1, r_rs2);
-        break;
-    case 1:
-        gen_helper_fcmped_fcc1(tcg_env, r_rs1, r_rs2);
-        break;
-    case 2:
-        gen_helper_fcmped_fcc2(tcg_env, r_rs1, r_rs2);
-        break;
-    case 3:
-        gen_helper_fcmped_fcc3(tcg_env, r_rs1, r_rs2);
-        break;
-    }
-}
-
-static void gen_op_fcmpeq(int fccno, TCGv_i128 r_rs1, TCGv_i128 r_rs2)
-{
-    switch (fccno) {
-    case 0:
-        gen_helper_fcmpeq(tcg_env, r_rs1, r_rs2);
-        break;
-    case 1:
-        gen_helper_fcmpeq_fcc1(tcg_env, r_rs1, r_rs2);
-        break;
-    case 2:
-        gen_helper_fcmpeq_fcc2(tcg_env, r_rs1, r_rs2);
-        break;
-    case 3:
-        gen_helper_fcmpeq_fcc3(tcg_env, r_rs1, r_rs2);
-        break;
-    }
-}
-
-#else
-
-static void gen_op_fcmps(int fccno, TCGv r_rs1, TCGv r_rs2)
-{
-    gen_helper_fcmps(tcg_env, r_rs1, r_rs2);
-}
-
-static void gen_op_fcmpd(int fccno, TCGv_i64 r_rs1, TCGv_i64 r_rs2)
-{
-    gen_helper_fcmpd(tcg_env, r_rs1, r_rs2);
-}
-
-static void gen_op_fcmpq(int fccno, TCGv_i128 r_rs1, TCGv_i128 r_rs2)
-{
-    gen_helper_fcmpq(tcg_env, r_rs1, r_rs2);
-}
-
-static void gen_op_fcmpes(int fccno, TCGv r_rs1, TCGv r_rs2)
-{
-    gen_helper_fcmpes(tcg_env, r_rs1, r_rs2);
-}
-
-static void gen_op_fcmped(int fccno, TCGv_i64 r_rs1, TCGv_i64 r_rs2)
-{
-    gen_helper_fcmped(tcg_env, r_rs1, r_rs2);
-}
-
-static void gen_op_fcmpeq(int fccno, TCGv_i128 r_rs1, TCGv_i128 r_rs2)
-{
-    gen_helper_fcmpeq(tcg_env, r_rs1, r_rs2);
-}
-#endif
-
 static void gen_op_fpexception_im(DisasContext *dc, int ftt)
 {
     /*
@@ -4400,11 +4084,10 @@ static bool trans_STDFQ(DisasContext *dc, arg_STDFQ *a)
     return true;
 }
 
-static bool do_ldfsr(DisasContext *dc, arg_r_r_ri *a, MemOp mop,
-                     target_ulong new_mask, target_ulong old_mask)
+static bool trans_LDFSR(DisasContext *dc, arg_r_r_ri *a)
 {
     TCGv addr = gen_ldst_addr(dc, a->rs1, a->imm, a->rs2_or_imm);
-    TCGv tnew, told;
+    TCGv_i32 tmp;
 
     if (addr == NULL) {
         return false;
@@ -4412,19 +4095,48 @@ static bool do_ldfsr(DisasContext *dc, arg_r_r_ri *a, MemOp mop,
     if (gen_trap_ifnofpu(dc)) {
         return true;
     }
-    tnew = tcg_temp_new();
-    told = tcg_temp_new();
-    tcg_gen_qemu_ld_tl(tnew, addr, dc->mem_idx, mop | MO_ALIGN);
-    tcg_gen_ld_tl(told, tcg_env, offsetof(CPUSPARCState, fsr));
-    tcg_gen_andi_tl(tnew, tnew, new_mask);
-    tcg_gen_andi_tl(told, told, old_mask);
-    tcg_gen_or_tl(tnew, tnew, told);
-    gen_helper_set_fsr_noftt(tcg_env, tnew);
+
+    tmp = tcg_temp_new_i32();
+    tcg_gen_qemu_ld_i32(tmp, addr, dc->mem_idx, MO_TEUL | MO_ALIGN);
+
+    tcg_gen_extract_i32(cpu_fcc[0], tmp, FSR_FCC0_SHIFT, 2);
+    /* LDFSR does not change FCC[1-3]. */
+
+    gen_helper_set_fsr_nofcc_noftt(tcg_env, tmp);
     return advance_pc(dc);
 }
 
-TRANS(LDFSR, ALL, do_ldfsr, a, MO_TEUL, FSR_LDFSR_MASK, FSR_LDFSR_OLDMASK)
-TRANS(LDXFSR, 64, do_ldfsr, a, MO_TEUQ, FSR_LDXFSR_MASK, FSR_LDXFSR_OLDMASK)
+static bool trans_LDXFSR(DisasContext *dc, arg_r_r_ri *a)
+{
+#ifdef TARGET_SPARC64
+    TCGv addr = gen_ldst_addr(dc, a->rs1, a->imm, a->rs2_or_imm);
+    TCGv_i64 t64;
+    TCGv_i32 lo, hi;
+
+    if (addr == NULL) {
+        return false;
+    }
+    if (gen_trap_ifnofpu(dc)) {
+        return true;
+    }
+
+    t64 = tcg_temp_new_i64();
+    tcg_gen_qemu_ld_i64(t64, addr, dc->mem_idx, MO_TEUQ | MO_ALIGN);
+
+    lo = tcg_temp_new_i32();
+    hi = cpu_fcc[3];
+    tcg_gen_extr_i64_i32(lo, hi, t64);
+    tcg_gen_extract_i32(cpu_fcc[0], lo, FSR_FCC0_SHIFT, 2);
+    tcg_gen_extract_i32(cpu_fcc[1], hi, FSR_FCC1_SHIFT - 32, 2);
+    tcg_gen_extract_i32(cpu_fcc[2], hi, FSR_FCC2_SHIFT - 32, 2);
+    tcg_gen_extract_i32(cpu_fcc[3], hi, FSR_FCC3_SHIFT - 32, 2);
+
+    gen_helper_set_fsr_nofcc_noftt(tcg_env, lo);
+    return advance_pc(dc);
+#else
+    return false;
+#endif
+}
 
 static bool do_stfsr(DisasContext *dc, arg_r_r_ri *a, MemOp mop)
 {
@@ -5075,9 +4787,9 @@ static bool do_fcmps(DisasContext *dc, arg_FCMPs *a, bool e)
     src1 = gen_load_fpr_F(dc, a->rs1);
     src2 = gen_load_fpr_F(dc, a->rs2);
     if (e) {
-        gen_op_fcmpes(a->cc, src1, src2);
+        gen_helper_fcmpes(cpu_fcc[a->cc], tcg_env, src1, src2);
     } else {
-        gen_op_fcmps(a->cc, src1, src2);
+        gen_helper_fcmps(cpu_fcc[a->cc], tcg_env, src1, src2);
     }
     return advance_pc(dc);
 }
@@ -5099,9 +4811,9 @@ static bool do_fcmpd(DisasContext *dc, arg_FCMPd *a, bool e)
     src1 = gen_load_fpr_D(dc, a->rs1);
     src2 = gen_load_fpr_D(dc, a->rs2);
     if (e) {
-        gen_op_fcmped(a->cc, src1, src2);
+        gen_helper_fcmped(cpu_fcc[a->cc], tcg_env, src1, src2);
     } else {
-        gen_op_fcmpd(a->cc, src1, src2);
+        gen_helper_fcmpd(cpu_fcc[a->cc], tcg_env, src1, src2);
     }
     return advance_pc(dc);
 }
@@ -5126,9 +4838,9 @@ static bool do_fcmpq(DisasContext *dc, arg_FCMPq *a, bool e)
     src1 = gen_load_fpr_Q(dc, a->rs1);
     src2 = gen_load_fpr_Q(dc, a->rs2);
     if (e) {
-        gen_op_fcmpeq(a->cc, src1, src2);
+        gen_helper_fcmpeq(cpu_fcc[a->cc], tcg_env, src1, src2);
     } else {
-        gen_op_fcmpq(a->cc, src1, src2);
+        gen_helper_fcmpq(cpu_fcc[a->cc], tcg_env, src1, src2);
     }
     return advance_pc(dc);
 }
@@ -5334,6 +5046,18 @@ void sparc_tcg_init(void)
         "f48", "f50", "f52", "f54", "f56", "f58", "f60", "f62",
     };
 
+    static const struct { TCGv_i32 *ptr; int off; const char *name; } r32[] = {
+#ifdef TARGET_SPARC64
+        { &cpu_fprs, offsetof(CPUSPARCState, fprs), "fprs" },
+        { &cpu_fcc[0], offsetof(CPUSPARCState, fcc[0]), "fcc0" },
+        { &cpu_fcc[1], offsetof(CPUSPARCState, fcc[1]), "fcc1" },
+        { &cpu_fcc[2], offsetof(CPUSPARCState, fcc[2]), "fcc2" },
+        { &cpu_fcc[3], offsetof(CPUSPARCState, fcc[3]), "fcc3" },
+#else
+        { &cpu_fcc[0], offsetof(CPUSPARCState, fcc[0]), "fcc" },
+#endif
+    };
+
     static const struct { TCGv *ptr; int off; const char *name; } rtl[] = {
 #ifdef TARGET_SPARC64
         { &cpu_gsr, offsetof(CPUSPARCState, gsr), "gsr" },
@@ -5357,6 +5081,10 @@ void sparc_tcg_init(void)
                                          offsetof(CPUSPARCState, regwptr),
                                          "regwptr");
 
+    for (i = 0; i < ARRAY_SIZE(r32); ++i) {
+        *r32[i].ptr = tcg_global_mem_new_i32(tcg_env, r32[i].off, r32[i].name);
+    }
+
     for (i = 0; i < ARRAY_SIZE(rtl); ++i) {
         *rtl[i].ptr = tcg_global_mem_new(tcg_env, rtl[i].off, rtl[i].name);
     }
@@ -5379,11 +5107,6 @@ void sparc_tcg_init(void)
                                             offsetof(CPUSPARCState, fpr[i]),
                                             fregnames[i]);
     }
-
-#ifdef TARGET_SPARC64
-    cpu_fprs = tcg_global_mem_new_i32(tcg_env,
-                                      offsetof(CPUSPARCState, fprs), "fprs");
-#endif
 }
 
 void sparc_restore_state_to_opc(CPUState *cs,
-- 
2.34.1

[PATCH 22/22] target/sparc: Remove FSR_FTT_NMASK, FSR_FTT_CEXC_NMASK

[Richard Henderson]

These macros are no longer used.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/sparc/cpu.h | 7 -------
 1 file changed, 7 deletions(-)

diff --git a/target/sparc/cpu.h b/target/sparc/cpu.h
index 30bab9a7b3..9ca9fb30ff 100644
--- a/target/sparc/cpu.h
+++ b/target/sparc/cpu.h
@@ -202,13 +202,6 @@ enum {
 #define FSR_FTT1   (1ULL << 15)
 #define FSR_FTT0   (1ULL << 14)
 #define FSR_FTT_MASK (FSR_FTT2 | FSR_FTT1 | FSR_FTT0)
-#ifdef TARGET_SPARC64
-#define FSR_FTT_NMASK      0xfffffffffffe3fffULL
-#define FSR_FTT_CEXC_NMASK 0xfffffffffffe3fe0ULL
-#else
-#define FSR_FTT_NMASK      0xfffe3fffULL
-#define FSR_FTT_CEXC_NMASK 0xfffe3fe0ULL
-#endif
 #define FSR_FTT_IEEE_EXCP (1ULL << 14)
 #define FSR_FTT_UNIMPFPOP (3ULL << 14)
 #define FSR_FTT_SEQ_ERROR (4ULL << 14)
-- 
2.34.1

Re: [PATCH 00/22] target/sparc: floating-point cleanup

[Richard Henderson]

On 11/4/23 03:38, Richard Henderson wrote:
> Major changes:
> 
> (1) Get rid of the env->qt[01] temporaries and use TCGv_i128 for float128.
> (2) Perform ieee exception check within the helpers, before any writeback
>      to the floating point registers.
> (3) Split env->fsr into pieces to simplify update, especially compares.
> 
> 
> r~
> 
> 
> Based-on: 20231101041132.174501-1-richard.henderson@linaro.org
> ("[PATCH v2 00/21] target/sparc: Cleanup condition codes etc")

Ping.

Prerequisites are upstream, and it rebases cleanly on master.
For reference,

   https://gitlab.com/rth7680/qemu/-/commits/tgt-sparc-fp


r~

> 
> 
> Richard Henderson (22):
>    target/sparc: Use tcg_gen_qemu_{ld,st}_i128 for ASI_M_BCOPY
>    target/sparc: Use tcg_gen_qemu_{ld,st}_i128 for ASI_M_BFILL
>    target/sparc: Remove gen_dest_fpr_F
>    target/sparc: Introduce gen_{load,store}_fpr_Q
>    target/sparc: Inline FNEG, FABS
>    target/sparc: Use i128 for FSQRTq
>    target/sparc: Use i128 for FADDq, FSUBq, FMULq, FDIVq
>    target/sparc: Use i128 for FqTOs, FqTOi
>    target/sparc: Use i128 for FqTOd, FqTOx
>    target/sparc: Use i128 for FCMPq, FCMPEq
>    target/sparc: Use i128 for FsTOq, FiTOq
>    target/sparc: Use i128 for FdTOq, FxTOq
>    target/sparc: Use i128 for Fdmulq
>    target/sparc: Remove qt0, qt1 temporaries
>    target/sparc: Introduce cpu_get_fsr, cpu_put_fsr
>    target/split: Split ver from env->fsr
>    target/sparc: Clear cexc and ftt in do_check_ieee_exceptions
>    target/sparc: Merge check_ieee_exceptions with FPop helpers
>    target/sparc: Split cexc and ftt from env->fsr
>    target/sparc: Remove cpu_fsr
>    target/sparc: Split fcc out of env->fsr
>    target/sparc: Remove FSR_FTT_NMASK, FSR_FTT_CEXC_NMASK
> 
>   target/sparc/cpu.h          |  39 +-
>   target/sparc/helper.h       | 116 ++----
>   linux-user/sparc/cpu_loop.c |   2 +-
>   linux-user/sparc/signal.c   |  14 +-
>   target/sparc/cpu.c          |  32 +-
>   target/sparc/fop_helper.c   | 510 +++++++++++++----------
>   target/sparc/gdbstub.c      |   8 +-
>   target/sparc/ldst_helper.c  |   3 -
>   target/sparc/machine.c      |  38 +-
>   target/sparc/translate.c    | 799 ++++++++++++------------------------
>   10 files changed, 680 insertions(+), 881 deletions(-)
> 

Re: [PATCH 00/22] target/sparc: floating-point cleanup

[Mark Cave-Ayland]

On 28/01/2024 06:49, Richard Henderson wrote:

> On 11/4/23 03:38, Richard Henderson wrote:
>> Major changes:
>>
>> (1) Get rid of the env->qt[01] temporaries and use TCGv_i128 for float128.
>> (2) Perform ieee exception check within the helpers, before any writeback
>>      to the floating point registers.
>> (3) Split env->fsr into pieces to simplify update, especially compares.
>>
>>
>> r~
>>
>>
>> Based-on: 20231101041132.174501-1-richard.henderson@linaro.org
>> ("[PATCH v2 00/21] target/sparc: Cleanup condition codes etc")
> 
> Ping.
> 
> Prerequisites are upstream, and it rebases cleanly on master.
> For reference,
> 
>    https://gitlab.com/rth7680/qemu/-/commits/tgt-sparc-fp
>  
> r~

Oops looks like I forgot about this series. I'm not sure I have any images that 
explicitly test floating point operations, but I can at least run it through my 
OpenBIOS test images and check for regressions over the next couple of days.

>> Richard Henderson (22):
>>    target/sparc: Use tcg_gen_qemu_{ld,st}_i128 for ASI_M_BCOPY
>>    target/sparc: Use tcg_gen_qemu_{ld,st}_i128 for ASI_M_BFILL
>>    target/sparc: Remove gen_dest_fpr_F
>>    target/sparc: Introduce gen_{load,store}_fpr_Q
>>    target/sparc: Inline FNEG, FABS
>>    target/sparc: Use i128 for FSQRTq
>>    target/sparc: Use i128 for FADDq, FSUBq, FMULq, FDIVq
>>    target/sparc: Use i128 for FqTOs, FqTOi
>>    target/sparc: Use i128 for FqTOd, FqTOx
>>    target/sparc: Use i128 for FCMPq, FCMPEq
>>    target/sparc: Use i128 for FsTOq, FiTOq
>>    target/sparc: Use i128 for FdTOq, FxTOq
>>    target/sparc: Use i128 for Fdmulq
>>    target/sparc: Remove qt0, qt1 temporaries
>>    target/sparc: Introduce cpu_get_fsr, cpu_put_fsr
>>    target/split: Split ver from env->fsr
>>    target/sparc: Clear cexc and ftt in do_check_ieee_exceptions
>>    target/sparc: Merge check_ieee_exceptions with FPop helpers
>>    target/sparc: Split cexc and ftt from env->fsr
>>    target/sparc: Remove cpu_fsr
>>    target/sparc: Split fcc out of env->fsr
>>    target/sparc: Remove FSR_FTT_NMASK, FSR_FTT_CEXC_NMASK
>>
>>   target/sparc/cpu.h          |  39 +-
>>   target/sparc/helper.h       | 116 ++----
>>   linux-user/sparc/cpu_loop.c |   2 +-
>>   linux-user/sparc/signal.c   |  14 +-
>>   target/sparc/cpu.c          |  32 +-
>>   target/sparc/fop_helper.c   | 510 +++++++++++++----------
>>   target/sparc/gdbstub.c      |   8 +-
>>   target/sparc/ldst_helper.c  |   3 -
>>   target/sparc/machine.c      |  38 +-
>>   target/sparc/translate.c    | 799 ++++++++++++------------------------
>>   10 files changed, 680 insertions(+), 881 deletions(-)


ATB,

Mark.

Re: [PATCH 00/22] target/sparc: floating-point cleanup

[Mark Cave-Ayland]

On 28/01/2024 06:49, Richard Henderson wrote:

> On 11/4/23 03:38, Richard Henderson wrote:
>> Major changes:
>>
>> (1) Get rid of the env->qt[01] temporaries and use TCGv_i128 for float128.
>> (2) Perform ieee exception check within the helpers, before any writeback
>>      to the floating point registers.
>> (3) Split env->fsr into pieces to simplify update, especially compares.
>>
>>
>> r~
>>
>>
>> Based-on: 20231101041132.174501-1-richard.henderson@linaro.org
>> ("[PATCH v2 00/21] target/sparc: Cleanup condition codes etc")
> 
> Ping.
> 
> Prerequisites are upstream, and it rebases cleanly on master.
> For reference,
> 
>    https://gitlab.com/rth7680/qemu/-/commits/tgt-sparc-fp
> 
> 
> r~

I've tested the above branch on my SPARC32 and SPARC64 images, and whilst I don't 
think they particularly exercise FP instructions, I don't see any regressions so:

Tested-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Acked-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>

I'm happy for you to take this via tcg-next if that's easiest for you.

>> Richard Henderson (22):
>>    target/sparc: Use tcg_gen_qemu_{ld,st}_i128 for ASI_M_BCOPY
>>    target/sparc: Use tcg_gen_qemu_{ld,st}_i128 for ASI_M_BFILL
>>    target/sparc: Remove gen_dest_fpr_F
>>    target/sparc: Introduce gen_{load,store}_fpr_Q
>>    target/sparc: Inline FNEG, FABS
>>    target/sparc: Use i128 for FSQRTq
>>    target/sparc: Use i128 for FADDq, FSUBq, FMULq, FDIVq
>>    target/sparc: Use i128 for FqTOs, FqTOi
>>    target/sparc: Use i128 for FqTOd, FqTOx
>>    target/sparc: Use i128 for FCMPq, FCMPEq
>>    target/sparc: Use i128 for FsTOq, FiTOq
>>    target/sparc: Use i128 for FdTOq, FxTOq
>>    target/sparc: Use i128 for Fdmulq
>>    target/sparc: Remove qt0, qt1 temporaries
>>    target/sparc: Introduce cpu_get_fsr, cpu_put_fsr
>>    target/split: Split ver from env->fsr
>>    target/sparc: Clear cexc and ftt in do_check_ieee_exceptions
>>    target/sparc: Merge check_ieee_exceptions with FPop helpers
>>    target/sparc: Split cexc and ftt from env->fsr
>>    target/sparc: Remove cpu_fsr
>>    target/sparc: Split fcc out of env->fsr
>>    target/sparc: Remove FSR_FTT_NMASK, FSR_FTT_CEXC_NMASK
>>
>>   target/sparc/cpu.h          |  39 +-
>>   target/sparc/helper.h       | 116 ++----
>>   linux-user/sparc/cpu_loop.c |   2 +-
>>   linux-user/sparc/signal.c   |  14 +-
>>   target/sparc/cpu.c          |  32 +-
>>   target/sparc/fop_helper.c   | 510 +++++++++++++----------
>>   target/sparc/gdbstub.c      |   8 +-
>>   target/sparc/ldst_helper.c  |   3 -
>>   target/sparc/machine.c      |  38 +-
>>   target/sparc/translate.c    | 799 ++++++++++++------------------------
>>   10 files changed, 680 insertions(+), 881 deletions(-)


ATB,

Mark.

Re: [PATCH 00/22] target/sparc: floating-point cleanup

[Richard Henderson]

On 2/1/24 07:49, Mark Cave-Ayland wrote:
> I'm happy for you to take this via tcg-next if that's easiest for you.

Yes, I can do that.


r~