Mesa (master): nir: Make nir_lower_clip_vs optionally work with variables.
GitLab Mirror
gitlab-mirror at kemper.freedesktop.org
Mon Nov 19 22:36:27 UTC 2018
Module: Mesa
Branch: master
Commit: 5b682143da1a24f4c66279fb5c6f9deefcb8a986
URL: http://cgit.freedesktop.org/mesa/mesa/commit/?id=5b682143da1a24f4c66279fb5c6f9deefcb8a986
Author: Kenneth Graunke <kenneth at whitecape.org>
Date: Sun May 21 19:26:15 2017 -0700
nir: Make nir_lower_clip_vs optionally work with variables.
The way nir_lower_clip_vs() works with store_output intrinsics makes a
ton of assumptions about the driver_location field.
In i965 and iris, I'd rather do this lowering early and work with
variables. v3d may want to switch to that as well, and ir3 could too,
but I'm not sure exactly what would need updating. For now, handle
both methods.
Reviewed-by: Eric Anholt <eric at anholt.net>
---
src/broadcom/compiler/vir.c | 3 +-
src/compiler/nir/nir.h | 2 +-
src/compiler/nir/nir_lower_clip.c | 45 ++++++++++++++++++++++-------
src/gallium/drivers/freedreno/ir3/ir3_nir.c | 2 +-
src/gallium/drivers/vc4/vc4_program.c | 3 +-
5 files changed, 40 insertions(+), 15 deletions(-)
diff --git a/src/broadcom/compiler/vir.c b/src/broadcom/compiler/vir.c
index 9a671c6c97..a52205b1a2 100644
--- a/src/broadcom/compiler/vir.c
+++ b/src/broadcom/compiler/vir.c
@@ -745,7 +745,8 @@ uint64_t *v3d_compile_vs(const struct v3d_compiler *compiler,
NIR_PASS_V(c->s, nir_lower_clamp_color_outputs);
if (key->base.ucp_enables) {
- NIR_PASS_V(c->s, nir_lower_clip_vs, key->base.ucp_enables);
+ NIR_PASS_V(c->s, nir_lower_clip_vs, key->base.ucp_enables,
+ false);
NIR_PASS_V(c->s, nir_lower_io_to_scalar,
nir_var_shader_out);
}
diff --git a/src/compiler/nir/nir.h b/src/compiler/nir/nir.h
index 1dd605010f..a292ec73e1 100644
--- a/src/compiler/nir/nir.h
+++ b/src/compiler/nir/nir.h
@@ -3014,7 +3014,7 @@ bool nir_lower_tex(nir_shader *shader,
bool nir_lower_idiv(nir_shader *shader);
-bool nir_lower_clip_vs(nir_shader *shader, unsigned ucp_enables);
+bool nir_lower_clip_vs(nir_shader *shader, unsigned ucp_enables, bool use_vars);
bool nir_lower_clip_fs(nir_shader *shader, unsigned ucp_enables);
bool nir_lower_clip_cull_distance_arrays(nir_shader *nir);
diff --git a/src/compiler/nir/nir_lower_clip.c b/src/compiler/nir/nir_lower_clip.c
index 496c39e1b6..880d65c617 100644
--- a/src/compiler/nir/nir_lower_clip.c
+++ b/src/compiler/nir/nir_lower_clip.c
@@ -152,9 +152,12 @@ find_output(nir_shader *shader, unsigned drvloc)
/* ucp_enables is bitmask of enabled ucps. Actual ucp values are
* passed in to shader via user_clip_plane system-values
+ *
+ * If use_vars is true, the pass will use variable loads and stores instead
+ * of working with store_output intrinsics.
*/
bool
-nir_lower_clip_vs(nir_shader *shader, unsigned ucp_enables)
+nir_lower_clip_vs(nir_shader *shader, unsigned ucp_enables, bool use_vars)
{
nir_function_impl *impl = nir_shader_get_entrypoint(shader);
nir_ssa_def *clipdist[MAX_CLIP_PLANES];
@@ -196,17 +199,30 @@ nir_lower_clip_vs(nir_shader *shader, unsigned ucp_enables)
/* if shader is already writing CLIPDIST, then
* there should be no user-clip-planes to deal
* with.
+ *
+ * We assume nir_remove_dead_variables has removed the clipdist
+ * variables if they're not written.
*/
return false;
}
}
- if (clipvertex)
- cv = find_output(shader, clipvertex->data.driver_location);
- else if (position)
- cv = find_output(shader, position->data.driver_location);
- else
- return false;
+ if (use_vars) {
+ cv = nir_load_var(&b, clipvertex ? clipvertex : position);
+
+ if (clipvertex) {
+ exec_node_remove(&clipvertex->node);
+ clipvertex->data.mode = nir_var_global;
+ exec_list_push_tail(&shader->globals, &clipvertex->node);
+ }
+ } else {
+ if (clipvertex)
+ cv = find_output(shader, clipvertex->data.driver_location);
+ else if (position)
+ cv = find_output(shader, position->data.driver_location);
+ else
+ return false;
+ }
/* insert CLIPDIST outputs: */
if (ucp_enables & 0x0f)
@@ -228,10 +244,17 @@ nir_lower_clip_vs(nir_shader *shader, unsigned ucp_enables)
}
}
- 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_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]);
+ }
nir_metadata_preserve(impl, nir_metadata_dominance);
diff --git a/src/gallium/drivers/freedreno/ir3/ir3_nir.c b/src/gallium/drivers/freedreno/ir3/ir3_nir.c
index 63866ae4d0..7c2a8f83b6 100644
--- a/src/gallium/drivers/freedreno/ir3/ir3_nir.c
+++ b/src/gallium/drivers/freedreno/ir3/ir3_nir.c
@@ -172,7 +172,7 @@ 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);
+ OPT_V(s, nir_lower_clip_vs, key->ucp_enables, false);
if (key->vclamp_color)
OPT_V(s, nir_lower_clamp_color_outputs);
} else if (s->info.stage == MESA_SHADER_FRAGMENT) {
diff --git a/src/gallium/drivers/vc4/vc4_program.c b/src/gallium/drivers/vc4/vc4_program.c
index bc9bd76ae9..b98baca30c 100644
--- a/src/gallium/drivers/vc4/vc4_program.c
+++ b/src/gallium/drivers/vc4/vc4_program.c
@@ -2363,7 +2363,8 @@ vc4_shader_ntq(struct vc4_context *vc4, enum qstage stage,
if (stage == QSTAGE_FRAG) {
NIR_PASS_V(c->s, nir_lower_clip_fs, c->key->ucp_enables);
} else {
- NIR_PASS_V(c->s, nir_lower_clip_vs, c->key->ucp_enables);
+ NIR_PASS_V(c->s, nir_lower_clip_vs,
+ c->key->ucp_enables, false);
NIR_PASS_V(c->s, nir_lower_io_to_scalar,
nir_var_shader_out);
}
More information about the mesa-commit
mailing list