On Monday, 2018-03-05 14:07:24 -0500, Jan Vesely wrote: > On Mon, 2018-03-05 at 10:04 +0000, Eric Engestrom wrote: > > On Saturday, 2018-03-03 18:35:02 -0500, Jan Vesely wrote: > > > Neither GL nor CL nor SPIRV needs the IEEE handling of sNaNs. > > > > > > Signed-off-by: Jan Vesely <jan.ves...@rutgers.edu> > > > --- > > > This is the 3rd way to handle broken fmin/fmax in clover CL. > > > It can be worked around in libclc (to not use > > > llvm.fminnum/llvm.fmaxnum)[0] > > > LLVM can be patched to not use IEEE_MODE[1] > > > Or it can be fixed in the config reg in mesa. > > > > > > The third option is arguably the worst (changes semantics of compiled > > > code), but since the first two efforts have been stalled for months. > > > Here's the third one. > > > > > > Jan > > > > > > [0] > > > http://lists.llvm.org/pipermail/libclc-dev/2017-November/thread.html#2692 > > > [1] https://reviews.llvm.org/D40514 > > > > > > > > > src/gallium/drivers/radeonsi/si_compute.c | 8 ++++++++ > > > src/gallium/drivers/radeonsi/si_shader.c | 6 +++++- > > > 2 files changed, 13 insertions(+), 1 deletion(-) > > > > > > diff --git a/src/gallium/drivers/radeonsi/si_compute.c > > > b/src/gallium/drivers/radeonsi/si_compute.c > > > index 41927988cf..d0fcba485b 100644 > > > --- a/src/gallium/drivers/radeonsi/si_compute.c > > > +++ b/src/gallium/drivers/radeonsi/si_compute.c > > > @@ -70,6 +70,10 @@ static void code_object_to_config(const > > > amd_kernel_code_t *code_object, > > > struct si_shader_config *out_config) { > > > > > > uint32_t rsrc1 = code_object->compute_pgm_resource_registers; > > > + /* Disable IEEE mode. IEEE_MODE determines sNaN handling by fmin/fmax > > > + * instructions. Nothing wants the IEEE rules here. > > > + * Ideally LLVM would not emit this configuration. */ > > > + rsrc1 &= C_00B848_IEEE_MODE; > > > > I don't know anything about this code, but this bit op looks odd to me; > > I would expect to see either: > > rsrc1 &= ~C_00B848_IEEE_MODE; > > C_00B848_IEEE_MODE is defined as 0xFF7FFFFF, so it's already bit- > negated.
Alright, looks reasonable then :) > > Jan > > > or: > > rsrc1 |= C_00B848_IEEE_MODE; > > > > > uint32_t rsrc2 = code_object->compute_pgm_resource_registers >> 32; > > > out_config->num_sgprs = code_object->wavefront_sgpr_count; > > > out_config->num_vgprs = code_object->workitem_vgpr_count; > > > @@ -137,6 +141,10 @@ static void si_create_compute_state_async(void *job, > > > int thread_index) > > > S_00B848_SGPRS((shader->config.num_sgprs - 1) / 8) | > > > S_00B848_DX10_CLAMP(1) | > > > S_00B848_FLOAT_MODE(shader->config.float_mode); > > > + /* Disable IEEE mode. IEEE_MODE determines sNaN handling by > > > + * fmin/fmax instructions. Nothing wants the IEEE rules. > > > + * Ideally LLVM would not emit this configuration. */ > > > + shader->config.rsrc1 &= C_00B848_IEEE_MODE; > > > > > > shader->config.rsrc2 = > > > S_00B84C_USER_SGPR(user_sgprs) | > > > diff --git a/src/gallium/drivers/radeonsi/si_shader.c > > > b/src/gallium/drivers/radeonsi/si_shader.c > > > index 2a50b266f6..01bad71a4e 100644 > > > --- a/src/gallium/drivers/radeonsi/si_shader.c > > > +++ b/src/gallium/drivers/radeonsi/si_shader.c > > > @@ -5178,7 +5178,11 @@ void si_shader_binary_read_config(struct > > > ac_shader_binary *binary, > > > conf->num_sgprs = MAX2(conf->num_sgprs, > > > (G_00B028_SGPRS(value) + 1) * 8); > > > conf->num_vgprs = MAX2(conf->num_vgprs, > > > (G_00B028_VGPRS(value) + 1) * 4); > > > conf->float_mode = G_00B028_FLOAT_MODE(value); > > > - conf->rsrc1 = value; > > > + /* Disable IEEE mode. IEEE_MODE determines sNaN > > > + * handling by fmin/fmax instructions. > > > + * Nothing wants the IEEE rules here. > > > + * Ideally LLVM would not emit this configuration. */ > > > + conf->rsrc1 = value & C_00B848_IEEE_MODE; > > > break; > > > case R_00B02C_SPI_SHADER_PGM_RSRC2_PS: > > > conf->lds_size = MAX2(conf->lds_size, > > > G_00B02C_EXTRA_LDS_SIZE(value)); > > > -- > > > 2.14.3 > > > > > > _______________________________________________ > > > mesa-dev mailing list > > > mesa-dev@lists.freedesktop.org > > > https://lists.freedesktop.org/mailman/listinfo/mesa-dev _______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev
