diff mbox series

[RFC,for,2.11,14/23] target/arm/translate-a64.c: add ARMv8.2 fadd scalar half-precision

Message ID 20170720150426.12393-15-alex.bennee@linaro.org
State New
Headers show
Series Implementing FP16 for ARMv8.2 using SoftFloat2a and 3c | expand

Commit Message

Alex Bennée July 20, 2017, 3:04 p.m. UTC 
This brings in the initial decode skeleton and the helpers for a
scalar half-precision fadd using SoftFloat3c.

Signed-off-by: Alex Bennée <alex.bennee@linaro.org>

---
 target/arm/advsimd_helper.c | 15 ++++++++++++
 target/arm/translate-a64.c  | 60 +++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 75 insertions(+)

-- 
2.13.0

Comments

Richard Henderson July 20, 2017, 7:40 p.m. UTC | #1 
On 07/20/2017 05:04 AM, Alex Bennée wrote:
> This brings in the initial decode skeleton and the helpers for a

> scalar half-precision fadd using SoftFloat3c.

> 

> Signed-off-by: Alex Bennée<alex.bennee@linaro.org>

> ---

>   target/arm/advsimd_helper.c | 15 ++++++++++++

>   target/arm/translate-a64.c  | 60 +++++++++++++++++++++++++++++++++++++++++++++

>   2 files changed, 75 insertions(+)


... and there it is.  Squash this with previous?  Anyway, do something so that 
they're bisectable.


r~
Alex Bennée July 21, 2017, 9:58 a.m. UTC | #2 
Richard Henderson <rth@twiddle.net> writes:

> On 07/20/2017 05:04 AM, Alex Bennée wrote:

>> This brings in the initial decode skeleton and the helpers for a

>> scalar half-precision fadd using SoftFloat3c.

>>

>> Signed-off-by: Alex Bennée<alex.bennee@linaro.org>

>> ---

>>   target/arm/advsimd_helper.c | 15 ++++++++++++

>>   target/arm/translate-a64.c  | 60 +++++++++++++++++++++++++++++++++++++++++++++

>>   2 files changed, 75 insertions(+)

>

> ... and there it is.  Squash this with previous?  Anyway, do something

> so that they're bisectable.


Yeah a re-basing failure on my part as I cleaned up the tree. It was
fairly messy by the time I got everything working ;-)

--
Alex Bennée
diff mbox series

Patch

diff --git a/target/arm/advsimd_helper.c b/target/arm/advsimd_helper.c
index ec875a83bb..4e577e9602 100644
--- a/target/arm/advsimd_helper.c
+++ b/target/arm/advsimd_helper.c
@@ -83,3 +83,18 @@  uint32_t HELPER(advsimd_acgt_f16)(uint32_t a, uint32_t b, void *fpstp)
     sync_softfloat_flags_to_2a(fpstp);
     return r;
 }
+
+
+/* Data processing - scalar floating-point and advanced SIMD */
+
+uint32_t HELPER(advsimd_addh)(uint32_t a, uint32_t b, void *fpstp)
+{
+    union ui16_f16 uA = { .ui = a };
+    union ui16_f16 uB = { .ui = b };
+    float16_t r;
+
+    sync_softfloat_flags_from_2a(fpstp);
+    r = f16_add(uA.f, uB.f);
+    sync_softfloat_flags_to_2a(fpstp);
+    return r.v;
+}
diff --git a/target/arm/translate-a64.c b/target/arm/translate-a64.c
index f6aca395bd..a8876e4bcc 100644
--- a/target/arm/translate-a64.c
+++ b/target/arm/translate-a64.c
@@ -581,6 +581,14 @@  static TCGv_i32 read_fp_sreg(DisasContext *s, int reg)
     return v;
 }
 
+static TCGv_i32 read_fp_hreg(DisasContext *s, int reg)
+{
+    TCGv_i32 v = tcg_temp_new_i32();
+
+    tcg_gen_ld16u_i32(v, cpu_env, fp_reg_offset(s, reg, MO_16));
+    return v;
+}
+
 static void write_fp_dreg(DisasContext *s, int reg, TCGv_i64 v)
 {
     TCGv_i64 tcg_zero = tcg_const_i64(0);
@@ -599,6 +607,15 @@  static void write_fp_sreg(DisasContext *s, int reg, TCGv_i32 v)
     tcg_temp_free_i64(tmp);
 }
 
+static void write_fp_hreg(DisasContext *s, int reg, TCGv_i32 v)
+{
+    TCGv_i64 tmp = tcg_temp_new_i64();
+
+    tcg_gen_extu_i32_i64(tmp, v);
+    write_fp_dreg(s, reg, tmp);
+    tcg_temp_free_i64(tmp);
+}
+
 static TCGv_ptr get_fpstatus_ptr(void)
 {
     TCGv_ptr statusptr = tcg_temp_new_ptr();
@@ -4710,6 +4727,39 @@  static void disas_fp_1src(DisasContext *s, uint32_t insn)
     }
 }
 
+/* ARMv8.2 floating-point data-processing (2 source) - half precision
+ */
+static void handle_fp_2src_half(DisasContext *s, int opcode,
+                                int rd, int rn, int rm)
+{
+    TCGv_i32 tcg_op1;
+    TCGv_i32 tcg_op2;
+    TCGv_i32 tcg_res;
+    TCGv_ptr fpst;
+
+    tcg_res = tcg_temp_new_i32();
+    fpst = get_fpstatus_ptr();
+    tcg_op1 = read_fp_hreg(s, rn);
+    tcg_op2 = read_fp_hreg(s, rm);
+
+    switch (opcode) {
+    case 0x2: /* FADD */
+        gen_helper_advsimd_addh(tcg_res, tcg_op1, tcg_op2, fpst);
+        break;
+    default:
+        fprintf(stderr, "%s: unhandled op %#x\n", __func__, opcode);
+        unsupported_encoding(s, opcode);
+        break;
+    }
+
+    write_fp_hreg(s, rd, tcg_res);
+
+    tcg_temp_free_ptr(fpst);
+    tcg_temp_free_i32(tcg_op1);
+    tcg_temp_free_i32(tcg_op2);
+    tcg_temp_free_i32(tcg_res);
+}
+
 /* C3.6.26 Floating-point data-processing (2 source) - single precision */
 static void handle_fp_2src_single(DisasContext *s, int opcode,
                                   int rd, int rn, int rm)
@@ -4848,6 +4898,16 @@  static void disas_fp_2src(DisasContext *s, uint32_t insn)
         }
         handle_fp_2src_double(s, opcode, rd, rn, rm);
         break;
+    case 3:
+        if (arm_dc_feature(s, ARM_FEATURE_V8_FP16)) {
+            if (!fp_access_check(s)) {
+                return;
+            }
+            handle_fp_2src_half(s, opcode, rd, rn, rm);
+        } else {
+            unallocated_encoding(s);
+        }
+        break;
     default:
         unallocated_encoding(s);
     }