[Qemu-devel] [PATCH 0/3] target/arm: Implement ARMv8.3-JSConv

[Qemu-devel] [PATCH 0/3] target/arm: Implement ARMv8.3-JSConv

[Richard Henderson]

Two cleanups to surrounding code preceed the actual implementation.

I did not find the ARM FPToFixedJS pseudo-function particularly helpful,
so I mostly cribed off of the Alpha cvttq implementation.  But I think
it all makes sense -- convert, produce the mod 2^32 result with NaN
getting INT32_MAX, set ZF iff the conversion is exact.

I've tested the aa64 version vs FVP with RISU.  I need to adjust my
FVP configuration in order to test aa32.


r~


Richard Henderson (3):
  target/arm: Force result size into dp after operation
  target/arm: Restructure disas_fp_int_conv
  target/arm: Implement ARMv8.3-JSConv

 target/arm/cpu.h           |  10 ++++
 target/arm/helper.h        |   2 +
 target/arm/cpu.c           |   1 +
 target/arm/cpu64.c         |   2 +
 target/arm/op_helper.c     |  91 ++++++++++++++++++++++++++++
 target/arm/translate-a64.c | 120 +++++++++++++++++++++++--------------
 target/arm/translate.c     |  47 ++++++++++-----
 7 files changed, 212 insertions(+), 61 deletions(-)

-- 
2.17.2

[Qemu-devel] [PATCH 1/3] target/arm: Force result size into dp after operation

[Richard Henderson]

Rather than a complex set of cases testing for writeback,
adjust DP after performing the operation.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/arm/translate.c | 32 ++++++++++++++++----------------
 1 file changed, 16 insertions(+), 16 deletions(-)

diff --git a/target/arm/translate.c b/target/arm/translate.c
index 66cf28c8cb..eb25895876 100644
--- a/target/arm/translate.c
+++ b/target/arm/translate.c
@@ -3970,6 +3970,7 @@ static int disas_vfp_insn(DisasContext *s, uint32_t insn)
                         tcg_gen_or_i32(tmp, tmp, tmp2);
                         tcg_temp_free_i32(tmp2);
                         gen_vfp_msr(tmp);
+                        dp = 0; /* always a single precision result */
                         break;
                     }
                     case 7: /* vcvtt.f16.f32, vcvtt.f16.f64 */
@@ -3993,20 +3994,25 @@ static int disas_vfp_insn(DisasContext *s, uint32_t insn)
                         tcg_gen_or_i32(tmp, tmp, tmp2);
                         tcg_temp_free_i32(tmp2);
                         gen_vfp_msr(tmp);
+                        dp = 0; /* always a single precision result */
                         break;
                     }
                     case 8: /* cmp */
                         gen_vfp_cmp(dp);
+                        dp = -1; /* no write back */
                         break;
                     case 9: /* cmpe */
                         gen_vfp_cmpe(dp);
+                        dp = -1; /* no write back */
                         break;
                     case 10: /* cmpz */
                         gen_vfp_cmp(dp);
+                        dp = -1; /* no write back */
                         break;
                     case 11: /* cmpez */
                         gen_vfp_F1_ld0(dp);
                         gen_vfp_cmpe(dp);
+                        dp = -1; /* no write back */
                         break;
                     case 12: /* vrintr */
                     {
@@ -4047,10 +4053,12 @@ static int disas_vfp_insn(DisasContext *s, uint32_t insn)
                         break;
                     }
                     case 15: /* single<->double conversion */
-                        if (dp)
+                        if (dp) {
                             gen_helper_vfp_fcvtsd(cpu_F0s, cpu_F0d, cpu_env);
-                        else
+                        } else {
                             gen_helper_vfp_fcvtds(cpu_F0d, cpu_F0s, cpu_env);
+                        }
+                        dp = !dp; /* result size is opposite */
                         break;
                     case 16: /* fuito */
                         gen_vfp_uito(dp, 0);
@@ -4084,15 +4092,19 @@ static int disas_vfp_insn(DisasContext *s, uint32_t insn)
                         break;
                     case 24: /* ftoui */
                         gen_vfp_toui(dp, 0);
+                        dp = 0; /* always an integer result */
                         break;
                     case 25: /* ftouiz */
                         gen_vfp_touiz(dp, 0);
+                        dp = 0; /* always an integer result */
                         break;
                     case 26: /* ftosi */
                         gen_vfp_tosi(dp, 0);
+                        dp = 0; /* always an integer result */
                         break;
                     case 27: /* ftosiz */
                         gen_vfp_tosiz(dp, 0);
+                        dp = 0; /* always an integer result */
                         break;
                     case 28: /* ftosh */
                         if (!arm_dc_feature(s, ARM_FEATURE_VFP3)) {
@@ -4126,20 +4138,8 @@ static int disas_vfp_insn(DisasContext *s, uint32_t insn)
                     return 1;
                 }
 
-                /* Write back the result.  */
-                if (op == 15 && (rn >= 8 && rn <= 11)) {
-                    /* Comparison, do nothing.  */
-                } else if (op == 15 && dp && ((rn & 0x1c) == 0x18 ||
-                                              (rn & 0x1e) == 0x6)) {
-                    /* VCVT double to int: always integer result.
-                     * VCVT double to half precision is always a single
-                     * precision result.
-                     */
-                    gen_mov_vreg_F0(0, rd);
-                } else if (op == 15 && rn == 15) {
-                    /* conversion */
-                    gen_mov_vreg_F0(!dp, rd);
-                } else {
+                /* Write back the result, if any.  */
+                if (dp >= 0) {
                     gen_mov_vreg_F0(dp, rd);
                 }
 
-- 
2.17.2

Re: [Qemu-devel] [PATCH 1/3] target/arm: Force result size into dp after operation

[Peter Maydell]

On Mon, 4 Feb 2019 at 05:27, Richard Henderson
<richard.henderson@linaro.org> wrote:
>
> Rather than a complex set of cases testing for writeback,
> adjust DP after performing the operation.
>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>

thanks
-- PMM

[Qemu-devel] [PATCH 2/3] target/arm: Restructure disas_fp_int_conv

[Richard Henderson]

For opcodes 0-5, move some if conditions into the structure
of a switch statement.  For opcodes 6 & 7, decode everything
at once with a second switch.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/arm/translate-a64.c | 94 ++++++++++++++++++++------------------
 1 file changed, 49 insertions(+), 45 deletions(-)

diff --git a/target/arm/translate-a64.c b/target/arm/translate-a64.c
index a1997e3ae2..90298ac562 100644
--- a/target/arm/translate-a64.c
+++ b/target/arm/translate-a64.c
@@ -6473,68 +6473,72 @@ static void disas_fp_int_conv(DisasContext *s, uint32_t insn)
     int type = extract32(insn, 22, 2);
     bool sbit = extract32(insn, 29, 1);
     bool sf = extract32(insn, 31, 1);
+    bool itof = false;
 
     if (sbit) {
-        unallocated_encoding(s);
-        return;
+        goto do_unallocated;
     }
 
-    if (opcode > 5) {
-        /* FMOV */
-        bool itof = opcode & 1;
-
-        if (rmode >= 2) {
-            unallocated_encoding(s);
-            return;
-        }
-
-        switch (sf << 3 | type << 1 | rmode) {
-        case 0x0: /* 32 bit */
-        case 0xa: /* 64 bit */
-        case 0xd: /* 64 bit to top half of quad */
-            break;
-        case 0x6: /* 16-bit float, 32-bit int */
-        case 0xe: /* 16-bit float, 64-bit int */
-            if (dc_isar_feature(aa64_fp16, s)) {
-                break;
-            }
-            /* fallthru */
-        default:
-            /* all other sf/type/rmode combinations are invalid */
-            unallocated_encoding(s);
-            return;
-        }
-
-        if (!fp_access_check(s)) {
-            return;
-        }
-        handle_fmov(s, rd, rn, type, itof);
-    } else {
-        /* actual FP conversions */
-        bool itof = extract32(opcode, 1, 1);
-
-        if (rmode != 0 && opcode > 1) {
-            unallocated_encoding(s);
-            return;
+    switch (opcode) {
+    case 2: /* SCVTF */
+    case 3: /* UCVTF */
+        itof = true;
+        /* fallthru */
+    case 4: /* FCVTAS */
+    case 5: /* FCVTAU */
+        if (rmode != 0) {
+            goto do_unallocated;
         }
+        /* fallthru */
+    case 0: /* FCVT[NPMZ]S */
+    case 1: /* FCVT[NPMZ]U */
         switch (type) {
         case 0: /* float32 */
         case 1: /* float64 */
             break;
         case 3: /* float16 */
-            if (dc_isar_feature(aa64_fp16, s)) {
-                break;
+            if (!dc_isar_feature(aa64_fp16, s)) {
+                goto do_unallocated;
             }
-            /* fallthru */
+            break;
         default:
-            unallocated_encoding(s);
-            return;
+            goto do_unallocated;
         }
-
         if (!fp_access_check(s)) {
             return;
         }
         handle_fpfpcvt(s, rd, rn, opcode, itof, rmode, 64, sf, type);
+        break;
+
+    default:
+        switch (sf << 6 | type << 5 | rmode << 3 | opcode) {
+        case 0b01100110: /* FMOV half <-> 32-bit int */
+        case 0b01100111:
+        case 0b11100110: /* FMOV half <-> 64-bit int */
+        case 0b11100111:
+            if (!dc_isar_feature(aa64_fp16, s)) {
+                goto do_unallocated;
+            }
+            /* fallthru */
+        case 0b00000110: /* FMOV 32-bit */
+        case 0b00000111:
+        case 0b10100110: /* FMOV 64-bit */
+        case 0b10100111:
+        case 0b11001110: /* FMOV top half of 128-bit */
+        case 0b11001111:
+            if (!fp_access_check(s)) {
+                return;
+            }
+            itof = opcode & 1;
+            handle_fmov(s, rd, rn, type, itof);
+            break;
+
+        default:
+        do_unallocated:
+            unallocated_encoding(s);
+            return;
+        }
+        break;
     }
 }
 
-- 
2.17.2

Re: [Qemu-devel] [PATCH 2/3] target/arm: Restructure disas_fp_int_conv

[Richard Henderson]

On 2/4/19 5:27 AM, Richard Henderson wrote:
> +    default:
> +        switch (sf << 6 | type << 5 | rmode << 3 | opcode) {
> +        case 0b01100110: /* FMOV half <-> 32-bit int */
> +        case 0b01100111:
> +        case 0b11100110: /* FMOV half <-> 64-bit int */
> +        case 0b11100111:
> +            if (!dc_isar_feature(aa64_fp16, s)) {
> +                goto do_unallocated;
> +            }
> +            /* fallthru */
> +        case 0b00000110: /* FMOV 32-bit */
> +        case 0b00000111:
> +        case 0b10100110: /* FMOV 64-bit */
> +        case 0b10100111:
> +        case 0b11001110: /* FMOV top half of 128-bit */
> +        case 0b11001111:

There's an error in here that a full risu run has diagnosed.


r~

[Qemu-devel] [PATCH 3/3] target/arm: Implement ARMv8.3-JSConv

[Richard Henderson]

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/arm/cpu.h           | 10 +++++
 target/arm/helper.h        |  2 +
 target/arm/cpu.c           |  1 +
 target/arm/cpu64.c         |  2 +
 target/arm/op_helper.c     | 91 ++++++++++++++++++++++++++++++++++++++
 target/arm/translate-a64.c | 26 +++++++++++
 target/arm/translate.c     | 15 +++++++
 7 files changed, 147 insertions(+)

diff --git a/target/arm/cpu.h b/target/arm/cpu.h
index a68bcc9fed..d2c2e2b0cf 100644
--- a/target/arm/cpu.h
+++ b/target/arm/cpu.h
@@ -3209,6 +3209,11 @@ static inline bool isar_feature_aa32_vcma(const ARMISARegisters *id)
     return FIELD_EX32(id->id_isar5, ID_ISAR5, VCMA) != 0;
 }
 
+static inline bool isar_feature_aa32_jscvt(const ARMISARegisters *id)
+{
+    return FIELD_EX32(id->id_isar6, ID_ISAR6, JSCVT) != 0;
+}
+
 static inline bool isar_feature_aa32_dp(const ARMISARegisters *id)
 {
     return FIELD_EX32(id->id_isar6, ID_ISAR6, DP) != 0;
@@ -3287,6 +3292,11 @@ static inline bool isar_feature_aa64_dp(const ARMISARegisters *id)
     return FIELD_EX64(id->id_aa64isar0, ID_AA64ISAR0, DP) != 0;
 }
 
+static inline bool isar_feature_aa64_jscvt(const ARMISARegisters *id)
+{
+    return FIELD_EX64(id->id_aa64isar1, ID_AA64ISAR1, JSCVT) != 0;
+}
+
 static inline bool isar_feature_aa64_fcma(const ARMISARegisters *id)
 {
     return FIELD_EX64(id->id_aa64isar1, ID_AA64ISAR1, FCMA) != 0;
diff --git a/target/arm/helper.h b/target/arm/helper.h
index 53a38188c6..6998f7e8d5 100644
--- a/target/arm/helper.h
+++ b/target/arm/helper.h
@@ -218,6 +218,8 @@ DEF_HELPER_FLAGS_2(rintd_exact, TCG_CALL_NO_RWG, f64, f64, ptr)
 DEF_HELPER_FLAGS_2(rints, TCG_CALL_NO_RWG, f32, f32, ptr)
 DEF_HELPER_FLAGS_2(rintd, TCG_CALL_NO_RWG, f64, f64, ptr)
 
+DEF_HELPER_FLAGS_2(fjcvtzs, TCG_CALL_NO_RWG, i64, f64, ptr)
+
 /* neon_helper.c */
 DEF_HELPER_FLAGS_3(neon_qadd_u8, TCG_CALL_NO_RWG, i32, env, i32, i32)
 DEF_HELPER_FLAGS_3(neon_qadd_s8, TCG_CALL_NO_RWG, i32, env, i32, i32)
diff --git a/target/arm/cpu.c b/target/arm/cpu.c
index 3874dc9875..2eb2ce6c8c 100644
--- a/target/arm/cpu.c
+++ b/target/arm/cpu.c
@@ -1995,6 +1995,7 @@ static void arm_max_initfn(Object *obj)
             cpu->isar.id_isar5 = t;
 
             t = cpu->isar.id_isar6;
+            t = FIELD_DP32(t, ID_ISAR6, JSCVT, 1);
             t = FIELD_DP32(t, ID_ISAR6, DP, 1);
             cpu->isar.id_isar6 = t;
 
diff --git a/target/arm/cpu64.c b/target/arm/cpu64.c
index 7107ec8d7e..43d8ff047c 100644
--- a/target/arm/cpu64.c
+++ b/target/arm/cpu64.c
@@ -311,6 +311,7 @@ static void aarch64_max_initfn(Object *obj)
         cpu->isar.id_aa64isar0 = t;
 
         t = cpu->isar.id_aa64isar1;
+        t = FIELD_DP64(t, ID_AA64ISAR1, JSCVT, 1);
         t = FIELD_DP64(t, ID_AA64ISAR1, FCMA, 1);
         t = FIELD_DP64(t, ID_AA64ISAR1, APA, 1); /* PAuth, architected only */
         t = FIELD_DP64(t, ID_AA64ISAR1, API, 0);
@@ -340,6 +341,7 @@ static void aarch64_max_initfn(Object *obj)
         cpu->isar.id_isar5 = u;
 
         u = cpu->isar.id_isar6;
+        u = FIELD_DP32(u, ID_ISAR6, JSCVT, 1);
         u = FIELD_DP32(u, ID_ISAR6, DP, 1);
         cpu->isar.id_isar6 = u;
 
diff --git a/target/arm/op_helper.c b/target/arm/op_helper.c
index c998eadfaa..a7259a7194 100644
--- a/target/arm/op_helper.c
+++ b/target/arm/op_helper.c
@@ -24,6 +24,7 @@
 #include "internals.h"
 #include "exec/exec-all.h"
 #include "exec/cpu_ldst.h"
+#include "fpu/softfloat.h"
 
 #define SIGNBIT (uint32_t)0x80000000
 #define SIGNBIT64 ((uint64_t)1 << 63)
@@ -1376,3 +1377,93 @@ uint32_t HELPER(ror_cc)(CPUARMState *env, uint32_t x, uint32_t i)
         return ((uint32_t)x >> shift) | (x << (32 - shift));
     }
 }
+
+/*
+ * Implement float64 to int32_t conversion without saturation;
+ * the result is supplied modulo 2^32.
+ */
+uint64_t HELPER(fjcvtzs)(float64 value, void *vstatus)
+{
+    float_status *status = vstatus;
+    uint32_t result, exp, sign;
+    uint64_t frac;
+    uint32_t inexact; /* !Z */
+
+    sign = extract64(value, 63, 1);
+    exp = extract64(value, 52, 11);
+    frac = extract64(value, 0, 52);
+
+    if (exp == 0) {
+        /* While not inexact for IEEE FP, -0.0 is inexact for JavaScript.  */
+        inexact = sign;
+        result = 0;
+        if (frac != 0) {
+            if (status->flush_inputs_to_zero) {
+                float_raise(float_flag_input_denormal, status);
+            } else {
+                float_raise(float_flag_inexact, status);
+                inexact = 1;
+            }
+        }
+    } else if (exp == 0x7ff) {
+        if (frac == 0) {
+            /* Infinity.  */
+            result = 0;
+        } else {
+            /* NaN */
+            result = INT32_MAX;
+        }
+        /* This operation raises Invalid for both NaN and overflow (Inf).  */
+        float_raise(float_flag_invalid, status);
+        inexact = 1;
+    } else {
+        int shift, true_exp;
+
+        true_exp = exp - 1023;
+        shift = 52 - true_exp;
+
+        /* Restore implicit bit.  */
+        frac |= 1ull << 52;
+
+        /* Shift the fraction into place.  */
+        if (shift <= -64) {
+            /*
+             * The number is so large the fraction is shifted out entirely.
+             * The result mod 2^32 is 0 and will match the overflow case.
+             */
+            inexact = 1;
+            frac = 0;
+        } else if (shift <= 0) {
+            /* The number is so large we must shift the fraction left.  */
+            inexact = 1;
+            frac <<= -shift;
+        } else if (shift < 64) {
+            /* Normal case -- shift right and notice if bits shift out.  */
+            inexact = (frac << (64 - shift)) != 0;
+            frac >>= shift;
+        } else {
+            /* The number is so small the fraction is shifted out entirely.  */
+            inexact = 1;
+            frac = 0;
+        }
+
+        /* Notice overflow or inexact exceptions.  */
+        if (true_exp > 31
+            || frac > (sign ? 0x80000000ull : 0x7fffffff)) {
+            /* Overflow, for which this operation raises invalid.  */
+            float_raise(float_flag_invalid, status);
+            inexact = 1;
+        } else if (inexact) {
+            float_raise(float_flag_inexact, status);
+        }
+
+        /* Produce the result mod 2^32.  */
+        if (sign) {
+            frac = -frac;
+        }
+        result = frac;
+    }
+
+    /* Pack the result and the env->ZF representation of Z together.  */
+    return deposit64(result, 32, 32, inexact);
+}
diff --git a/target/arm/translate-a64.c b/target/arm/translate-a64.c
index 90298ac562..49289bf1d8 100644
--- a/target/arm/translate-a64.c
+++ b/target/arm/translate-a64.c
@@ -6458,6 +6458,24 @@ static void handle_fmov(DisasContext *s, int rd, int rn, int type, bool itof)
     }
 }
 
+static void handle_fjcvtzs(DisasContext *s, int rd, int rn)
+{
+    TCGv_i64 t = read_fp_dreg(s, rn);
+    TCGv_ptr fpstatus = get_fpstatus_ptr(false);
+
+    gen_helper_fjcvtzs(t, t, fpstatus);
+
+    tcg_temp_free_ptr(fpstatus);
+
+    tcg_gen_ext32u_i64(cpu_reg(s, rd), t);
+    tcg_gen_extrh_i64_i32(cpu_ZF, t);
+    tcg_gen_movi_i32(cpu_CF, 0);
+    tcg_gen_movi_i32(cpu_NF, 0);
+    tcg_gen_movi_i32(cpu_VF, 0);
+
+    tcg_temp_free_i64(t);
+}
+
 /* Floating point <-> integer conversions
  *   31   30  29 28       24 23  22  21 20   19 18 16 15         10 9  5 4  0
  * +----+---+---+-----------+------+---+-------+-----+-------------+----+----+
@@ -6533,6 +6551,14 @@ static void disas_fp_int_conv(DisasContext *s, uint32_t insn)
             handle_fmov(s, rd, rn, type, itof);
             break;
 
+        case 0b00111110: /* FJCVTZS */
+            if (!dc_isar_feature(aa64_jscvt, s)) {
+                goto do_unallocated;
+            } else if (fp_access_check(s)) {
+                handle_fjcvtzs(s, rd, rn);
+            }
+            break;
+
         default:
         do_unallocated:
             unallocated_encoding(s);
diff --git a/target/arm/translate.c b/target/arm/translate.c
index eb25895876..a92d06b05b 100644
--- a/target/arm/translate.c
+++ b/target/arm/translate.c
@@ -4066,6 +4066,21 @@ static int disas_vfp_insn(DisasContext *s, uint32_t insn)
                     case 17: /* fsito */
                         gen_vfp_sito(dp, 0);
                         break;
+                    case 19: /* vjcvt */
+                        if (!dp || !dc_isar_feature(aa32_jscvt, s)) {
+                            return 1;
+                        } else {
+                            TCGv_ptr fpst = get_fpstatus_ptr(0);
+                            gen_helper_fjcvtzs(cpu_F0d, cpu_F0d, fpst);
+                            tcg_temp_free_ptr(fpst);
+
+                            tcg_gen_extr_i64_i32(cpu_F0s, cpu_ZF, cpu_F0d);
+                            tcg_gen_movi_i32(cpu_NF, 0);
+                            tcg_gen_movi_i32(cpu_CF, 0);
+                            tcg_gen_movi_i32(cpu_VF, 0);
+                            dp = 0; /* always a single precision result */
+                        }
+                        break;
                     case 20: /* fshto */
                         if (!arm_dc_feature(s, ARM_FEATURE_VFP3)) {
                             return 1;
-- 
2.17.2

Re: [Qemu-devel] [PATCH 3/3] target/arm: Implement ARMv8.3-JSConv

[Laurent Desnogues]

Hello,

On Mon, Feb 4, 2019 at 6:38 AM Richard Henderson
<richard.henderson@linaro.org> wrote:
>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
>  target/arm/cpu.h           | 10 +++++
>  target/arm/helper.h        |  2 +
>  target/arm/cpu.c           |  1 +
>  target/arm/cpu64.c         |  2 +
>  target/arm/op_helper.c     | 91 ++++++++++++++++++++++++++++++++++++++
>  target/arm/translate-a64.c | 26 +++++++++++
>  target/arm/translate.c     | 15 +++++++
>  7 files changed, 147 insertions(+)
>
> diff --git a/target/arm/cpu.h b/target/arm/cpu.h
> index a68bcc9fed..d2c2e2b0cf 100644
> --- a/target/arm/cpu.h
> +++ b/target/arm/cpu.h
> @@ -3209,6 +3209,11 @@ static inline bool isar_feature_aa32_vcma(const ARMISARegisters *id)
>      return FIELD_EX32(id->id_isar5, ID_ISAR5, VCMA) != 0;
>  }
>
> +static inline bool isar_feature_aa32_jscvt(const ARMISARegisters *id)
> +{
> +    return FIELD_EX32(id->id_isar6, ID_ISAR6, JSCVT) != 0;
> +}
> +
>  static inline bool isar_feature_aa32_dp(const ARMISARegisters *id)
>  {
>      return FIELD_EX32(id->id_isar6, ID_ISAR6, DP) != 0;
> @@ -3287,6 +3292,11 @@ static inline bool isar_feature_aa64_dp(const ARMISARegisters *id)
>      return FIELD_EX64(id->id_aa64isar0, ID_AA64ISAR0, DP) != 0;
>  }
>
> +static inline bool isar_feature_aa64_jscvt(const ARMISARegisters *id)
> +{
> +    return FIELD_EX64(id->id_aa64isar1, ID_AA64ISAR1, JSCVT) != 0;
> +}
> +
>  static inline bool isar_feature_aa64_fcma(const ARMISARegisters *id)
>  {
>      return FIELD_EX64(id->id_aa64isar1, ID_AA64ISAR1, FCMA) != 0;
> diff --git a/target/arm/helper.h b/target/arm/helper.h
> index 53a38188c6..6998f7e8d5 100644
> --- a/target/arm/helper.h
> +++ b/target/arm/helper.h
> @@ -218,6 +218,8 @@ DEF_HELPER_FLAGS_2(rintd_exact, TCG_CALL_NO_RWG, f64, f64, ptr)
>  DEF_HELPER_FLAGS_2(rints, TCG_CALL_NO_RWG, f32, f32, ptr)
>  DEF_HELPER_FLAGS_2(rintd, TCG_CALL_NO_RWG, f64, f64, ptr)
>
> +DEF_HELPER_FLAGS_2(fjcvtzs, TCG_CALL_NO_RWG, i64, f64, ptr)
> +
>  /* neon_helper.c */
>  DEF_HELPER_FLAGS_3(neon_qadd_u8, TCG_CALL_NO_RWG, i32, env, i32, i32)
>  DEF_HELPER_FLAGS_3(neon_qadd_s8, TCG_CALL_NO_RWG, i32, env, i32, i32)
> diff --git a/target/arm/cpu.c b/target/arm/cpu.c
> index 3874dc9875..2eb2ce6c8c 100644
> --- a/target/arm/cpu.c
> +++ b/target/arm/cpu.c
> @@ -1995,6 +1995,7 @@ static void arm_max_initfn(Object *obj)
>              cpu->isar.id_isar5 = t;
>
>              t = cpu->isar.id_isar6;
> +            t = FIELD_DP32(t, ID_ISAR6, JSCVT, 1);
>              t = FIELD_DP32(t, ID_ISAR6, DP, 1);
>              cpu->isar.id_isar6 = t;
>
> diff --git a/target/arm/cpu64.c b/target/arm/cpu64.c
> index 7107ec8d7e..43d8ff047c 100644
> --- a/target/arm/cpu64.c
> +++ b/target/arm/cpu64.c
> @@ -311,6 +311,7 @@ static void aarch64_max_initfn(Object *obj)
>          cpu->isar.id_aa64isar0 = t;
>
>          t = cpu->isar.id_aa64isar1;
> +        t = FIELD_DP64(t, ID_AA64ISAR1, JSCVT, 1);
>          t = FIELD_DP64(t, ID_AA64ISAR1, FCMA, 1);
>          t = FIELD_DP64(t, ID_AA64ISAR1, APA, 1); /* PAuth, architected only */
>          t = FIELD_DP64(t, ID_AA64ISAR1, API, 0);
> @@ -340,6 +341,7 @@ static void aarch64_max_initfn(Object *obj)
>          cpu->isar.id_isar5 = u;
>
>          u = cpu->isar.id_isar6;
> +        u = FIELD_DP32(u, ID_ISAR6, JSCVT, 1);
>          u = FIELD_DP32(u, ID_ISAR6, DP, 1);
>          cpu->isar.id_isar6 = u;
>
> diff --git a/target/arm/op_helper.c b/target/arm/op_helper.c
> index c998eadfaa..a7259a7194 100644
> --- a/target/arm/op_helper.c
> +++ b/target/arm/op_helper.c
> @@ -24,6 +24,7 @@
>  #include "internals.h"
>  #include "exec/exec-all.h"
>  #include "exec/cpu_ldst.h"
> +#include "fpu/softfloat.h"
>
>  #define SIGNBIT (uint32_t)0x80000000
>  #define SIGNBIT64 ((uint64_t)1 << 63)
> @@ -1376,3 +1377,93 @@ uint32_t HELPER(ror_cc)(CPUARMState *env, uint32_t x, uint32_t i)
>          return ((uint32_t)x >> shift) | (x << (32 - shift));
>      }
>  }
> +
> +/*
> + * Implement float64 to int32_t conversion without saturation;
> + * the result is supplied modulo 2^32.
> + */
> +uint64_t HELPER(fjcvtzs)(float64 value, void *vstatus)
> +{
> +    float_status *status = vstatus;
> +    uint32_t result, exp, sign;
> +    uint64_t frac;
> +    uint32_t inexact; /* !Z */
> +
> +    sign = extract64(value, 63, 1);
> +    exp = extract64(value, 52, 11);
> +    frac = extract64(value, 0, 52);
> +
> +    if (exp == 0) {
> +        /* While not inexact for IEEE FP, -0.0 is inexact for JavaScript.  */
> +        inexact = sign;
> +        result = 0;
> +        if (frac != 0) {
> +            if (status->flush_inputs_to_zero) {
> +                float_raise(float_flag_input_denormal, status);
> +            } else {
> +                float_raise(float_flag_inexact, status);
> +                inexact = 1;
> +            }
> +        }
> +    } else if (exp == 0x7ff) {
> +        if (frac == 0) {
> +            /* Infinity.  */
> +            result = 0;
> +        } else {
> +            /* NaN */
> +            result = INT32_MAX;
> +        }

The pseudo code asks for NaN inputs to return 0 since FPUnpack returns
a value of 0 for NaN.

The rest looks correct.

Thanks,

Laurent

> +        /* This operation raises Invalid for both NaN and overflow (Inf).  */
> +        float_raise(float_flag_invalid, status);
> +        inexact = 1;
> +    } else {
> +        int shift, true_exp;
> +
> +        true_exp = exp - 1023;
> +        shift = 52 - true_exp;
> +
> +        /* Restore implicit bit.  */
> +        frac |= 1ull << 52;
> +
> +        /* Shift the fraction into place.  */
> +        if (shift <= -64) {
> +            /*
> +             * The number is so large the fraction is shifted out entirely.
> +             * The result mod 2^32 is 0 and will match the overflow case.
> +             */
> +            inexact = 1;
> +            frac = 0;
> +        } else if (shift <= 0) {
> +            /* The number is so large we must shift the fraction left.  */
> +            inexact = 1;
> +            frac <<= -shift;
> +        } else if (shift < 64) {
> +            /* Normal case -- shift right and notice if bits shift out.  */
> +            inexact = (frac << (64 - shift)) != 0;
> +            frac >>= shift;
> +        } else {
> +            /* The number is so small the fraction is shifted out entirely.  */
> +            inexact = 1;
> +            frac = 0;
> +        }
> +
> +        /* Notice overflow or inexact exceptions.  */
> +        if (true_exp > 31
> +            || frac > (sign ? 0x80000000ull : 0x7fffffff)) {
> +            /* Overflow, for which this operation raises invalid.  */
> +            float_raise(float_flag_invalid, status);
> +            inexact = 1;
> +        } else if (inexact) {
> +            float_raise(float_flag_inexact, status);
> +        }
> +
> +        /* Produce the result mod 2^32.  */
> +        if (sign) {
> +            frac = -frac;
> +        }
> +        result = frac;
> +    }
> +
> +    /* Pack the result and the env->ZF representation of Z together.  */
> +    return deposit64(result, 32, 32, inexact);
> +}
> diff --git a/target/arm/translate-a64.c b/target/arm/translate-a64.c
> index 90298ac562..49289bf1d8 100644
> --- a/target/arm/translate-a64.c
> +++ b/target/arm/translate-a64.c
> @@ -6458,6 +6458,24 @@ static void handle_fmov(DisasContext *s, int rd, int rn, int type, bool itof)
>      }
>  }
>
> +static void handle_fjcvtzs(DisasContext *s, int rd, int rn)
> +{
> +    TCGv_i64 t = read_fp_dreg(s, rn);
> +    TCGv_ptr fpstatus = get_fpstatus_ptr(false);
> +
> +    gen_helper_fjcvtzs(t, t, fpstatus);
> +
> +    tcg_temp_free_ptr(fpstatus);
> +
> +    tcg_gen_ext32u_i64(cpu_reg(s, rd), t);
> +    tcg_gen_extrh_i64_i32(cpu_ZF, t);
> +    tcg_gen_movi_i32(cpu_CF, 0);
> +    tcg_gen_movi_i32(cpu_NF, 0);
> +    tcg_gen_movi_i32(cpu_VF, 0);
> +
> +    tcg_temp_free_i64(t);
> +}
> +
>  /* Floating point <-> integer conversions
>   *   31   30  29 28       24 23  22  21 20   19 18 16 15         10 9  5 4  0
>   * +----+---+---+-----------+------+---+-------+-----+-------------+----+----+
> @@ -6533,6 +6551,14 @@ static void disas_fp_int_conv(DisasContext *s, uint32_t insn)
>              handle_fmov(s, rd, rn, type, itof);
>              break;
>
> +        case 0b00111110: /* FJCVTZS */
> +            if (!dc_isar_feature(aa64_jscvt, s)) {
> +                goto do_unallocated;
> +            } else if (fp_access_check(s)) {
> +                handle_fjcvtzs(s, rd, rn);
> +            }
> +            break;
> +
>          default:
>          do_unallocated:
>              unallocated_encoding(s);
> diff --git a/target/arm/translate.c b/target/arm/translate.c
> index eb25895876..a92d06b05b 100644
> --- a/target/arm/translate.c
> +++ b/target/arm/translate.c
> @@ -4066,6 +4066,21 @@ static int disas_vfp_insn(DisasContext *s, uint32_t insn)
>                      case 17: /* fsito */
>                          gen_vfp_sito(dp, 0);
>                          break;
> +                    case 19: /* vjcvt */
> +                        if (!dp || !dc_isar_feature(aa32_jscvt, s)) {
> +                            return 1;
> +                        } else {
> +                            TCGv_ptr fpst = get_fpstatus_ptr(0);
> +                            gen_helper_fjcvtzs(cpu_F0d, cpu_F0d, fpst);
> +                            tcg_temp_free_ptr(fpst);
> +
> +                            tcg_gen_extr_i64_i32(cpu_F0s, cpu_ZF, cpu_F0d);
> +                            tcg_gen_movi_i32(cpu_NF, 0);
> +                            tcg_gen_movi_i32(cpu_CF, 0);
> +                            tcg_gen_movi_i32(cpu_VF, 0);
> +                            dp = 0; /* always a single precision result */
> +                        }
> +                        break;
>                      case 20: /* fshto */
>                          if (!arm_dc_feature(s, ARM_FEATURE_VFP3)) {
>                              return 1;
> --
> 2.17.2
>
>

Re: [Qemu-devel] [PATCH 3/3] target/arm: Implement ARMv8.3-JSConv

[Richard Henderson]

On 2/4/19 8:00 AM, Laurent Desnogues wrote:
> Hello,
> 
> On Mon, Feb 4, 2019 at 6:38 AM Richard Henderson
> <richard.henderson@linaro.org> wrote:
>>
>> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
>> ---
>>  target/arm/cpu.h           | 10 +++++
>>  target/arm/helper.h        |  2 +
>>  target/arm/cpu.c           |  1 +
>>  target/arm/cpu64.c         |  2 +
>>  target/arm/op_helper.c     | 91 ++++++++++++++++++++++++++++++++++++++
>>  target/arm/translate-a64.c | 26 +++++++++++
>>  target/arm/translate.c     | 15 +++++++
>>  7 files changed, 147 insertions(+)
>>
>> diff --git a/target/arm/cpu.h b/target/arm/cpu.h
>> index a68bcc9fed..d2c2e2b0cf 100644
>> --- a/target/arm/cpu.h
>> +++ b/target/arm/cpu.h
>> @@ -3209,6 +3209,11 @@ static inline bool isar_feature_aa32_vcma(const ARMISARegisters *id)
>>      return FIELD_EX32(id->id_isar5, ID_ISAR5, VCMA) != 0;
>>  }
>>
>> +static inline bool isar_feature_aa32_jscvt(const ARMISARegisters *id)
>> +{
>> +    return FIELD_EX32(id->id_isar6, ID_ISAR6, JSCVT) != 0;
>> +}
>> +
>>  static inline bool isar_feature_aa32_dp(const ARMISARegisters *id)
>>  {
>>      return FIELD_EX32(id->id_isar6, ID_ISAR6, DP) != 0;
>> @@ -3287,6 +3292,11 @@ static inline bool isar_feature_aa64_dp(const ARMISARegisters *id)
>>      return FIELD_EX64(id->id_aa64isar0, ID_AA64ISAR0, DP) != 0;
>>  }
>>
>> +static inline bool isar_feature_aa64_jscvt(const ARMISARegisters *id)
>> +{
>> +    return FIELD_EX64(id->id_aa64isar1, ID_AA64ISAR1, JSCVT) != 0;
>> +}
>> +
>>  static inline bool isar_feature_aa64_fcma(const ARMISARegisters *id)
>>  {
>>      return FIELD_EX64(id->id_aa64isar1, ID_AA64ISAR1, FCMA) != 0;
>> diff --git a/target/arm/helper.h b/target/arm/helper.h
>> index 53a38188c6..6998f7e8d5 100644
>> --- a/target/arm/helper.h
>> +++ b/target/arm/helper.h
>> @@ -218,6 +218,8 @@ DEF_HELPER_FLAGS_2(rintd_exact, TCG_CALL_NO_RWG, f64, f64, ptr)
>>  DEF_HELPER_FLAGS_2(rints, TCG_CALL_NO_RWG, f32, f32, ptr)
>>  DEF_HELPER_FLAGS_2(rintd, TCG_CALL_NO_RWG, f64, f64, ptr)
>>
>> +DEF_HELPER_FLAGS_2(fjcvtzs, TCG_CALL_NO_RWG, i64, f64, ptr)
>> +
>>  /* neon_helper.c */
>>  DEF_HELPER_FLAGS_3(neon_qadd_u8, TCG_CALL_NO_RWG, i32, env, i32, i32)
>>  DEF_HELPER_FLAGS_3(neon_qadd_s8, TCG_CALL_NO_RWG, i32, env, i32, i32)
>> diff --git a/target/arm/cpu.c b/target/arm/cpu.c
>> index 3874dc9875..2eb2ce6c8c 100644
>> --- a/target/arm/cpu.c
>> +++ b/target/arm/cpu.c
>> @@ -1995,6 +1995,7 @@ static void arm_max_initfn(Object *obj)
>>              cpu->isar.id_isar5 = t;
>>
>>              t = cpu->isar.id_isar6;
>> +            t = FIELD_DP32(t, ID_ISAR6, JSCVT, 1);
>>              t = FIELD_DP32(t, ID_ISAR6, DP, 1);
>>              cpu->isar.id_isar6 = t;
>>
>> diff --git a/target/arm/cpu64.c b/target/arm/cpu64.c
>> index 7107ec8d7e..43d8ff047c 100644
>> --- a/target/arm/cpu64.c
>> +++ b/target/arm/cpu64.c
>> @@ -311,6 +311,7 @@ static void aarch64_max_initfn(Object *obj)
>>          cpu->isar.id_aa64isar0 = t;
>>
>>          t = cpu->isar.id_aa64isar1;
>> +        t = FIELD_DP64(t, ID_AA64ISAR1, JSCVT, 1);
>>          t = FIELD_DP64(t, ID_AA64ISAR1, FCMA, 1);
>>          t = FIELD_DP64(t, ID_AA64ISAR1, APA, 1); /* PAuth, architected only */
>>          t = FIELD_DP64(t, ID_AA64ISAR1, API, 0);
>> @@ -340,6 +341,7 @@ static void aarch64_max_initfn(Object *obj)
>>          cpu->isar.id_isar5 = u;
>>
>>          u = cpu->isar.id_isar6;
>> +        u = FIELD_DP32(u, ID_ISAR6, JSCVT, 1);
>>          u = FIELD_DP32(u, ID_ISAR6, DP, 1);
>>          cpu->isar.id_isar6 = u;
>>
>> diff --git a/target/arm/op_helper.c b/target/arm/op_helper.c
>> index c998eadfaa..a7259a7194 100644
>> --- a/target/arm/op_helper.c
>> +++ b/target/arm/op_helper.c
>> @@ -24,6 +24,7 @@
>>  #include "internals.h"
>>  #include "exec/exec-all.h"
>>  #include "exec/cpu_ldst.h"
>> +#include "fpu/softfloat.h"
>>
>>  #define SIGNBIT (uint32_t)0x80000000
>>  #define SIGNBIT64 ((uint64_t)1 << 63)
>> @@ -1376,3 +1377,93 @@ uint32_t HELPER(ror_cc)(CPUARMState *env, uint32_t x, uint32_t i)
>>          return ((uint32_t)x >> shift) | (x << (32 - shift));
>>      }
>>  }
>> +
>> +/*
>> + * Implement float64 to int32_t conversion without saturation;
>> + * the result is supplied modulo 2^32.
>> + */
>> +uint64_t HELPER(fjcvtzs)(float64 value, void *vstatus)
>> +{
>> +    float_status *status = vstatus;
>> +    uint32_t result, exp, sign;
>> +    uint64_t frac;
>> +    uint32_t inexact; /* !Z */
>> +
>> +    sign = extract64(value, 63, 1);
>> +    exp = extract64(value, 52, 11);
>> +    frac = extract64(value, 0, 52);
>> +
>> +    if (exp == 0) {
>> +        /* While not inexact for IEEE FP, -0.0 is inexact for JavaScript.  */
>> +        inexact = sign;
>> +        result = 0;
>> +        if (frac != 0) {
>> +            if (status->flush_inputs_to_zero) {
>> +                float_raise(float_flag_input_denormal, status);
>> +            } else {
>> +                float_raise(float_flag_inexact, status);
>> +                inexact = 1;
>> +            }
>> +        }
>> +    } else if (exp == 0x7ff) {
>> +        if (frac == 0) {
>> +            /* Infinity.  */
>> +            result = 0;
>> +        } else {
>> +            /* NaN */
>> +            result = INT32_MAX;
>> +        }
> 
> The pseudo code asks for NaN inputs to return 0 since FPUnpack returns
> a value of 0 for NaN.
> 
> The rest looks correct.

So it does.  I apparently copied the INT_MAX thing from softfloat.c without
double-checking.

Thanks,


r~

Re: [Qemu-devel] [PATCH 3/3] target/arm: Implement ARMv8.3-JSConv

[Peter Maydell]

On Mon, 4 Feb 2019 at 10:17, Richard Henderson
<richard.henderson@linaro.org> wrote:
>
> On 2/4/19 8:00 AM, Laurent Desnogues wrote:
> > The pseudo code asks for NaN inputs to return 0 since FPUnpack returns
> > a value of 0 for NaN.
> >
> > The rest looks correct.
>
> So it does.  I apparently copied the INT_MAX thing from softfloat.c without
> double-checking.

I'm a little surprised risu didn't catch that -- it's usually
quite keen on checking NaN behaviour...

thanks
-- PMM

Re: [Qemu-devel] [PATCH 3/3] target/arm: Implement ARMv8.3-JSConv

[Richard Henderson]

On 2/4/19 11:21 AM, Peter Maydell wrote:
> On Mon, 4 Feb 2019 at 10:17, Richard Henderson
> <richard.henderson@linaro.org> wrote:
>>
>> On 2/4/19 8:00 AM, Laurent Desnogues wrote:
>>> The pseudo code asks for NaN inputs to return 0 since FPUnpack returns
>>> a value of 0 for NaN.
>>>
>>> The rest looks correct.
>>
>> So it does.  I apparently copied the INT_MAX thing from softfloat.c without
>> double-checking.
> 
> I'm a little surprised risu didn't catch that -- it's usually
> quite keen on checking NaN behaviour...

I'm not sure either...  Because of other issues, I regenerated the .bin file
and the nan case does trigger now.  I'm not sure what happened before.


r~

Re: [Qemu-devel] [PATCH 0/3] target/arm: Implement ARMv8.3-JSConv

[no-reply]

Patchew URL: https://patchew.org/QEMU/20190204052712.30833-1-richard.henderson@linaro.org/



Hi,

This series failed the docker-mingw@fedora build test. Please find the testing commands and
their output below. If you have Docker installed, you can probably reproduce it
locally.

=== TEST SCRIPT BEGIN ===
#!/bin/bash
time make docker-test-mingw@fedora SHOW_ENV=1 J=14
=== TEST SCRIPT END ===


Configure options:
--enable-werror --target-list=x86_64-softmmu,aarch64-softmmu --prefix=/tmp/qemu-test/install --python=/usr/bin/python3 --cross-prefix=x86_64-w64-mingw32- --enable-trace-backends=simple --enable-gnutls --enable-nettle --enable-curl --enable-vnc --enable-bzip2 --enable-guest-agent --with-sdlabi=2.0
ERROR: unknown option --with-sdlabi=2.0
Try '/tmp/qemu-test/src/configure --help' for more information
# QEMU configure log Mon Feb  4 05:48:37 UTC 2019
# Configured with: '/tmp/qemu-test/src/configure' '--enable-werror' '--target-list=x86_64-softmmu,aarch64-softmmu' '--prefix=/tmp/qemu-test/install' '--python=/usr/bin/python3' '--cross-prefix=x86_64-w64-mingw32-' '--enable-trace-backends=simple' '--enable-gnutls' '--enable-nettle' '--enable-curl' '--enable-vnc' '--enable-bzip2' '--enable-guest-agent' '--with-sdlabi=2.0'
---
funcs: do_compiler do_cc compile_object check_define main
lines: 92 122 617 634 0
x86_64-w64-mingw32-gcc -D_GNU_SOURCE -D_FILE_OFFSET_BITS=64 -D_LARGEFILE_SOURCE -Wstrict-prototypes -Wredundant-decls -Wall -Wundef -Wwrite-strings -Wmissing-prototypes -fno-strict-aliasing -fno-common -fwrapv -std=gnu99 -c -o config-temp/qemu-conf.o config-temp/qemu-conf.c
config-temp/qemu-conf.c:2:2: error: #error __linux__ not defined
 #error __linux__ not defined
  ^~~~~

---
funcs: do_compiler do_cc compile_object check_define main
lines: 92 122 617 686 0
x86_64-w64-mingw32-gcc -D_GNU_SOURCE -D_FILE_OFFSET_BITS=64 -D_LARGEFILE_SOURCE -Wstrict-prototypes -Wredundant-decls -Wall -Wundef -Wwrite-strings -Wmissing-prototypes -fno-strict-aliasing -fno-common -fwrapv -std=gnu99 -c -o config-temp/qemu-conf.o config-temp/qemu-conf.c
config-temp/qemu-conf.c:2:2: error: #error __i386__ not defined
 #error __i386__ not defined
  ^~~~~

---
funcs: do_compiler do_cc compile_object check_define main
lines: 92 122 617 689 0
x86_64-w64-mingw32-gcc -D_GNU_SOURCE -D_FILE_OFFSET_BITS=64 -D_LARGEFILE_SOURCE -Wstrict-prototypes -Wredundant-decls -Wall -Wundef -Wwrite-strings -Wmissing-prototypes -fno-strict-aliasing -fno-common -fwrapv -std=gnu99 -c -o config-temp/qemu-conf.o config-temp/qemu-conf.c
config-temp/qemu-conf.c:2:2: error: #error __ILP32__ not defined
 #error __ILP32__ not defined
  ^~~~~

---
lines: 92 128 920 0
x86_64-w64-mingw32-gcc -mthreads -D_GNU_SOURCE -D_FILE_OFFSET_BITS=64 -D_LARGEFILE_SOURCE -Wstrict-prototypes -Wredundant-decls -Wall -Wundef -Wwrite-strings -Wmissing-prototypes -fno-strict-aliasing -fno-common -fwrapv -std=gnu99 -o config-temp/qemu-conf.exe config-temp/qemu-conf.c -g -liberty
/usr/lib/gcc/x86_64-w64-mingw32/8.2.0/../../../../x86_64-w64-mingw32/bin/ld: cannot find -liberty
collect2: error: ld returned 1 exit status
Failed to run 'configure'
Traceback (most recent call last):
  File "./tests/docker/docker.py", line 563, in <module>


The full log is available at
http://patchew.org/logs/20190204052712.30833-1-richard.henderson@linaro.org/testing.docker-mingw@fedora/?type=message.
---
Email generated automatically by Patchew [http://patchew.org/].
Please send your feedback to patchew-devel@redhat.com

Re: [Qemu-devel] [PATCH 0/3] target/arm: Implement ARMv8.3-JSConv

[Peter Maydell]

On Mon, 4 Feb 2019 at 05:27, Richard Henderson
<richard.henderson@linaro.org> wrote:
>
> Two cleanups to surrounding code preceed the actual implementation.
>
> I did not find the ARM FPToFixedJS pseudo-function particularly helpful,
> so I mostly cribed off of the Alpha cvttq implementation.  But I think
> it all makes sense -- convert, produce the mod 2^32 result with NaN
> getting INT32_MAX, set ZF iff the conversion is exact.
>
> I've tested the aa64 version vs FVP with RISU.  I need to adjust my
> FVP configuration in order to test aa32.

I've reviewed patch 1; I'll leave 2 and 3 til you post v2.

thanks
-- PMM