From 1f46163acb4853b3e159e4a0cb91d61c8a348553 Mon Sep 17 00:00:00 2001 From: Kenneth Graunke Date: Tue, 10 Nov 2015 01:17:04 -0800 Subject: i965: Add tessellation shader VUE map code. Based on a patch by Chris Forbes, but largely rewritten by Ken. Signed-off-by: Kenneth Graunke Reviewed-by: Jordan Justen --- src/mesa/drivers/dri/i965/brw_compiler.h | 20 ++++++- src/mesa/drivers/dri/i965/brw_vue_map.c | 94 ++++++++++++++++++++++++++++++-- 2 files changed, 107 insertions(+), 7 deletions(-) (limited to 'src/mesa/drivers') diff --git a/src/mesa/drivers/dri/i965/brw_compiler.h b/src/mesa/drivers/dri/i965/brw_compiler.h index 514788c2174..c9e03175010 100644 --- a/src/mesa/drivers/dri/i965/brw_compiler.h +++ b/src/mesa/drivers/dri/i965/brw_compiler.h @@ -442,7 +442,7 @@ struct brw_vue_map { * additional processing is applied before storing them in the VUE), the * value is -1. */ - signed char varying_to_slot[BRW_VARYING_SLOT_COUNT]; + signed char varying_to_slot[VARYING_SLOT_TESS_MAX]; /** * Map from VUE slot to gl_varying_slot value. For slots that do not @@ -451,12 +451,24 @@ struct brw_vue_map { * * For slots that are not in use, the value is BRW_VARYING_SLOT_PAD. */ - signed char slot_to_varying[BRW_VARYING_SLOT_COUNT]; + signed char slot_to_varying[VARYING_SLOT_TESS_MAX]; /** * Total number of VUE slots in use */ int num_slots; + + /** + * Number of per-patch VUE slots. Only valid for tessellation control + * shader outputs and tessellation evaluation shader inputs. + */ + int num_per_patch_slots; + + /** + * Number of per-vertex VUE slots. Only valid for tessellation control + * shader outputs and tessellation evaluation shader inputs. + */ + int num_per_vertex_slots; }; void brw_print_vue_map(FILE *fp, const struct brw_vue_map *vue_map); @@ -484,6 +496,10 @@ void brw_compute_vue_map(const struct brw_device_info *devinfo, GLbitfield64 slots_valid, bool separate_shader); +void brw_compute_tess_vue_map(struct brw_vue_map *const vue_map, + const GLbitfield64 slots_valid, + const GLbitfield is_patch); + enum shader_dispatch_mode { DISPATCH_MODE_4X1_SINGLE = 0, DISPATCH_MODE_4X2_DUAL_INSTANCE = 1, diff --git a/src/mesa/drivers/dri/i965/brw_vue_map.c b/src/mesa/drivers/dri/i965/brw_vue_map.c index 6cb3da46995..09eadbcb54f 100644 --- a/src/mesa/drivers/dri/i965/brw_vue_map.c +++ b/src/mesa/drivers/dri/i965/brw_vue_map.c @@ -176,6 +176,73 @@ brw_compute_vue_map(const struct brw_device_info *devinfo, } vue_map->num_slots = separate ? slot + 1 : slot; + vue_map->num_per_vertex_slots = 0; + vue_map->num_per_patch_slots = 0; +} + +/** + * Compute the VUE map for tessellation control shader outputs and + * tessellation evaluation shader inputs. + */ +void +brw_compute_tess_vue_map(struct brw_vue_map *vue_map, + GLbitfield64 vertex_slots, + GLbitfield patch_slots) +{ + /* I don't think anything actually uses this... */ + vue_map->slots_valid = vertex_slots; + + vertex_slots &= ~(VARYING_BIT_TESS_LEVEL_OUTER | + VARYING_BIT_TESS_LEVEL_INNER); + + /* Make sure that the values we store in vue_map->varying_to_slot and + * vue_map->slot_to_varying won't overflow the signed chars that are used + * to store them. Note that since vue_map->slot_to_varying sometimes holds + * values equal to VARYING_SLOT_TESS_MAX , we need to ensure that + * VARYING_SLOT_TESS_MAX is <= 127, not 128. + */ + STATIC_ASSERT(VARYING_SLOT_TESS_MAX <= 127); + + for (int i = 0; i < VARYING_SLOT_TESS_MAX ; ++i) { + vue_map->varying_to_slot[i] = -1; + vue_map->slot_to_varying[i] = BRW_VARYING_SLOT_PAD; + } + + int slot = 0; + + /* The first 8 DWords are reserved for the "Patch Header". + * + * VARYING_SLOT_TESS_LEVEL_OUTER / INNER live here, but the exact layout + * depends on the domain type. They might not be in slots 0 and 1 as + * described here, but pretending they're separate allows us to uniquely + * identify them by distinct slot locations. + */ + assign_vue_slot(vue_map, VARYING_SLOT_TESS_LEVEL_INNER, slot++); + assign_vue_slot(vue_map, VARYING_SLOT_TESS_LEVEL_OUTER, slot++); + + /* first assign per-patch varyings */ + while (patch_slots != 0) { + const int varying = ffsll(patch_slots) - 1; + if (vue_map->varying_to_slot[varying + VARYING_SLOT_PATCH0] == -1) { + assign_vue_slot(vue_map, varying + VARYING_SLOT_PATCH0, slot++); + } + patch_slots &= ~BITFIELD64_BIT(varying); + } + + /* apparently, including the patch header... */ + vue_map->num_per_patch_slots = slot; + + /* then assign per-vertex varyings for each vertex in our patch */ + while (vertex_slots != 0) { + const int varying = ffsll(vertex_slots) - 1; + if (vue_map->varying_to_slot[varying] == -1) { + assign_vue_slot(vue_map, varying, slot++); + } + vertex_slots &= ~BITFIELD64_BIT(varying); + } + + vue_map->num_per_vertex_slots = slot - vue_map->num_per_patch_slots; + vue_map->num_slots = slot; } static const char * @@ -196,11 +263,28 @@ varying_name(brw_varying_slot slot) void brw_print_vue_map(FILE *fp, const struct brw_vue_map *vue_map) { - fprintf(fp, "VUE map (%d slots, %s)\n", - vue_map->num_slots, vue_map->separate ? "SSO" : "non-SSO"); - for (int i = 0; i < vue_map->num_slots; i++) { - fprintf(fp, " [%d] %s\n", i, - varying_name(vue_map->slot_to_varying[i])); + if (vue_map->num_per_vertex_slots > 0 || vue_map->num_per_patch_slots > 0) { + fprintf(fp, "PUE map (%d slots, %d/patch, %d/vertex, %s)\n", + vue_map->num_slots, + vue_map->num_per_patch_slots, + vue_map->num_per_vertex_slots, + vue_map->separate ? "SSO" : "non-SSO"); + for (int i = 0; i < vue_map->num_slots; i++) { + if (vue_map->slot_to_varying[i] >= VARYING_SLOT_PATCH0) { + fprintf(fp, " [%d] VARYING_SLOT_PATCH%d\n", i, + vue_map->slot_to_varying[i] - VARYING_SLOT_PATCH0); + } else { + fprintf(fp, " [%d] %s\n", i, + varying_name(vue_map->slot_to_varying[i])); + } + } + } else { + fprintf(fp, "VUE map (%d slots, %s)\n", + vue_map->num_slots, vue_map->separate ? "SSO" : "non-SSO"); + for (int i = 0; i < vue_map->num_slots; i++) { + fprintf(fp, " [%d] %s\n", i, + varying_name(vue_map->slot_to_varying[i])); + } } fprintf(fp, "\n"); } -- cgit v1.2.3