diff mbox series

[v2,14/27] target/arm: squash FZ16 behaviour for conversions

Message ID 20180512004311.9299-15-richard.henderson@linaro.org
State Superseded
Headers show
Series softfloat patch roundup | expand

Commit Message

Richard Henderson May 12, 2018, 12:42 a.m. UTC 
From: Alex Bennée <alex.bennee@linaro.org>


The ARM ARM specifies FZ16 is suppressed for conversions. Rather than
pushing this logic into the softfloat code we can simply save the FZ
state and temporarily disable it for the softfloat call.

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

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>


---
v4
  - float16_to_floatX squished the wrong softfloat bit for FZ16;
    need to adjust input denormals in this case.
---
 target/arm/helper.c | 40 ++++++++++++++++++++++++++++++++++++----
 1 file changed, 36 insertions(+), 4 deletions(-)

-- 
2.17.0

Comments

Peter Maydell May 14, 2018, 1:53 p.m. UTC | #1 
On 12 May 2018 at 01:42, Richard Henderson <richard.henderson@linaro.org> wrote:
> From: Alex Bennée <alex.bennee@linaro.org>

>

> The ARM ARM specifies FZ16 is suppressed for conversions. Rather than

> pushing this logic into the softfloat code we can simply save the FZ

> state and temporarily disable it for the softfloat call.

>

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

> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

>

> ---

> v4

>   - float16_to_floatX squished the wrong softfloat bit for FZ16;

>     need to adjust input denormals in this case.

> ---

>  target/arm/helper.c | 40 ++++++++++++++++++++++++++++++++++++----

>  1 file changed, 36 insertions(+), 4 deletions(-)


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


(once you've fixed up which fp status pointer we pass to these
helpers, then they do the right thing)

thanks
-- PMM
diff mbox series

Patch

diff --git a/target/arm/helper.c b/target/arm/helper.c
index a1c1dc5bbe..e05c7230d4 100644
--- a/target/arm/helper.c
+++ b/target/arm/helper.c
@@ -11542,22 +11542,54 @@  uint32_t HELPER(set_neon_rmode)(uint32_t rmode, CPUARMState *env)
 /* Half precision conversions.  */
 float32 HELPER(vfp_fcvt_f16_to_f32)(float16 a, void *fpstp, uint32_t ahp_mode)
 {
-    return float16_to_float32(a, !ahp_mode, fpstp);
+    /* Squash FZ16 to 0 for the duration of conversion.  In this case,
+     * it would affect flushing input denormals.
+     */
+    float_status *fpst = fpstp;
+    flag save = get_flush_inputs_to_zero(fpst);
+    set_flush_inputs_to_zero(false, fpst);
+    float32 r = float16_to_float32(a, !ahp_mode, fpst);
+    set_flush_inputs_to_zero(save, fpst);
+    return r;
 }
 
 float16 HELPER(vfp_fcvt_f32_to_f16)(float32 a, void *fpstp, uint32_t ahp_mode)
 {
-    return float32_to_float16(a, !ahp_mode, fpstp);
+    /* Squash FZ16 to 0 for the duration of conversion.  In this case,
+     * it would affect flushing output denormals.
+     */
+    float_status *fpst = fpstp;
+    flag save = get_flush_to_zero(fpst);
+    set_flush_to_zero(false, fpst);
+    float16 r = float32_to_float16(a, !ahp_mode, fpst);
+    set_flush_to_zero(save, fpst);
+    return r;
 }
 
 float64 HELPER(vfp_fcvt_f16_to_f64)(float16 a, void *fpstp, uint32_t ahp_mode)
 {
-    return float16_to_float64(a, !ahp_mode, fpstp);
+    /* Squash FZ16 to 0 for the duration of conversion.  In this case,
+     * it would affect flushing input denormals.
+     */
+    float_status *fpst = fpstp;
+    flag save = get_flush_inputs_to_zero(fpst);
+    set_flush_inputs_to_zero(false, fpst);
+    float64 r = float16_to_float64(a, !ahp_mode, fpst);
+    set_flush_inputs_to_zero(save, fpst);
+    return r;
 }
 
 float16 HELPER(vfp_fcvt_f64_to_f16)(float64 a, void *fpstp, uint32_t ahp_mode)
 {
-    return float64_to_float16(a, !ahp_mode, fpstp);
+    /* Squash FZ16 to 0 for the duration of conversion.  In this case,
+     * it would affect flushing output denormals.
+     */
+    float_status *fpst = fpstp;
+    flag save = get_flush_to_zero(fpst);
+    set_flush_to_zero(false, fpst);
+    float16 r = float64_to_float16(a, !ahp_mode, fpst);
+    set_flush_to_zero(save, fpst);
+    return r;
 }
 
 #define float32_two make_float32(0x40000000)