On 2/14/19 1:16 AM, Laurent Desnogues wrote:
> Hello,
>
> On Thu, Feb 14, 2019 at 5:00 AM Richard Henderson
> <richard.hender...@linaro.org> wrote:
>>
>> Note that float16_to_float32 rightly squashes SNaN to QNaN.
>> But of course pickNaNMulAdd, for ARM, selects SNaNs first.
>> So we have to preserve SNaN long enough for the correct NaN
>> to be selected. Thus float16_to_float32_by_bits.
>>
>> Signed-off-by: Richard Henderson <richard.hender...@linaro.org>
>> ---
>> target/arm/helper.h | 9 +++
>> target/arm/vec_helper.c | 154 ++++++++++++++++++++++++++++++++++++++++
>> 2 files changed, 163 insertions(+)
>>
>> diff --git a/target/arm/helper.h b/target/arm/helper.h
>> index 53a38188c6..0302e13604 100644
>> --- a/target/arm/helper.h
>> +++ b/target/arm/helper.h
>> @@ -653,6 +653,15 @@ DEF_HELPER_FLAGS_6(gvec_fmla_idx_s, TCG_CALL_NO_RWG,
>> DEF_HELPER_FLAGS_6(gvec_fmla_idx_d, TCG_CALL_NO_RWG,
>> void, ptr, ptr, ptr, ptr, ptr, i32)
>>
>> +DEF_HELPER_FLAGS_5(gvec_fmlal_h, TCG_CALL_NO_RWG,
>> + void, ptr, ptr, ptr, ptr, i32)
>> +DEF_HELPER_FLAGS_5(gvec_fmlsl_h, TCG_CALL_NO_RWG,
>> + void, ptr, ptr, ptr, ptr, i32)
>> +DEF_HELPER_FLAGS_5(gvec_fmlal_idx_h, TCG_CALL_NO_RWG,
>> + void, ptr, ptr, ptr, ptr, i32)
>> +DEF_HELPER_FLAGS_5(gvec_fmlsl_idx_h, TCG_CALL_NO_RWG,
>> + void, ptr, ptr, ptr, ptr, i32)
>> +
>> #ifdef TARGET_AARCH64
>> #include "helper-a64.h"
>> #include "helper-sve.h"
>> diff --git a/target/arm/vec_helper.c b/target/arm/vec_helper.c
>> index 37f338732e..0c3b3de961 100644
>> --- a/target/arm/vec_helper.c
>> +++ b/target/arm/vec_helper.c
>> @@ -766,3 +766,157 @@ DO_FMLA_IDX(gvec_fmla_idx_s, float32, H4)
>> DO_FMLA_IDX(gvec_fmla_idx_d, float64, )
>>
>> #undef DO_FMLA_IDX
>> +
>> +/*
>> + * Convert float16 to float32, raising no exceptions and
>> + * preserving exceptional values, including SNaN.
>> + * This is effectively an unpack+repack operation.
>> + */
>> +static float32 float16_to_float32_by_bits(uint32_t f16)
>> +{
>> + const int f16_bias = 15;
>> + const int f32_bias = 127;
>> + uint32_t sign = extract32(f16, 15, 1);
>> + uint32_t exp = extract32(f16, 10, 5);
>> + uint32_t frac = extract32(f16, 0, 10);
>> +
>> + if (exp == 0x1f) {
>> + /* Inf or NaN */
>> + exp = 0xff;
>> + } else if (exp == 0) {
>> + /* Zero or denormal. */
>> + if (frac != 0) {
>> + /*
>> + * Denormal; these are all normal float32.
>> + * Shift the fraction so that the msb is at bit 11,
>> + * then remove bit 11 as the implicit bit of the
>> + * normalized float32. Note that we still go through
>> + * the shift for normal numbers below, to put the
>> + * float32 fraction at the right place.
>> + */
>> + int shift = clz32(frac) - 21;
>> + frac = (frac << shift) & 0x3ff;
>> + exp = f32_bias - f16_bias - shift + 1;
>
> If FZ16 is set, this should flush to zero.
Ho, hum, yes it should.
> This means you will have to use both fp_status (for the muladd) and
> fp_status_f16 (for this function) and so you should pass cpu_env to
> the helpers rather than the fp_status.
It's not quite as simple as that, because aa32 mode would pass
standard_fp_status. I'll figure something out...
r~
