/* Generates the lowering code for user-clip-planes, generating CLIPDIST
* from UCP[n] + CLIPVERTEX or POSITION. Additionally, an optional pass
- * for fragment shaders to insert conditional kill's based on the inter-
+ * for fragment shaders to insert conditional kills based on the inter-
* polated CLIPDIST
*
* NOTE: should be run after nir_lower_outputs_to_temporaries() (or at
static nir_variable *
-create_clipdist_var(nir_shader *shader, unsigned drvloc,
- bool output, gl_varying_slot slot)
+create_clipdist_var(nir_shader *shader,
+ 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);
+ /* TODO use type_size() for num_inputs/outputs */
+ if (output) {
+ var->data.driver_location = shader->num_outputs++;
+ var->data.mode = nir_var_shader_out;
+ } else {
+ var->data.driver_location = shader->num_inputs++;
+ var->data.mode = nir_var_shader_in;
+ }
+ var->name = ralloc_asprintf(var, "clipdist_%d", var->data.driver_location);
var->data.index = 0;
var->data.location = slot;
- if (output) {
- exec_list_push_tail(&shader->outputs, &var->node);
- shader->num_outputs++; /* TODO use type_size() */
- }
- else {
- exec_list_push_tail(&shader->inputs, &var->node);
- shader->num_inputs++; /* TODO use type_size() */
- }
+ 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();
+
+ nir_shader_add_variable(shader, var);
return var;
}
+static void
+create_clipdist_vars(nir_shader *shader, nir_variable **io_vars,
+ unsigned ucp_enables, bool output,
+ bool use_clipdist_array)
+{
+ if (use_clipdist_array) {
+ io_vars[0] =
+ create_clipdist_var(shader, output,
+ VARYING_SLOT_CLIP_DIST0,
+ util_last_bit(ucp_enables));
+ } else {
+ if (ucp_enables & 0x0f)
+ io_vars[0] =
+ create_clipdist_var(shader, output,
+ VARYING_SLOT_CLIP_DIST0, 0);
+ if (ucp_enables & 0xf0)
+ io_vars[1] =
+ create_clipdist_var(shader, output,
+ VARYING_SLOT_CLIP_DIST1, 0);
+ }
+}
+
static void
store_clipdist_output(nir_builder *b, nir_variable *out, nir_ssa_def **val)
{
}
static void
-load_clipdist_input(nir_builder *b, nir_variable *in, nir_ssa_def **val)
+load_clipdist_input(nir_builder *b, nir_variable *in, int location_offset,
+ nir_ssa_def **val)
{
nir_intrinsic_instr *load;
load = nir_intrinsic_instr_create(b->shader, nir_intrinsic_load_input);
load->num_components = 4;
- nir_intrinsic_set_base(load, in->data.driver_location);
+ nir_intrinsic_set_base(load, in->data.driver_location + location_offset);
load->src[0] = nir_src_for_ssa(nir_imm_int(b, 0));
- nir_ssa_dest_init(&load->instr, &load->dest, 4, NULL);
+ nir_ssa_dest_init(&load->instr, &load->dest, 4, 32, NULL);
nir_builder_instr_insert(b, &load->instr);
val[0] = nir_channel(b, &load->dest.ssa, 0);
val[3] = nir_channel(b, &load->dest.ssa, 3);
}
-struct find_output_state
-{
- unsigned drvloc;
- nir_ssa_def *def;
-};
-
-static bool
-find_output_in_block(nir_block *block, void *void_state)
+static nir_ssa_def *
+find_output_in_block(nir_block *block, unsigned drvloc)
{
- struct find_output_state *state = void_state;
- nir_foreach_instr(block, instr) {
+ nir_foreach_instr(instr, block) {
if (instr->type == nir_instr_type_intrinsic) {
nir_intrinsic_instr *intr = nir_instr_as_intrinsic(instr);
if ((intr->intrinsic == nir_intrinsic_store_output) &&
- nir_intrinsic_base(intr) == state->drvloc) {
- assert(state->def == NULL);
+ nir_intrinsic_base(intr) == drvloc) {
assert(intr->src[0].is_ssa);
- assert(nir_src_as_const_value(intr->src[1]));
- state->def = intr->src[0].ssa;
+ assert(nir_src_is_const(intr->src[1]));
+ return intr->src[0].ssa;
+ }
+ }
+ }
+ return NULL;
+}
+
+/* TODO: maybe this would be a useful helper?
+ * NOTE: assumes each output is written exactly once (and unconditionally)
+ * so if needed nir_lower_outputs_to_temporaries()
+ */
+static nir_ssa_def *
+find_output(nir_shader *shader, unsigned drvloc)
+{
+ nir_ssa_def *def = NULL;
+ nir_foreach_function(function, shader) {
+ if (function->impl) {
+ nir_foreach_block_reverse(block, function->impl) {
+ nir_ssa_def *new_def = find_output_in_block(block, drvloc);
+ assert(!(new_def && def));
+ def = new_def;
#if !defined(DEBUG)
/* for debug builds, scan entire shader to assert
* if output is written multiple times. For release
* builds just assume all is well and bail when we
* find first:
*/
- return false;
+ if (def)
+ break;
#endif
}
}
}
- return true;
+ return def;
+}
+
+static bool
+find_clipvertex_and_position_outputs(nir_shader *shader,
+ nir_variable **clipvertex,
+ nir_variable **position)
+{
+ nir_foreach_shader_out_variable(var, shader) {
+ switch (var->data.location) {
+ case VARYING_SLOT_POS:
+ *position = var;
+ break;
+ case VARYING_SLOT_CLIP_VERTEX:
+ *clipvertex = var;
+ break;
+ case VARYING_SLOT_CLIP_DIST0:
+ case VARYING_SLOT_CLIP_DIST1:
+ /* 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;
+ }
+ }
+
+ return *clipvertex || *position;
}
-/* TODO: maybe this would be a useful helper?
- * NOTE: assumes each output is written exactly once (and unconditionally)
- * so if needed nir_lower_outputs_to_temporaries()
- */
static nir_ssa_def *
-find_output(nir_shader *shader, unsigned drvloc)
+get_ucp(nir_builder *b, int plane,
+ const gl_state_index16 clipplane_state_tokens[][STATE_LENGTH])
{
- struct find_output_state state = {
- .drvloc = drvloc,
- };
+ if (clipplane_state_tokens) {
+ char tmp[100];
+ snprintf(tmp, ARRAY_SIZE(tmp), "gl_ClipPlane%dMESA", plane);
+ nir_variable *var = nir_variable_create(b->shader,
+ nir_var_uniform,
+ glsl_vec4_type(),
+ tmp);
+
+ var->num_state_slots = 1;
+ var->state_slots = ralloc_array(var, nir_state_slot, 1);
+ memcpy(var->state_slots[0].tokens,
+ clipplane_state_tokens[plane],
+ sizeof(var->state_slots[0].tokens));
+ return nir_load_var(b, var);
+ } else
+ return nir_load_user_clip_plane(b, plane);
+}
- nir_foreach_function(shader, function) {
- if (function->impl) {
- nir_foreach_block_reverse(function->impl,
- find_output_in_block, &state);
+
+static void
+lower_clip_outputs(nir_builder *b, nir_variable *position,
+ nir_variable *clipvertex, nir_variable **out,
+ unsigned ucp_enables, bool use_vars,
+ bool use_clipdist_array,
+ const gl_state_index16 clipplane_state_tokens[][STATE_LENGTH])
+{
+ nir_ssa_def *clipdist[MAX_CLIP_PLANES];
+ nir_ssa_def *cv;
+
+ if (use_vars) {
+ cv = nir_load_var(b, clipvertex ? clipvertex : position);
+
+ if (clipvertex) {
+ clipvertex->data.mode = nir_var_shader_temp;
+ nir_fixup_deref_modes(b->shader);
+ }
+ } else {
+ if (clipvertex)
+ cv = find_output(b->shader, clipvertex->data.driver_location);
+ else {
+ assert(position);
+ cv = find_output(b->shader, position->data.driver_location);
+ }
+ }
+
+ for (int plane = 0; plane < MAX_CLIP_PLANES; plane++) {
+ if (ucp_enables & (1 << plane)) {
+ nir_ssa_def *ucp = get_ucp(b, plane, clipplane_state_tokens);
+
+ /* calculate clipdist[plane] - dot(ucp, cv): */
+ clipdist[plane] = nir_fdot4(b, ucp, cv);
+ } else {
+ /* 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);
}
}
- return state.def;
+ 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]);
+ }
+ }
}
/*
* VS lowering
*/
-static void
-lower_clip_vs(nir_function_impl *impl, unsigned ucp_enables,
- nir_ssa_def *cv, nir_variable **out)
+/* 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.
+ *
+ * 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,
+ bool use_clipdist_array,
+ const gl_state_index16 clipplane_state_tokens[][STATE_LENGTH])
{
- nir_ssa_def *clipdist[MAX_CLIP_PLANES];
+ nir_function_impl *impl = nir_shader_get_entrypoint(shader);
nir_builder b;
+ nir_variable *position = NULL;
+ nir_variable *clipvertex = NULL;
+ nir_variable *out[2] = { NULL };
+
+ if (!ucp_enables)
+ return false;
nir_builder_init(&b, impl);
* should be only a single predecessor block to end_block, which
* makes the perfect place to insert the clipdist calculations.
*
- * NOTE: in case of early return's, these would have to be lowered
+ * NOTE: in case of early returns, these would have to be lowered
* to jumps to end_block predecessor in a previous pass. Not sure
* if there is a good way to sanity check this, but for now the
* users of this pass don't support sub-routines.
assert(impl->end_block->predecessors->entries == 1);
b.cursor = nir_after_cf_list(&impl->body);
- for (int plane = 0; plane < MAX_CLIP_PLANES; plane++) {
- if (ucp_enables & (1 << plane)) {
- nir_ssa_def *ucp =
- nir_load_system_value(&b, nir_intrinsic_load_user_clip_plane, plane);
+ /* find clipvertex/position outputs */
+ if (!find_clipvertex_and_position_outputs(shader, &clipvertex, &position))
+ return false;
- /* calculate clipdist[plane] - dot(ucp, cv): */
- clipdist[plane] = nir_fdot4(&b, ucp, cv);
- }
- else {
- /* 0.0 == don't-clip == disabled: */
- clipdist[plane] = nir_imm_float(&b, 0.0);
- }
- }
+ /* insert CLIPDIST outputs */
+ create_clipdist_vars(shader, out, ucp_enables, true,
+ use_clipdist_array);
- if (ucp_enables & 0x0f)
- store_clipdist_output(&b, out[0], &clipdist[0]);
- if (ucp_enables & 0xf0)
- store_clipdist_output(&b, out[1], &clipdist[4]);
+ lower_clip_outputs(&b, position, clipvertex, out, ucp_enables, use_vars,
+ use_clipdist_array, clipplane_state_tokens);
nir_metadata_preserve(impl, nir_metadata_dominance);
+
+ return true;
}
-/* ucp_enables is bitmask of enabled ucp's. Actual ucp values are
- * passed in to shader via user_clip_plane system-values
+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, bool use_clipdist_array,
+ const gl_state_index16 clipplane_state_tokens[][STATE_LENGTH])
+{
+ nir_foreach_instr_safe(instr, block) {
+ if (instr->type != nir_instr_type_intrinsic)
+ continue;
+
+ nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
+ switch (intrin->intrinsic) {
+ 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,
+ use_clipdist_array, clipplane_state_tokens);
+ break;
+ default:
+ /* not interesting; skip this */
+ break;
+ }
+ }
+}
+
+/*
+ * GS lowering
*/
-void
-nir_lower_clip_vs(nir_shader *shader, unsigned ucp_enables)
+
+bool
+nir_lower_clip_gs(nir_shader *shader, unsigned ucp_enables,
+ bool use_clipdist_array,
+ const gl_state_index16 clipplane_state_tokens[][STATE_LENGTH])
{
- int clipvertex = -1;
- int position = -1;
- int maxloc = -1;
- nir_ssa_def *cv;
+ nir_function_impl *impl = nir_shader_get_entrypoint(shader);
+ nir_builder b;
+ nir_variable *position = NULL;
+ nir_variable *clipvertex = NULL;
nir_variable *out[2] = { NULL };
if (!ucp_enables)
- return;
+ return false;
- /* find clipvertex/position outputs: */
- nir_foreach_variable(var, &shader->outputs) {
- int loc = var->data.driver_location;
+ /* find clipvertex/position outputs */
+ if (!find_clipvertex_and_position_outputs(shader, &clipvertex, &position))
+ return false;
- /* keep track of last used driver-location.. we'll be
- * appending CLIP_DIST0/CLIP_DIST1 after last existing
- * output:
- */
- maxloc = MAX2(maxloc, loc);
+ /* insert CLIPDIST outputs */
+ create_clipdist_vars(shader, out, ucp_enables, true,
+ use_clipdist_array);
- switch (var->data.location) {
- case VARYING_SLOT_POS:
- position = loc;
- break;
- case VARYING_SLOT_CLIP_VERTEX:
- clipvertex = loc;
- break;
- case VARYING_SLOT_CLIP_DIST0:
- case VARYING_SLOT_CLIP_DIST1:
- /* if shader is already writing CLIPDIST, then
- * there should be no user-clip-planes to deal
- * with.
- */
- return;
- }
- }
+ nir_builder_init(&b, impl);
- if (clipvertex != -1)
- cv = find_output(shader, clipvertex);
- else if (position != -1)
- cv = find_output(shader, position);
- else
- return;
+ nir_foreach_block(block, impl)
+ lower_clip_in_gs_block(&b, block, position, clipvertex, out,
+ ucp_enables, use_clipdist_array,
+ clipplane_state_tokens);
- /* insert CLIPDIST outputs: */
- if (ucp_enables & 0x0f)
- out[0] =
- create_clipdist_var(shader, ++maxloc, true, VARYING_SLOT_CLIP_DIST0);
- if (ucp_enables & 0xf0)
- out[1] =
- create_clipdist_var(shader, ++maxloc, true, VARYING_SLOT_CLIP_DIST1);
+ nir_metadata_preserve(impl, nir_metadata_dominance);
- nir_foreach_function(shader, function) {
- if (!strcmp(function->name, "main"))
- lower_clip_vs(function->impl, ucp_enables, cv, out);
- }
+ return true;
}
/*
static void
lower_clip_fs(nir_function_impl *impl, unsigned ucp_enables,
- nir_variable **in)
+ nir_variable **in, bool use_clipdist_array)
{
nir_ssa_def *clipdist[MAX_CLIP_PLANES];
nir_builder b;
nir_builder_init(&b, impl);
b.cursor = nir_before_cf_list(&impl->body);
- if (ucp_enables & 0x0f)
- load_clipdist_input(&b, in[0], &clipdist[0]);
- if (ucp_enables & 0xf0)
- load_clipdist_input(&b, in[1], &clipdist[4]);
+ if (!use_clipdist_array) {
+ if (ucp_enables & 0x0f)
+ load_clipdist_input(&b, in[0], 0, &clipdist[0]);
+ if (ucp_enables & 0xf0)
+ load_clipdist_input(&b, in[1], 0, &clipdist[4]);
+ } else {
+ if (ucp_enables & 0x0f)
+ load_clipdist_input(&b, in[0], 0, &clipdist[0]);
+ if (ucp_enables & 0xf0)
+ load_clipdist_input(&b, in[0], 1, &clipdist[4]);
+ }
for (int plane = 0; plane < MAX_CLIP_PLANES; plane++) {
if (ucp_enables & (1 << plane)) {
nir_intrinsic_discard_if);
discard->src[0] = nir_src_for_ssa(cond);
nir_builder_instr_insert(&b, &discard->instr);
+
+ b.shader->info.fs.uses_discard = true;
}
}
+
+ nir_metadata_preserve(impl, nir_metadata_dominance);
+}
+
+static bool
+fs_has_clip_dist_input_var(nir_shader *shader, nir_variable **io_vars,
+ unsigned *ucp_enables)
+{
+ assert(shader->info.stage == MESA_SHADER_FRAGMENT);
+ nir_foreach_shader_in_variable(var, shader) {
+ switch (var->data.location) {
+ case VARYING_SLOT_CLIP_DIST0:
+ assert(var->data.compact);
+ io_vars[0] = var;
+ *ucp_enables &= (1 << glsl_get_length(var->type)) - 1;
+ return true;
+ default:
+ break;
+ }
+ }
+ return false;
}
/* insert conditional kill based on interpolated CLIPDIST
*/
-void
-nir_lower_clip_fs(nir_shader *shader, unsigned ucp_enables)
+bool
+nir_lower_clip_fs(nir_shader *shader, unsigned ucp_enables,
+ bool use_clipdist_array)
{
- nir_variable *in[2];
- int maxloc = -1;
+ nir_variable *in[2] = {0};
if (!ucp_enables)
- return;
-
- nir_foreach_variable(var, &shader->inputs) {
- int loc = var->data.driver_location;
-
- /* keep track of last used driver-location.. we'll be
- * appending CLIP_DIST0/CLIP_DIST1 after last existing
- * input:
- */
- maxloc = MAX2(maxloc, loc);
- }
+ return false;
- /* The shader won't normally have CLIPDIST inputs, so we
- * must add our own:
+ /* Fragment shaders can't read gl_ClipDistance[] in OpenGL so it will not
+ * have the variable defined, but Vulkan allows this, in which case the
+ * SPIR-V compiler would have already added it as a compact array.
*/
- /* insert CLIPDIST outputs: */
- if (ucp_enables & 0x0f)
- in[0] =
- create_clipdist_var(shader, ++maxloc, false,
- VARYING_SLOT_CLIP_DIST0);
- if (ucp_enables & 0xf0)
- in[1] =
- create_clipdist_var(shader, ++maxloc, false,
- VARYING_SLOT_CLIP_DIST1);
-
- nir_foreach_function(shader, function) {
+ if (!fs_has_clip_dist_input_var(shader, in, &ucp_enables))
+ create_clipdist_vars(shader, in, ucp_enables, false, use_clipdist_array);
+ else
+ assert(use_clipdist_array);
+
+ nir_foreach_function(function, shader) {
if (!strcmp(function->name, "main"))
- lower_clip_fs(function->impl, ucp_enables, in);
+ lower_clip_fs(function->impl, ucp_enables, in, use_clipdist_array);
}
+
+ return true;
}
projects / mesa.git / blobdiff