aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorErik Faye-Lund <[email protected]>2019-10-02 16:19:08 -0400
committerErik Faye-Lund <[email protected]>2019-10-17 10:41:36 +0200
commiteb3047c094abfa03e071453d7c373e9c2c574370 (patch)
tree2be3fea726a40dc3334160ce8b72e83eb66439f0
parent28543f1640d459404413c244740291fe68a75a5c (diff)
nir: support lowering clipdist to arrays
This allows us to make sure clipdist is emitted as a scalar array rather than two vec4s. This matches SPIR-V semantics, and will be useful for Zink. Reviewed-by: Marek Olšák <[email protected]>
-rw-r--r--src/broadcom/compiler/vir.c5
-rw-r--r--src/compiler/nir/nir.h10
-rw-r--r--src/compiler/nir/nir_lower_clip.c99
-rw-r--r--src/freedreno/ir3/ir3_nir.c4
-rw-r--r--src/gallium/drivers/iris/iris_program.c8
-rw-r--r--src/gallium/drivers/vc4/vc4_program.c5
-rw-r--r--src/mesa/drivers/dri/i965/brw_nir_uniforms.cpp2
7 files changed, 89 insertions, 44 deletions
diff --git a/src/broadcom/compiler/vir.c b/src/broadcom/compiler/vir.c
index 5d929c2412a..afba863b0ec 100644
--- a/src/broadcom/compiler/vir.c
+++ b/src/broadcom/compiler/vir.c
@@ -825,7 +825,7 @@ v3d_nir_lower_vs_late(struct v3d_compile *c)
if (c->key->ucp_enables) {
NIR_PASS_V(c->s, nir_lower_clip_vs, c->key->ucp_enables,
- false);
+ false, false);
NIR_PASS_V(c->s, nir_lower_io_to_scalar,
nir_var_shader_out);
}
@@ -850,7 +850,8 @@ v3d_nir_lower_fs_late(struct v3d_compile *c)
}
if (c->key->ucp_enables)
- NIR_PASS_V(c->s, nir_lower_clip_fs, c->key->ucp_enables);
+ NIR_PASS_V(c->s, nir_lower_clip_fs, c->key->ucp_enables,
+ false);
/* Note: FS input scalarizing must happen after
* nir_lower_two_sided_color, which only handles a vec4 at a time.
diff --git a/src/compiler/nir/nir.h b/src/compiler/nir/nir.h
index 77c877d6a79..a1f233917bd 100644
--- a/src/compiler/nir/nir.h
+++ b/src/compiler/nir/nir.h
@@ -3906,9 +3906,13 @@ bool nir_lower_idiv(nir_shader *shader);
bool nir_lower_input_attachments(nir_shader *shader, bool use_fragcoord_sysval);
-bool nir_lower_clip_vs(nir_shader *shader, unsigned ucp_enables, bool use_vars);
-bool nir_lower_clip_gs(nir_shader *shader, unsigned ucp_enables);
-bool nir_lower_clip_fs(nir_shader *shader, unsigned ucp_enables);
+bool nir_lower_clip_vs(nir_shader *shader, unsigned ucp_enables,
+ bool use_vars,
+ bool use_clipdist_array);
+bool nir_lower_clip_gs(nir_shader *shader, unsigned ucp_enables,
+ bool use_clipdist_array);
+bool nir_lower_clip_fs(nir_shader *shader, unsigned ucp_enables,
+ bool use_clipdist_array);
bool nir_lower_clip_cull_distance_arrays(nir_shader *nir);
void nir_lower_point_size_mov(nir_shader *shader,
diff --git a/src/compiler/nir/nir_lower_clip.c b/src/compiler/nir/nir_lower_clip.c
index 26c3e64856b..0da36d79ba0 100644
--- a/src/compiler/nir/nir_lower_clip.c
+++ b/src/compiler/nir/nir_lower_clip.c
@@ -42,17 +42,23 @@
static nir_variable *
create_clipdist_var(nir_shader *shader, unsigned drvloc,
- bool output, gl_varying_slot slot)
+ bool output, gl_varying_slot slot, unsigned array_size)
{
nir_variable *var = rzalloc(shader, nir_variable);
var->data.driver_location = drvloc;
- var->type = glsl_vec4_type();
var->data.mode = output ? nir_var_shader_out : nir_var_shader_in;
var->name = ralloc_asprintf(var, "clipdist_%d", drvloc);
var->data.index = 0;
var->data.location = slot;
+ if (array_size > 0) {
+ var->type = glsl_array_type(glsl_float_type(), array_size,
+ sizeof(float));
+ var->data.compact = 1;
+ } else
+ var->type = glsl_vec4_type();
+
if (output) {
exec_list_push_tail(&shader->outputs, &var->node);
shader->num_outputs++; /* TODO use type_size() */
@@ -66,16 +72,24 @@ create_clipdist_var(nir_shader *shader, unsigned drvloc,
static void
create_clipdist_vars(nir_shader *shader, nir_variable **io_vars,
- unsigned ucp_enables, int *drvloc, bool output)
+ unsigned ucp_enables, int *drvloc, bool output,
+ bool use_clipdist_array)
{
- if (ucp_enables & 0x0f)
+ if (use_clipdist_array) {
io_vars[0] =
- create_clipdist_var(shader, ++(*drvloc), output,
- VARYING_SLOT_CLIP_DIST0);
- if (ucp_enables & 0xf0)
- io_vars[1] =
- create_clipdist_var(shader, ++(*drvloc), output,
- VARYING_SLOT_CLIP_DIST1);
+ create_clipdist_var(shader, ++(*drvloc), true,
+ VARYING_SLOT_CLIP_DIST0,
+ util_last_bit(ucp_enables));
+ } else {
+ if (ucp_enables & 0x0f)
+ io_vars[0] =
+ create_clipdist_var(shader, ++(*drvloc), output,
+ VARYING_SLOT_CLIP_DIST0, 0);
+ if (ucp_enables & 0xf0)
+ io_vars[1] =
+ create_clipdist_var(shader, ++(*drvloc), output,
+ VARYING_SLOT_CLIP_DIST1, 0);
+ }
}
static void
@@ -192,7 +206,8 @@ find_clipvertex_and_position_outputs(nir_shader *shader,
static void
lower_clip_outputs(nir_builder *b, nir_variable *position,
nir_variable *clipvertex, nir_variable **out,
- unsigned ucp_enables, bool use_vars)
+ unsigned ucp_enables, bool use_vars,
+ bool use_clipdist_array)
{
nir_ssa_def *clipdist[MAX_CLIP_PLANES];
nir_ssa_def *cv;
@@ -224,18 +239,28 @@ lower_clip_outputs(nir_builder *b, nir_variable *position,
/* 0.0 == don't-clip == disabled: */
clipdist[plane] = nir_imm_float(b, 0.0);
}
+ if (use_clipdist_array && plane < util_last_bit(ucp_enables)) {
+ assert(use_vars);
+ nir_deref_instr *deref;
+ deref = nir_build_deref_array_imm(b,
+ nir_build_deref_var(b, out[0]),
+ plane);
+ nir_store_deref(b, deref, clipdist[plane], 1);
+ }
}
- if (use_vars) {
- if (ucp_enables & 0x0f)
- nir_store_var(b, out[0], nir_vec(b, clipdist, 4), 0xf);
- if (ucp_enables & 0xf0)
- nir_store_var(b, out[1], nir_vec(b, &clipdist[4], 4), 0xf);
- } else {
- if (ucp_enables & 0x0f)
- store_clipdist_output(b, out[0], &clipdist[0]);
- if (ucp_enables & 0xf0)
- store_clipdist_output(b, out[1], &clipdist[4]);
+ if (!use_clipdist_array) {
+ if (use_vars) {
+ if (ucp_enables & 0x0f)
+ nir_store_var(b, out[0], nir_vec(b, clipdist, 4), 0xf);
+ if (ucp_enables & 0xf0)
+ nir_store_var(b, out[1], nir_vec(b, &clipdist[4], 4), 0xf);
+ } else {
+ if (ucp_enables & 0x0f)
+ store_clipdist_output(b, out[0], &clipdist[0]);
+ if (ucp_enables & 0xf0)
+ store_clipdist_output(b, out[1], &clipdist[4]);
+ }
}
}
@@ -248,9 +273,13 @@ lower_clip_outputs(nir_builder *b, nir_variable *position,
*
* If use_vars is true, the pass will use variable loads and stores instead
* of working with store_output intrinsics.
+ *
+ * If use_clipdist_array is true, the pass will use compact arrays for the
+ * clipdist output instead of two vec4s.
*/
bool
-nir_lower_clip_vs(nir_shader *shader, unsigned ucp_enables, bool use_vars)
+nir_lower_clip_vs(nir_shader *shader, unsigned ucp_enables, bool use_vars,
+ bool use_clipdist_array)
{
nir_function_impl *impl = nir_shader_get_entrypoint(shader);
nir_builder b;
@@ -281,9 +310,11 @@ nir_lower_clip_vs(nir_shader *shader, unsigned ucp_enables, bool use_vars)
return false;
/* insert CLIPDIST outputs */
- create_clipdist_vars(shader, out, ucp_enables, &maxloc, true);
+ create_clipdist_vars(shader, out, ucp_enables, &maxloc, true,
+ use_clipdist_array);
- lower_clip_outputs(&b, position, clipvertex, out, ucp_enables, use_vars);
+ lower_clip_outputs(&b, position, clipvertex, out, ucp_enables, use_vars,
+ use_clipdist_array);
nir_metadata_preserve(impl, nir_metadata_dominance);
@@ -293,7 +324,7 @@ nir_lower_clip_vs(nir_shader *shader, unsigned ucp_enables, bool use_vars)
static void
lower_clip_in_gs_block(nir_builder *b, nir_block *block, nir_variable *position,
nir_variable *clipvertex, nir_variable **out,
- unsigned ucp_enables)
+ unsigned ucp_enables, bool use_clipdist_array)
{
nir_foreach_instr_safe(instr, block) {
if (instr->type != nir_instr_type_intrinsic)
@@ -304,7 +335,8 @@ lower_clip_in_gs_block(nir_builder *b, nir_block *block, nir_variable *position,
case nir_intrinsic_emit_vertex_with_counter:
case nir_intrinsic_emit_vertex:
b->cursor = nir_before_instr(instr);
- lower_clip_outputs(b, position, clipvertex, out, ucp_enables, true);
+ lower_clip_outputs(b, position, clipvertex, out, ucp_enables, true,
+ use_clipdist_array);
break;
default:
/* not interesting; skip this */
@@ -318,7 +350,8 @@ lower_clip_in_gs_block(nir_builder *b, nir_block *block, nir_variable *position,
*/
bool
-nir_lower_clip_gs(nir_shader *shader, unsigned ucp_enables)
+nir_lower_clip_gs(nir_shader *shader, unsigned ucp_enables,
+ bool use_clipdist_array)
{
nir_function_impl *impl = nir_shader_get_entrypoint(shader);
nir_builder b;
@@ -335,12 +368,14 @@ nir_lower_clip_gs(nir_shader *shader, unsigned ucp_enables)
return false;
/* insert CLIPDIST outputs */
- create_clipdist_vars(shader, out, ucp_enables, &maxloc, true);
+ create_clipdist_vars(shader, out, ucp_enables, &maxloc, true,
+ use_clipdist_array);
nir_builder_init(&b, impl);
nir_foreach_block(block, impl)
- lower_clip_in_gs_block(&b, block, position, clipvertex, out, ucp_enables);
+ lower_clip_in_gs_block(&b, block, position, clipvertex, out,
+ ucp_enables, use_clipdist_array);
nir_metadata_preserve(impl, nir_metadata_dominance);
@@ -388,7 +423,8 @@ lower_clip_fs(nir_function_impl *impl, unsigned ucp_enables,
/* insert conditional kill based on interpolated CLIPDIST
*/
bool
-nir_lower_clip_fs(nir_shader *shader, unsigned ucp_enables)
+nir_lower_clip_fs(nir_shader *shader, unsigned ucp_enables,
+ bool use_clipdist_array)
{
nir_variable *in[2];
int maxloc = -1;
@@ -410,7 +446,8 @@ nir_lower_clip_fs(nir_shader *shader, unsigned ucp_enables)
* must add our own:
*/
/* insert CLIPDIST inputs */
- create_clipdist_vars(shader, in, ucp_enables, &maxloc, false);
+ create_clipdist_vars(shader, in, ucp_enables, &maxloc, false,
+ use_clipdist_array);
nir_foreach_function(function, shader) {
if (!strcmp(function->name, "main"))
diff --git a/src/freedreno/ir3/ir3_nir.c b/src/freedreno/ir3/ir3_nir.c
index b729692bdff..dbf25926ac5 100644
--- a/src/freedreno/ir3/ir3_nir.c
+++ b/src/freedreno/ir3/ir3_nir.c
@@ -223,11 +223,11 @@ ir3_optimize_nir(struct ir3_shader *shader, nir_shader *s,
if (key) {
if (s->info.stage == MESA_SHADER_VERTEX) {
- OPT_V(s, nir_lower_clip_vs, key->ucp_enables, false);
+ OPT_V(s, nir_lower_clip_vs, key->ucp_enables, false, false);
if (key->vclamp_color)
OPT_V(s, nir_lower_clamp_color_outputs);
} else if (s->info.stage == MESA_SHADER_FRAGMENT) {
- OPT_V(s, nir_lower_clip_fs, key->ucp_enables);
+ OPT_V(s, nir_lower_clip_fs, key->ucp_enables, false);
if (key->fclamp_color)
OPT_V(s, nir_lower_clamp_color_outputs);
}
diff --git a/src/gallium/drivers/iris/iris_program.c b/src/gallium/drivers/iris/iris_program.c
index fbf6d685382..fc4f2d2949d 100644
--- a/src/gallium/drivers/iris/iris_program.c
+++ b/src/gallium/drivers/iris/iris_program.c
@@ -938,7 +938,8 @@ iris_compile_vs(struct iris_context *ice,
if (key->nr_userclip_plane_consts) {
nir_function_impl *impl = nir_shader_get_entrypoint(nir);
- nir_lower_clip_vs(nir, (1 << key->nr_userclip_plane_consts) - 1, true);
+ nir_lower_clip_vs(nir, (1 << key->nr_userclip_plane_consts) - 1, true,
+ false);
nir_lower_io_to_temporaries(nir, impl, true, false);
nir_lower_global_vars_to_local(nir);
nir_lower_vars_to_ssa(nir);
@@ -1281,7 +1282,8 @@ iris_compile_tes(struct iris_context *ice,
if (key->nr_userclip_plane_consts) {
nir_function_impl *impl = nir_shader_get_entrypoint(nir);
- nir_lower_clip_vs(nir, (1 << key->nr_userclip_plane_consts) - 1, true);
+ nir_lower_clip_vs(nir, (1 << key->nr_userclip_plane_consts) - 1, true,
+ false);
nir_lower_io_to_temporaries(nir, impl, true, false);
nir_lower_global_vars_to_local(nir);
nir_lower_vars_to_ssa(nir);
@@ -1401,7 +1403,7 @@ iris_compile_gs(struct iris_context *ice,
if (key->nr_userclip_plane_consts) {
nir_function_impl *impl = nir_shader_get_entrypoint(nir);
- nir_lower_clip_gs(nir, (1 << key->nr_userclip_plane_consts) - 1);
+ nir_lower_clip_gs(nir, (1 << key->nr_userclip_plane_consts) - 1, false);
nir_lower_io_to_temporaries(nir, impl, true, false);
nir_lower_global_vars_to_local(nir);
nir_lower_vars_to_ssa(nir);
diff --git a/src/gallium/drivers/vc4/vc4_program.c b/src/gallium/drivers/vc4/vc4_program.c
index 08487bea3a6..4568100c104 100644
--- a/src/gallium/drivers/vc4/vc4_program.c
+++ b/src/gallium/drivers/vc4/vc4_program.c
@@ -2313,10 +2313,11 @@ vc4_shader_ntq(struct vc4_context *vc4, enum qstage stage,
if (c->key->ucp_enables) {
if (stage == QSTAGE_FRAG) {
- NIR_PASS_V(c->s, nir_lower_clip_fs, c->key->ucp_enables);
+ NIR_PASS_V(c->s, nir_lower_clip_fs,
+ c->key->ucp_enables, false);
} else {
NIR_PASS_V(c->s, nir_lower_clip_vs,
- c->key->ucp_enables, false);
+ c->key->ucp_enables, false, false);
NIR_PASS_V(c->s, nir_lower_io_to_scalar,
nir_var_shader_out);
}
diff --git a/src/mesa/drivers/dri/i965/brw_nir_uniforms.cpp b/src/mesa/drivers/dri/i965/brw_nir_uniforms.cpp
index e31c91175c9..995b229e101 100644
--- a/src/mesa/drivers/dri/i965/brw_nir_uniforms.cpp
+++ b/src/mesa/drivers/dri/i965/brw_nir_uniforms.cpp
@@ -407,7 +407,7 @@ brw_nir_lower_legacy_clipping(nir_shader *nir, int nr_userclip_plane_consts,
nir_function_impl *impl = nir_shader_get_entrypoint(nir);
- nir_lower_clip_vs(nir, (1 << nr_userclip_plane_consts) - 1, true);
+ nir_lower_clip_vs(nir, (1 << nr_userclip_plane_consts) - 1, true, false);
nir_lower_io_to_temporaries(nir, impl, true, false);
nir_lower_global_vars_to_local(nir);
nir_lower_vars_to_ssa(nir);