Message ID | 20241025141254.2141506-17-peter.maydell@linaro.org (mailing list archive) |
---|---|
State | New |
Headers | show |
Series | softfloat: Set 2-NaN propagation rule in float_status, not at compile time | expand |
On 25/10/24 11:12, Peter Maydell wrote: > Set the NaN propagation rule explicitly for the float_status word > used in this target. > > This is a no-behaviour-change commit, so we retain the existing > behaviour of x87-style pick-largest-significand NaN propagation. > This is however not the architecturally correct handling, so we leave > a TODO note to that effect. > > We also leave a TODO note pointing out that all this code in the cpu > initfn (including the existing setting up of env->flags and the FPCR) > should be in a currently non-existent CPU reset function. > > Signed-off-by: Peter Maydell <peter.maydell@linaro.org> > --- > target/alpha/cpu.c | 11 +++++++++++ > fpu/softfloat-specialize.c.inc | 2 +- > 2 files changed, 12 insertions(+), 1 deletion(-) Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
On 10/25/24 15:12, Peter Maydell wrote: > Set the NaN propagation rule explicitly for the float_status word > used in this target. > > This is a no-behaviour-change commit, so we retain the existing > behaviour of x87-style pick-largest-significand NaN propagation. > This is however not the architecturally correct handling, so we leave > a TODO note to that effect. > > We also leave a TODO note pointing out that all this code in the cpu > initfn (including the existing setting up of env->flags and the FPCR) > should be in a currently non-existent CPU reset function. > > Signed-off-by: Peter Maydell<peter.maydell@linaro.org> > --- > target/alpha/cpu.c | 11 +++++++++++ > fpu/softfloat-specialize.c.inc | 2 +- > 2 files changed, 12 insertions(+), 1 deletion(-) Reviewed-by: Richard Henderson <richard.henderson@linaro.org> r~
diff --git a/target/alpha/cpu.c b/target/alpha/cpu.c index 9db1dffc03e..5d75c941f7a 100644 --- a/target/alpha/cpu.c +++ b/target/alpha/cpu.c @@ -24,6 +24,7 @@ #include "qemu/qemu-print.h" #include "cpu.h" #include "exec/exec-all.h" +#include "fpu/softfloat.h" static void alpha_cpu_set_pc(CPUState *cs, vaddr value) @@ -187,7 +188,17 @@ static void alpha_cpu_initfn(Object *obj) { CPUAlphaState *env = cpu_env(CPU(obj)); + /* TODO all this should be done in reset, not init */ + env->lock_addr = -1; + + /* + * TODO: this is incorrect. The Alpha Architecture Handbook version 4 + * describes NaN propagation in section 4.7.10.4. We should prefer + * the operand in Fb (whether it is a QNaN or an SNaN), then the + * operand in Fa. That is float_2nan_prop_ba. + */ + set_float_2nan_prop_rule(float_2nan_prop_x87, &env->fp_status); #if defined(CONFIG_USER_ONLY) env->flags = ENV_FLAG_PS_USER | ENV_FLAG_FEN; cpu_alpha_store_fpcr(env, (uint64_t)(FPCR_INVD | FPCR_DZED | FPCR_OVFD diff --git a/fpu/softfloat-specialize.c.inc b/fpu/softfloat-specialize.c.inc index 77ebc8216f6..a5c3e2b8de5 100644 --- a/fpu/softfloat-specialize.c.inc +++ b/fpu/softfloat-specialize.c.inc @@ -406,7 +406,7 @@ static int pickNaN(FloatClass a_cls, FloatClass b_cls, || defined(TARGET_LOONGARCH64) || defined(TARGET_HPPA) \ || defined(TARGET_S390X) || defined(TARGET_PPC) || defined(TARGET_M68K) \ || defined(TARGET_SPARC) || defined(TARGET_XTENSA) \ - || defined(TARGET_I386) + || defined(TARGET_I386) || defined(TARGET_ALPHA) g_assert_not_reached(); #else rule = float_2nan_prop_x87;
Set the NaN propagation rule explicitly for the float_status word used in this target. This is a no-behaviour-change commit, so we retain the existing behaviour of x87-style pick-largest-significand NaN propagation. This is however not the architecturally correct handling, so we leave a TODO note to that effect. We also leave a TODO note pointing out that all this code in the cpu initfn (including the existing setting up of env->flags and the FPCR) should be in a currently non-existent CPU reset function. Signed-off-by: Peter Maydell <peter.maydell@linaro.org> --- target/alpha/cpu.c | 11 +++++++++++ fpu/softfloat-specialize.c.inc | 2 +- 2 files changed, 12 insertions(+), 1 deletion(-)