[Mesa-dev] [PATCH v2 1/8] nir: add builtin builder
Karol Herbst
kherbst at redhat.com
Mon Jul 16 14:28:19 UTC 2018
also move some of the GLSL builtins over we will need for implementing
some OpenCL builtins
v2: replace NIR_IMM_FP by nir_imm_floatN_t in ported code
fix up changes caused by swizzle rework
Signed-off-by: Karol Herbst <kherbst at redhat.com>
---
src/compiler/Makefile.sources | 2 +
src/compiler/nir/meson.build | 2 +
src/compiler/nir/nir_builtin_builder.c | 65 +++++++++++++++++++++
src/compiler/nir/nir_builtin_builder.h | 79 ++++++++++++++++++++++++++
src/compiler/spirv/vtn_glsl450.c | 58 +++++--------------
5 files changed, 161 insertions(+), 45 deletions(-)
create mode 100644 src/compiler/nir/nir_builtin_builder.c
create mode 100644 src/compiler/nir/nir_builtin_builder.h
diff --git a/src/compiler/Makefile.sources b/src/compiler/Makefile.sources
index db4dd1e89f4..cc147218c4e 100644
--- a/src/compiler/Makefile.sources
+++ b/src/compiler/Makefile.sources
@@ -203,6 +203,8 @@ NIR_FILES = \
nir/nir.c \
nir/nir.h \
nir/nir_builder.h \
+ nir/nir_builtin_builder.c \
+ nir/nir_builtin_builder.h \
nir/nir_clone.c \
nir/nir_constant_expressions.h \
nir/nir_control_flow.c \
diff --git a/src/compiler/nir/meson.build b/src/compiler/nir/meson.build
index 28aa8de7014..a1bb19356ce 100644
--- a/src/compiler/nir/meson.build
+++ b/src/compiler/nir/meson.build
@@ -87,6 +87,8 @@ files_libnir = files(
'nir.c',
'nir.h',
'nir_builder.h',
+ 'nir_builtin_builder.c',
+ 'nir_builtin_builder.h',
'nir_clone.c',
'nir_constant_expressions.h',
'nir_control_flow.c',
diff --git a/src/compiler/nir/nir_builtin_builder.c b/src/compiler/nir/nir_builtin_builder.c
new file mode 100644
index 00000000000..252a7691f36
--- /dev/null
+++ b/src/compiler/nir/nir_builtin_builder.c
@@ -0,0 +1,65 @@
+/*
+ * Copyright © 2018 Red Hat Inc.
+ *
+ * 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 "nir.h"
+#include "nir_builtin_builder.h"
+
+nir_ssa_def*
+nir_cross(nir_builder *b, nir_ssa_def *x, nir_ssa_def *y)
+{
+ unsigned yzx[3] = { 1, 2, 0 };
+ unsigned zxy[3] = { 2, 0, 1 };
+
+ return nir_fsub(b, nir_fmul(b, nir_swizzle(b, x, yzx, 3, true),
+ nir_swizzle(b, y, zxy, 3, true)),
+ nir_fmul(b, nir_swizzle(b, x, zxy, 3, true),
+ nir_swizzle(b, y, yzx, 3, true)));
+}
+
+nir_ssa_def*
+nir_fast_length(nir_builder *b, nir_ssa_def *vec)
+{
+ switch (vec->num_components) {
+ case 1: return nir_fsqrt(b, nir_fmul(b, vec, vec));
+ case 2: return nir_fsqrt(b, nir_fdot2(b, vec, vec));
+ case 3: return nir_fsqrt(b, nir_fdot3(b, vec, vec));
+ case 4: return nir_fsqrt(b, nir_fdot4(b, vec, vec));
+ default:
+ unreachable("Invalid number of components");
+ }
+}
+
+nir_ssa_def*
+nir_smoothstep(nir_builder *b, nir_ssa_def *edge0, nir_ssa_def *edge1, nir_ssa_def *x)
+{
+ nir_ssa_def *f2 = nir_imm_floatN_t(b, 2.0, x->bit_size);
+ nir_ssa_def *f3 = nir_imm_floatN_t(b, 3.0, x->bit_size);
+
+ /* t = clamp((x - edge0) / (edge1 - edge0), 0, 1) */
+ nir_ssa_def *t =
+ nir_fsat(b, nir_fdiv(b, nir_fsub(b, x, edge0),
+ nir_fsub(b, edge1, edge0)));
+
+ /* result = t * t * (3 - 2 * t) */
+ return nir_fmul(b, t, nir_fmul(b, t, nir_fsub(b, f3, nir_fmul(b, f2, t))));
+}
diff --git a/src/compiler/nir/nir_builtin_builder.h b/src/compiler/nir/nir_builtin_builder.h
new file mode 100644
index 00000000000..5b76d012c85
--- /dev/null
+++ b/src/compiler/nir/nir_builtin_builder.h
@@ -0,0 +1,79 @@
+/*
+ * Copyright © 2018 Red Hat Inc.
+ *
+ * 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.
+ */
+
+#ifndef NIR_BUILTIN_BUILDER_H
+#define NIR_BUILTIN_BUILDER_H
+
+#include "nir/nir_builder.h"
+
+nir_ssa_def* nir_cross(nir_builder *b, nir_ssa_def *x, nir_ssa_def *y);
+nir_ssa_def* nir_fast_length(nir_builder *b, nir_ssa_def *vec);
+nir_ssa_def* nir_smoothstep(nir_builder *b, nir_ssa_def *edge0,
+ nir_ssa_def *edge1, nir_ssa_def *x);
+
+static inline nir_ssa_def *
+nir_fclamp(nir_builder *b,
+ nir_ssa_def *x, nir_ssa_def *min_val, nir_ssa_def *max_val)
+{
+ return nir_fmin(b, nir_fmax(b, x, min_val), max_val);
+}
+
+static inline nir_ssa_def *
+nir_iclamp(nir_builder *b,
+ nir_ssa_def *x, nir_ssa_def *min_val, nir_ssa_def *max_val)
+{
+ return nir_imin(b, nir_imax(b, x, min_val), max_val);
+}
+
+static inline nir_ssa_def *
+nir_uclamp(nir_builder *b,
+ nir_ssa_def *x, nir_ssa_def *min_val, nir_ssa_def *max_val)
+{
+ return nir_umin(b, nir_umax(b, x, min_val), max_val);
+}
+
+static inline nir_ssa_def *
+nir_degrees(nir_builder *b, nir_ssa_def *val)
+{
+ return nir_fmul(b, val, nir_imm_float(b, 57.2957795131));
+}
+
+static inline nir_ssa_def *
+nir_fast_distance(nir_builder *b, nir_ssa_def *x, nir_ssa_def *y)
+{
+ return nir_fast_length(b, nir_fsub(b, x, y));
+}
+
+static inline nir_ssa_def*
+nir_fast_normalize(nir_builder *b, nir_ssa_def *vec)
+{
+ return nir_fdiv(b, vec, nir_fast_length(b, vec));
+}
+
+static inline nir_ssa_def *
+nir_radians(nir_builder *b, nir_ssa_def *val)
+{
+ return nir_fmul(b, val, nir_imm_float(b, 0.01745329251));
+}
+
+#endif /* NIR_BUILTIN_BUILDER_H */
diff --git a/src/compiler/spirv/vtn_glsl450.c b/src/compiler/spirv/vtn_glsl450.c
index 9ef066e8e6d..06a49e48e3f 100644
--- a/src/compiler/spirv/vtn_glsl450.c
+++ b/src/compiler/spirv/vtn_glsl450.c
@@ -26,6 +26,9 @@
*/
#include <math.h>
+
+#include "nir/nir_builtin_builder.h"
+
#include "vtn_private.h"
#include "GLSL.std.450.h"
@@ -168,26 +171,6 @@ matrix_inverse(struct vtn_builder *b, struct vtn_ssa_value *src)
return val;
}
-static nir_ssa_def*
-build_length(nir_builder *b, nir_ssa_def *vec)
-{
- switch (vec->num_components) {
- case 1: return nir_fsqrt(b, nir_fmul(b, vec, vec));
- case 2: return nir_fsqrt(b, nir_fdot2(b, vec, vec));
- case 3: return nir_fsqrt(b, nir_fdot3(b, vec, vec));
- case 4: return nir_fsqrt(b, nir_fdot4(b, vec, vec));
- default:
- unreachable("Invalid number of components");
- }
-}
-
-static inline nir_ssa_def *
-build_fclamp(nir_builder *b,
- nir_ssa_def *x, nir_ssa_def *min_val, nir_ssa_def *max_val)
-{
- return nir_fmin(b, nir_fmax(b, x, min_val), max_val);
-}
-
/**
* Return e^x.
*/
@@ -540,10 +523,10 @@ handle_glsl450_alu(struct vtn_builder *b, enum GLSLstd450 entrypoint,
switch (entrypoint) {
case GLSLstd450Radians:
- val->ssa->def = nir_fmul(nb, src[0], nir_imm_float(nb, 0.01745329251));
+ val->ssa->def = nir_radians(nb, src[0]);
return;
case GLSLstd450Degrees:
- val->ssa->def = nir_fmul(nb, src[0], nir_imm_float(nb, 57.2957795131));
+ val->ssa->def = nir_degrees(nb, src[0]);
return;
case GLSLstd450Tan:
val->ssa->def = nir_fdiv(nb, nir_fsin(nb, src[0]),
@@ -573,13 +556,13 @@ handle_glsl450_alu(struct vtn_builder *b, enum GLSLstd450 entrypoint,
return;
case GLSLstd450Length:
- val->ssa->def = build_length(nb, src[0]);
+ val->ssa->def = nir_fast_length(nb, src[0]);
return;
case GLSLstd450Distance:
- val->ssa->def = build_length(nb, nir_fsub(nb, src[0], src[1]));
+ val->ssa->def = nir_fast_distance(nb, src[0], src[1]);
return;
case GLSLstd450Normalize:
- val->ssa->def = nir_fdiv(nb, src[0], build_length(nb, src[0]));
+ val->ssa->def = nir_fast_normalize(nb, src[0]);
return;
case GLSLstd450Exp:
@@ -592,37 +575,22 @@ handle_glsl450_alu(struct vtn_builder *b, enum GLSLstd450 entrypoint,
case GLSLstd450FClamp:
case GLSLstd450NClamp:
- val->ssa->def = build_fclamp(nb, src[0], src[1], src[2]);
+ val->ssa->def = nir_fclamp(nb, src[0], src[1], src[2]);
return;
case GLSLstd450UClamp:
- val->ssa->def = nir_umin(nb, nir_umax(nb, src[0], src[1]), src[2]);
+ val->ssa->def = nir_uclamp(nb, src[0], src[1], src[2]);
return;
case GLSLstd450SClamp:
- val->ssa->def = nir_imin(nb, nir_imax(nb, src[0], src[1]), src[2]);
+ val->ssa->def = nir_iclamp(nb, src[0], src[1], src[2]);
return;
case GLSLstd450Cross: {
- unsigned yzx[3] = { 1, 2, 0 };
- unsigned zxy[3] = { 2, 0, 1 };
- val->ssa->def =
- nir_fsub(nb, nir_fmul(nb, nir_swizzle(nb, src[0], yzx, 3, true),
- nir_swizzle(nb, src[1], zxy, 3, true)),
- nir_fmul(nb, nir_swizzle(nb, src[0], zxy, 3, true),
- nir_swizzle(nb, src[1], yzx, 3, true)));
+ val->ssa->def = nir_cross(nb, src[0], src[1]);
return;
}
case GLSLstd450SmoothStep: {
- /* t = clamp((x - edge0) / (edge1 - edge0), 0, 1) */
- nir_ssa_def *t =
- build_fclamp(nb, nir_fdiv(nb, nir_fsub(nb, src[2], src[0]),
- nir_fsub(nb, src[1], src[0])),
- NIR_IMM_FP(nb, 0.0), NIR_IMM_FP(nb, 1.0));
- /* result = t * t * (3 - 2 * t) */
- val->ssa->def =
- nir_fmul(nb, t, nir_fmul(nb, t,
- nir_fsub(nb, NIR_IMM_FP(nb, 3.0),
- nir_fmul(nb, NIR_IMM_FP(nb, 2.0), t))));
+ val->ssa->def = nir_smoothstep(nb, src[0], src[1], src[2]);
return;
}
--
2.17.1
More information about the mesa-dev
mailing list