Re: [Mesa-dev] [PATCH v4 02/40] intel/compiler: add a NIR pass to lower conversions

2019-02-15 Thread Iago Toral 
On Thu, 2019-02-14 at 16:59 -0600, Jason Ekstrand wrote:
> On Tue, Feb 12, 2019 at 5:56 AM Iago Toral Quiroga  > wrote:
> > Some conversions are not directly supported in hardware and need to
> > be
> > 
> > split in two conversion instructions going through an intermediary
> > type.
> > 
> > Doing this at the NIR level simplifies a bit the complexity in the
> > backend.
> > 
> > 
> > 
> > v2:
> > 
> >  - Consider fp16 rounding conversion opcodes
> > 
> >  - Properly handle swizzles on conversion sources.
> > 
> > 
> > 
> > v3
> > 
> >  - Run the pass earlier, right after nir_opt_algebraic_late (Jason)
> > 
> >  - NIR alu output types already have the bit-size (Jason)
> > 
> >  - Use the new is_conversion field in nir_op_info to select
> > conversion
> > 
> >instructions (Jason)
> > 
> >  - Use 'is_conversion' to identify conversion operations (Jason)
> > 
> > 
> > 
> > Reviewed-by: Topi Pohjolainen  (v1)
> > 
> > ---
> > 
> >  src/intel/Makefile.sources|   1 +
> > 
> >  src/intel/compiler/brw_nir.c  |   2 +
> > 
> >  src/intel/compiler/brw_nir.h  |   2 +
> > 
> >  .../compiler/brw_nir_lower_conversions.c  | 158
> > ++
> > 
> >  src/intel/compiler/meson.build|   1 +
> > 
> >  5 files changed, 164 insertions(+)
> > 
> >  create mode 100644 src/intel/compiler/brw_nir_lower_conversions.c
> > 
> > 
> > 
> > diff --git a/src/intel/Makefile.sources
> > b/src/intel/Makefile.sources
> > 
> > index 94a28d370e8..9975daa3ad1 100644
> > 
> > --- a/src/intel/Makefile.sources
> > 
> > +++ b/src/intel/Makefile.sources
> > 
> > @@ -83,6 +83,7 @@ COMPILER_FILES = \
> > 
> > compiler/brw_nir_analyze_boolean_resolves.c \
> > 
> > compiler/brw_nir_analyze_ubo_ranges.c \
> > 
> > compiler/brw_nir_attribute_workarounds.c \
> > 
> > +   compiler/brw_nir_lower_conversions.c \
> > 
> > compiler/brw_nir_lower_cs_intrinsics.c \
> > 
> > compiler/brw_nir_lower_image_load_store.c \
> > 
> > compiler/brw_nir_lower_mem_access_bit_sizes.c \
> > 
> > diff --git a/src/intel/compiler/brw_nir.c
> > b/src/intel/compiler/brw_nir.c
> > 
> > index 9dbf06004a4..7e3dbc9e447 100644
> > 
> > --- a/src/intel/compiler/brw_nir.c
> > 
> > +++ b/src/intel/compiler/brw_nir.c
> > 
> > @@ -876,6 +876,8 @@ brw_postprocess_nir(nir_shader *nir, const
> > struct brw_compiler *compiler,
> > 
> > 
> > 
> > OPT(nir_opt_algebraic_late);
> > 
> > 
> > 
> > +   OPT(brw_nir_lower_conversions);
> > 
> > +
> > 
> > OPT(nir_lower_to_source_mods, nir_lower_all_source_mods);
> > 
> > OPT(nir_copy_prop);
> > 
> > OPT(nir_opt_dce);
> > 
> > diff --git a/src/intel/compiler/brw_nir.h
> > b/src/intel/compiler/brw_nir.h
> > 
> > index bc81950d47e..662b2627e95 100644
> > 
> > --- a/src/intel/compiler/brw_nir.h
> > 
> > +++ b/src/intel/compiler/brw_nir.h
> > 
> > @@ -114,6 +114,8 @@ void brw_nir_lower_tcs_outputs(nir_shader *nir,
> > const struct brw_vue_map *vue,
> > 
> > GLenum tes_primitive_mode);
> > 
> >  void brw_nir_lower_fs_outputs(nir_shader *nir);
> > 
> > 
> > 
> > +bool brw_nir_lower_conversions(nir_shader *nir);
> > 
> > +
> > 
> >  bool brw_nir_lower_image_load_store(nir_shader *nir,
> > 
> >  const struct gen_device_info
> > *devinfo);
> > 
> >  void brw_nir_rewrite_image_intrinsic(nir_intrinsic_instr *intrin,
> > 
> > diff --git a/src/intel/compiler/brw_nir_lower_conversions.c
> > b/src/intel/compiler/brw_nir_lower_conversions.c
> > 
> > new file mode 100644
> > 
> > index 000..bb1312ad428
> > 
> > --- /dev/null
> > 
> > +++ b/src/intel/compiler/brw_nir_lower_conversions.c
> > 
> > @@ -0,0 +1,158 @@
> > 
> > +/*
> > 
> > + * Copyright © 2018 Intel Corporation
> > 
> > + *
> > 
> > + * Permission is hereby granted, free of charge, to any person
> > obtaining a
> > 
> > + * copy of this software and associated documentation files (the
> > "Software"),
> > 
> > + * to deal in the Software without restriction, including without
> > limitation
> > 
> > + * the rights to use, copy, modify, merge, publish, distribute,
> > sublicense,
> > 
> > + * and/or sell copies of the Software, and to permit persons to
> > whom the
> > 
> > + * Software is furnished to do so, subject to the following
> > conditions:
> > 
> > + *
> > 
> > + * The above copyright notice and this permission notice
> > (including the next
> > 
> > + * paragraph) shall be included in all copies or substantial
> > portions of the
> > 
> > + * Software.
> > 
> > + *
> > 
> > + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
> > EXPRESS OR
> > 
> > + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
> > MERCHANTABILITY,
> > 
> > + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO
> > EVENT SHALL
> > 
> > + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM,
> > DAMAGES OR OTHER
> > 
> > + * LIABILITY, WHETHER IN AN ACTION OF CONTRA

Re: [Mesa-dev] [PATCH v4 02/40] intel/compiler: add a NIR pass to lower conversions

2019-02-15 Thread Jason Ekstrand 
On Tue, Feb 12, 2019 at 5:56 AM Iago Toral Quiroga 
wrote:

> Some conversions are not directly supported in hardware and need to be
> split in two conversion instructions going through an intermediary type.
> Doing this at the NIR level simplifies a bit the complexity in the backend.
>
> v2:
>  - Consider fp16 rounding conversion opcodes
>  - Properly handle swizzles on conversion sources.
>
> v3
>  - Run the pass earlier, right after nir_opt_algebraic_late (Jason)
>  - NIR alu output types already have the bit-size (Jason)
>  - Use the new is_conversion field in nir_op_info to select conversion
>instructions (Jason)
>  - Use 'is_conversion' to identify conversion operations (Jason)
>
> Reviewed-by: Topi Pohjolainen  (v1)
> ---
>  src/intel/Makefile.sources|   1 +
>  src/intel/compiler/brw_nir.c  |   2 +
>  src/intel/compiler/brw_nir.h  |   2 +
>  .../compiler/brw_nir_lower_conversions.c  | 158 ++
>  src/intel/compiler/meson.build|   1 +
>  5 files changed, 164 insertions(+)
>  create mode 100644 src/intel/compiler/brw_nir_lower_conversions.c
>
> diff --git a/src/intel/Makefile.sources b/src/intel/Makefile.sources
> index 94a28d370e8..9975daa3ad1 100644
> --- a/src/intel/Makefile.sources
> +++ b/src/intel/Makefile.sources
> @@ -83,6 +83,7 @@ COMPILER_FILES = \
> compiler/brw_nir_analyze_boolean_resolves.c \
> compiler/brw_nir_analyze_ubo_ranges.c \
> compiler/brw_nir_attribute_workarounds.c \
> +   compiler/brw_nir_lower_conversions.c \
> compiler/brw_nir_lower_cs_intrinsics.c \
> compiler/brw_nir_lower_image_load_store.c \
> compiler/brw_nir_lower_mem_access_bit_sizes.c \
> diff --git a/src/intel/compiler/brw_nir.c b/src/intel/compiler/brw_nir.c
> index 9dbf06004a4..7e3dbc9e447 100644
> --- a/src/intel/compiler/brw_nir.c
> +++ b/src/intel/compiler/brw_nir.c
> @@ -876,6 +876,8 @@ brw_postprocess_nir(nir_shader *nir, const struct
> brw_compiler *compiler,
>
> OPT(nir_opt_algebraic_late);
>
> +   OPT(brw_nir_lower_conversions);
> +
> OPT(nir_lower_to_source_mods, nir_lower_all_source_mods);
> OPT(nir_copy_prop);
> OPT(nir_opt_dce);
> diff --git a/src/intel/compiler/brw_nir.h b/src/intel/compiler/brw_nir.h
> index bc81950d47e..662b2627e95 100644
> --- a/src/intel/compiler/brw_nir.h
> +++ b/src/intel/compiler/brw_nir.h
> @@ -114,6 +114,8 @@ void brw_nir_lower_tcs_outputs(nir_shader *nir, const
> struct brw_vue_map *vue,
> GLenum tes_primitive_mode);
>  void brw_nir_lower_fs_outputs(nir_shader *nir);
>
> +bool brw_nir_lower_conversions(nir_shader *nir);
> +
>  bool brw_nir_lower_image_load_store(nir_shader *nir,
>  const struct gen_device_info
> *devinfo);
>  void brw_nir_rewrite_image_intrinsic(nir_intrinsic_instr *intrin,
> diff --git a/src/intel/compiler/brw_nir_lower_conversions.c
> b/src/intel/compiler/brw_nir_lower_conversions.c
> new file mode 100644
> index 000..bb1312ad428
> --- /dev/null
> +++ b/src/intel/compiler/brw_nir_lower_conversions.c
> @@ -0,0 +1,158 @@
> +/*
> + * Copyright © 2018 Intel Corporation
> + *
> + * Permission is hereby granted, free of charge, to any person obtaining a
> + * copy of this software and associated documentation files (the
> "Software"),
> + * to deal in the Software without restriction, including without
> limitation
> + * the rights to use, copy, modify, merge, publish, distribute,
> sublicense,
> + * and/or sell copies of the Software, and to permit persons to whom the
> + * Software is furnished to do so, subject to the following conditions:
> + *
> + * The above copyright notice and this permission notice (including the
> next
> + * paragraph) shall be included in all copies or substantial portions of
> the
> + * Software.
> + *
> + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
> EXPRESS OR
> + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
> MERCHANTABILITY,
> + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT
> SHALL
> + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
> OTHER
> + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
> + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
> DEALINGS
> + * IN THE SOFTWARE.
> + */
> +
> +#include "brw_nir.h"
> +#include "compiler/nir/nir_builder.h"
> +
> +static nir_op
> +get_conversion_op(nir_alu_type src_type,
> +  unsigned src_bit_size,
> +  nir_alu_type dst_type,
> +  unsigned dst_bit_size,
> +  nir_rounding_mode rounding_mode)
> +{
> +   nir_alu_type src_full_type = (nir_alu_type) (src_type | src_bit_size);
> +   nir_alu_type dst_full_type = (nir_alu_type) (dst_type | dst_bit_size);
> +
> +   return nir_type_conversion_op(src_full_type, dst_full_type,
> rounding_mode);
> +}
> +
> +static nir_rounding_mod

[Mesa-dev] [PATCH v4 02/40] intel/compiler: add a NIR pass to lower conversions

2019-02-12 Thread Iago Toral Quiroga 
Some conversions are not directly supported in hardware and need to be
split in two conversion instructions going through an intermediary type.
Doing this at the NIR level simplifies a bit the complexity in the backend.

v2:
 - Consider fp16 rounding conversion opcodes
 - Properly handle swizzles on conversion sources.

v3
 - Run the pass earlier, right after nir_opt_algebraic_late (Jason)
 - NIR alu output types already have the bit-size (Jason)
 - Use the new is_conversion field in nir_op_info to select conversion
   instructions (Jason)
 - Use 'is_conversion' to identify conversion operations (Jason)

Reviewed-by: Topi Pohjolainen  (v1)
---
 src/intel/Makefile.sources|   1 +
 src/intel/compiler/brw_nir.c  |   2 +
 src/intel/compiler/brw_nir.h  |   2 +
 .../compiler/brw_nir_lower_conversions.c  | 158 ++
 src/intel/compiler/meson.build|   1 +
 5 files changed, 164 insertions(+)
 create mode 100644 src/intel/compiler/brw_nir_lower_conversions.c

diff --git a/src/intel/Makefile.sources b/src/intel/Makefile.sources
index 94a28d370e8..9975daa3ad1 100644
--- a/src/intel/Makefile.sources
+++ b/src/intel/Makefile.sources
@@ -83,6 +83,7 @@ COMPILER_FILES = \
compiler/brw_nir_analyze_boolean_resolves.c \
compiler/brw_nir_analyze_ubo_ranges.c \
compiler/brw_nir_attribute_workarounds.c \
+   compiler/brw_nir_lower_conversions.c \
compiler/brw_nir_lower_cs_intrinsics.c \
compiler/brw_nir_lower_image_load_store.c \
compiler/brw_nir_lower_mem_access_bit_sizes.c \
diff --git a/src/intel/compiler/brw_nir.c b/src/intel/compiler/brw_nir.c
index 9dbf06004a4..7e3dbc9e447 100644
--- a/src/intel/compiler/brw_nir.c
+++ b/src/intel/compiler/brw_nir.c
@@ -876,6 +876,8 @@ brw_postprocess_nir(nir_shader *nir, const struct 
brw_compiler *compiler,
 
OPT(nir_opt_algebraic_late);
 
+   OPT(brw_nir_lower_conversions);
+
OPT(nir_lower_to_source_mods, nir_lower_all_source_mods);
OPT(nir_copy_prop);
OPT(nir_opt_dce);
diff --git a/src/intel/compiler/brw_nir.h b/src/intel/compiler/brw_nir.h
index bc81950d47e..662b2627e95 100644
--- a/src/intel/compiler/brw_nir.h
+++ b/src/intel/compiler/brw_nir.h
@@ -114,6 +114,8 @@ void brw_nir_lower_tcs_outputs(nir_shader *nir, const 
struct brw_vue_map *vue,
GLenum tes_primitive_mode);
 void brw_nir_lower_fs_outputs(nir_shader *nir);
 
+bool brw_nir_lower_conversions(nir_shader *nir);
+
 bool brw_nir_lower_image_load_store(nir_shader *nir,
 const struct gen_device_info *devinfo);
 void brw_nir_rewrite_image_intrinsic(nir_intrinsic_instr *intrin,
diff --git a/src/intel/compiler/brw_nir_lower_conversions.c 
b/src/intel/compiler/brw_nir_lower_conversions.c
new file mode 100644
index 000..bb1312ad428
--- /dev/null
+++ b/src/intel/compiler/brw_nir_lower_conversions.c
@@ -0,0 +1,158 @@
+/*
+ * Copyright © 2018 Intel Corporation
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+ * IN THE SOFTWARE.
+ */
+
+#include "brw_nir.h"
+#include "compiler/nir/nir_builder.h"
+
+static nir_op
+get_conversion_op(nir_alu_type src_type,
+  unsigned src_bit_size,
+  nir_alu_type dst_type,
+  unsigned dst_bit_size,
+  nir_rounding_mode rounding_mode)
+{
+   nir_alu_type src_full_type = (nir_alu_type) (src_type | src_bit_size);
+   nir_alu_type dst_full_type = (nir_alu_type) (dst_type | dst_bit_size);
+
+   return nir_type_conversion_op(src_full_type, dst_full_type, rounding_mode);
+}
+
+static nir_rounding_mode
+get_opcode_rounding_mode(nir_op op)
+{
+   switch (op) {
+   case nir_op_f2f16_rtz:
+  return nir_rounding_mode_rtz;
+   case nir_op_f2f16_rtne:
+  return nir_rounding_mode_rtne;
+   default:
+  return nir_rounding_mode_undef;
+   }
+}
+
+static void
+split_conversion(nir_builder