summaryrefslogtreecommitdiffstats
path: root/src/gallium
diff options
context:
space:
mode:
authorChristian König <[email protected]>2012-07-18 17:39:15 +0200
committerChristian König <[email protected]>2012-07-24 12:29:30 +0200
commit3c09f11e5cefd437bb8185539430786dc245c96f (patch)
tree2ef17dbda9e1e89ca17d9d07d77175759e077d96 /src/gallium
parentbd2a5cf328c21f2d5b243442ee2eac73a996b15c (diff)
radeonsi: move shaders to new handling
Signed-off-by: Christian König <[email protected]>
Diffstat (limited to 'src/gallium')
-rw-r--r--src/gallium/drivers/radeonsi/evergreen_state.c214
-rw-r--r--src/gallium/drivers/radeonsi/r600_state_common.c77
-rw-r--r--src/gallium/drivers/radeonsi/radeonsi_pipe.h44
-rw-r--r--src/gallium/drivers/radeonsi/radeonsi_shader.c4
-rw-r--r--src/gallium/drivers/radeonsi/si_state.c250
-rw-r--r--src/gallium/drivers/radeonsi/si_state.h42
6 files changed, 295 insertions, 336 deletions
diff --git a/src/gallium/drivers/radeonsi/evergreen_state.c b/src/gallium/drivers/radeonsi/evergreen_state.c
index b04c1cb9ecf..bb027eee448 100644
--- a/src/gallium/drivers/radeonsi/evergreen_state.c
+++ b/src/gallium/drivers/radeonsi/evergreen_state.c
@@ -1054,20 +1054,14 @@ static void evergreen_set_sample_mask(struct pipe_context *pipe, unsigned sample
void cayman_init_state_functions(struct r600_context *rctx)
{
si_init_state_functions(rctx);
- rctx->context.create_fs_state = si_create_shader_state;
rctx->context.create_sampler_state = si_create_sampler_state;
rctx->context.create_sampler_view = evergreen_create_sampler_view;
rctx->context.create_vertex_elements_state = si_create_vertex_elements;
- rctx->context.create_vs_state = si_create_shader_state;
rctx->context.bind_fragment_sampler_states = evergreen_bind_ps_sampler;
- rctx->context.bind_fs_state = r600_bind_ps_shader;
rctx->context.bind_vertex_elements_state = r600_bind_vertex_elements;
rctx->context.bind_vertex_sampler_states = evergreen_bind_vs_sampler;
- rctx->context.bind_vs_state = r600_bind_vs_shader;
- rctx->context.delete_fs_state = r600_delete_ps_shader;
rctx->context.delete_sampler_state = si_delete_sampler_state;
rctx->context.delete_vertex_elements_state = r600_delete_vertex_element;
- rctx->context.delete_vs_state = r600_delete_vs_shader;
rctx->context.set_constant_buffer = r600_set_constant_buffer;
rctx->context.set_fragment_sampler_views = evergreen_set_ps_sampler_view;
rctx->context.set_polygon_stipple = evergreen_set_polygon_stipple;
@@ -1081,211 +1075,3 @@ void cayman_init_state_functions(struct r600_context *rctx)
rctx->context.stream_output_target_destroy = r600_so_target_destroy;
rctx->context.set_stream_output_targets = r600_set_so_targets;
}
-
-void si_pipe_shader_ps(struct pipe_context *ctx, struct si_pipe_shader *shader)
-{
- struct r600_context *rctx = (struct r600_context *)ctx;
- struct r600_pipe_state *rstate = &shader->rstate;
- struct r600_shader *rshader = &shader->shader;
- unsigned i, exports_ps, num_cout, spi_ps_in_control, db_shader_control;
- unsigned num_sgprs, num_user_sgprs;
- int pos_index = -1, face_index = -1;
- int ninterp = 0;
- boolean have_linear = FALSE, have_centroid = FALSE, have_perspective = FALSE;
- unsigned spi_baryc_cntl;
- uint64_t va;
-
- if (si_pipe_shader_create(ctx, shader))
- return;
-
- rstate->nregs = 0;
-
- db_shader_control = S_02880C_Z_ORDER(V_02880C_EARLY_Z_THEN_LATE_Z);
- for (i = 0; i < rshader->ninput; i++) {
- ninterp++;
- /* XXX: Flat shading hangs the GPU */
- if (rshader->input[i].interpolate == TGSI_INTERPOLATE_CONSTANT ||
- (rshader->input[i].interpolate == TGSI_INTERPOLATE_COLOR &&
- rctx->queued.named.rasterizer->flatshade))
- have_linear = TRUE;
- if (rshader->input[i].interpolate == TGSI_INTERPOLATE_LINEAR)
- have_linear = TRUE;
- if (rshader->input[i].interpolate == TGSI_INTERPOLATE_PERSPECTIVE)
- have_perspective = TRUE;
- if (rshader->input[i].centroid)
- have_centroid = TRUE;
- }
-
- for (i = 0; i < rshader->noutput; i++) {
- if (rshader->output[i].name == TGSI_SEMANTIC_POSITION)
- db_shader_control |= S_02880C_Z_EXPORT_ENABLE(1);
- if (rshader->output[i].name == TGSI_SEMANTIC_STENCIL)
- db_shader_control |= 0; // XXX OP_VAL or TEST_VAL?
- }
- if (rshader->uses_kill)
- db_shader_control |= S_02880C_KILL_ENABLE(1);
-
- exports_ps = 0;
- num_cout = 0;
- for (i = 0; i < rshader->noutput; i++) {
- if (rshader->output[i].name == TGSI_SEMANTIC_POSITION ||
- rshader->output[i].name == TGSI_SEMANTIC_STENCIL)
- exports_ps |= 1;
- else if (rshader->output[i].name == TGSI_SEMANTIC_COLOR) {
- if (rshader->fs_write_all)
- num_cout = rshader->nr_cbufs;
- else
- num_cout++;
- }
- }
- if (!exports_ps) {
- /* always at least export 1 component per pixel */
- exports_ps = 2;
- }
-
- spi_ps_in_control = S_0286D8_NUM_INTERP(ninterp);
-
- spi_baryc_cntl = 0;
- if (have_perspective)
- spi_baryc_cntl |= have_centroid ?
- S_0286E0_PERSP_CENTROID_CNTL(1) : S_0286E0_PERSP_CENTER_CNTL(1);
- if (have_linear)
- spi_baryc_cntl |= have_centroid ?
- S_0286E0_LINEAR_CENTROID_CNTL(1) : S_0286E0_LINEAR_CENTER_CNTL(1);
-
- r600_pipe_state_add_reg(rstate,
- R_0286E0_SPI_BARYC_CNTL,
- spi_baryc_cntl,
- NULL, 0);
-
- r600_pipe_state_add_reg(rstate,
- R_0286CC_SPI_PS_INPUT_ENA,
- shader->spi_ps_input_ena,
- NULL, 0);
-
- r600_pipe_state_add_reg(rstate,
- R_0286D0_SPI_PS_INPUT_ADDR,
- shader->spi_ps_input_ena,
- NULL, 0);
-
- r600_pipe_state_add_reg(rstate,
- R_0286D8_SPI_PS_IN_CONTROL,
- spi_ps_in_control,
- NULL, 0);
-
- /* XXX: Depends on Z buffer format? */
- r600_pipe_state_add_reg(rstate,
- R_028710_SPI_SHADER_Z_FORMAT,
- 0,
- NULL, 0);
-
- /* XXX: Depends on color buffer format? */
- r600_pipe_state_add_reg(rstate,
- R_028714_SPI_SHADER_COL_FORMAT,
- S_028714_COL0_EXPORT_FORMAT(V_028714_SPI_SHADER_32_ABGR),
- NULL, 0);
-
- va = r600_resource_va(ctx->screen, (void *)shader->bo);
- r600_pipe_state_add_reg(rstate,
- R_00B020_SPI_SHADER_PGM_LO_PS,
- va >> 8,
- shader->bo, RADEON_USAGE_READ);
- r600_pipe_state_add_reg(rstate,
- R_00B024_SPI_SHADER_PGM_HI_PS,
- va >> 40,
- shader->bo, RADEON_USAGE_READ);
-
- num_user_sgprs = 6;
- num_sgprs = shader->num_sgprs;
- if (num_user_sgprs > num_sgprs)
- num_sgprs = num_user_sgprs;
- /* Last 2 reserved SGPRs are used for VCC */
- num_sgprs += 2;
- assert(num_sgprs <= 104);
-
- r600_pipe_state_add_reg(rstate,
- R_00B028_SPI_SHADER_PGM_RSRC1_PS,
- S_00B028_VGPRS((shader->num_vgprs - 1) / 4) |
- S_00B028_SGPRS((num_sgprs - 1) / 8),
- NULL, 0);
- r600_pipe_state_add_reg(rstate,
- R_00B02C_SPI_SHADER_PGM_RSRC2_PS,
- S_00B02C_USER_SGPR(num_user_sgprs),
- NULL, 0);
-
- r600_pipe_state_add_reg(rstate, R_02880C_DB_SHADER_CONTROL,
- db_shader_control,
- NULL, 0);
-
- shader->sprite_coord_enable = rctx->sprite_coord_enable;
-}
-
-void si_pipe_shader_vs(struct pipe_context *ctx, struct si_pipe_shader *shader)
-{
- struct r600_context *rctx = (struct r600_context *)ctx;
- struct r600_pipe_state *rstate = &shader->rstate;
- struct r600_shader *rshader = &shader->shader;
- unsigned num_sgprs, num_user_sgprs;
- unsigned nparams, i;
- uint64_t va;
-
- if (si_pipe_shader_create(ctx, shader))
- return;
-
- /* clear previous register */
- rstate->nregs = 0;
-
- /* Certain attributes (position, psize, etc.) don't count as params.
- * VS is required to export at least one param and r600_shader_from_tgsi()
- * takes care of adding a dummy export.
- */
- for (nparams = 0, i = 0 ; i < rshader->noutput; i++) {
- if (rshader->output[i].name != TGSI_SEMANTIC_POSITION)
- nparams++;
- }
- if (nparams < 1)
- nparams = 1;
-
- r600_pipe_state_add_reg(rstate,
- R_0286C4_SPI_VS_OUT_CONFIG,
- S_0286C4_VS_EXPORT_COUNT(nparams - 1),
- NULL, 0);
-
- r600_pipe_state_add_reg(rstate,
- R_02870C_SPI_SHADER_POS_FORMAT,
- S_02870C_POS0_EXPORT_FORMAT(V_02870C_SPI_SHADER_4COMP) |
- S_02870C_POS1_EXPORT_FORMAT(V_02870C_SPI_SHADER_NONE) |
- S_02870C_POS2_EXPORT_FORMAT(V_02870C_SPI_SHADER_NONE) |
- S_02870C_POS3_EXPORT_FORMAT(V_02870C_SPI_SHADER_NONE),
- NULL, 0);
-
- va = r600_resource_va(ctx->screen, (void *)shader->bo);
- r600_pipe_state_add_reg(rstate,
- R_00B120_SPI_SHADER_PGM_LO_VS,
- va >> 8,
- shader->bo, RADEON_USAGE_READ);
- r600_pipe_state_add_reg(rstate,
- R_00B124_SPI_SHADER_PGM_HI_VS,
- va >> 40,
- shader->bo, RADEON_USAGE_READ);
-
- num_user_sgprs = 8;
- num_sgprs = shader->num_sgprs;
- if (num_user_sgprs > num_sgprs)
- num_sgprs = num_user_sgprs;
- /* Last 2 reserved SGPRs are used for VCC */
- num_sgprs += 2;
- assert(num_sgprs <= 104);
-
- r600_pipe_state_add_reg(rstate,
- R_00B128_SPI_SHADER_PGM_RSRC1_VS,
- S_00B128_VGPRS((shader->num_vgprs - 1) / 4) |
- S_00B128_SGPRS((num_sgprs - 1) / 8),
- NULL, 0);
- r600_pipe_state_add_reg(rstate,
- R_00B12C_SPI_SHADER_PGM_RSRC2_VS,
- S_00B12C_USER_SGPR(num_user_sgprs),
- NULL, 0);
-}
-
-
diff --git a/src/gallium/drivers/radeonsi/r600_state_common.c b/src/gallium/drivers/radeonsi/r600_state_common.c
index 3ad105bc963..734826c592b 100644
--- a/src/gallium/drivers/radeonsi/r600_state_common.c
+++ b/src/gallium/drivers/radeonsi/r600_state_common.c
@@ -119,7 +119,7 @@ void r600_delete_state(struct pipe_context *ctx, void *state)
void r600_bind_vertex_elements(struct pipe_context *ctx, void *state)
{
struct r600_context *rctx = (struct r600_context *)ctx;
- struct r600_vertex_element *v = (struct r600_vertex_element*)state;
+ struct si_vertex_element *v = (struct r600_vertex_element*)state;
rctx->vertex_elements = v;
if (v) {
@@ -164,7 +164,7 @@ void *si_create_vertex_elements(struct pipe_context *ctx,
const struct pipe_vertex_element *elements)
{
struct r600_context *rctx = (struct r600_context *)ctx;
- struct r600_vertex_element *v = CALLOC_STRUCT(r600_vertex_element);
+ struct si_vertex_element *v = CALLOC_STRUCT(si_vertex_element);
assert(count < 32);
if (!v)
@@ -176,75 +176,6 @@ void *si_create_vertex_elements(struct pipe_context *ctx,
return v;
}
-void *si_create_shader_state(struct pipe_context *ctx,
- const struct pipe_shader_state *state)
-{
- struct si_pipe_shader *shader = CALLOC_STRUCT(si_pipe_shader);
-
- shader->tokens = tgsi_dup_tokens(state->tokens);
- shader->so = state->stream_output;
-
- return shader;
-}
-
-void r600_bind_ps_shader(struct pipe_context *ctx, void *state)
-{
- struct r600_context *rctx = (struct r600_context *)ctx;
-
- if (rctx->ps_shader != state)
- rctx->shader_dirty = true;
-
- /* TODO delete old shader */
- rctx->ps_shader = (struct si_pipe_shader *)state;
- if (state) {
- r600_inval_shader_cache(rctx);
- r600_context_pipe_state_set(rctx, &rctx->ps_shader->rstate);
- }
-}
-
-void r600_bind_vs_shader(struct pipe_context *ctx, void *state)
-{
- struct r600_context *rctx = (struct r600_context *)ctx;
-
- if (rctx->vs_shader != state)
- rctx->shader_dirty = true;
-
- /* TODO delete old shader */
- rctx->vs_shader = (struct si_pipe_shader *)state;
- if (state) {
- r600_inval_shader_cache(rctx);
- r600_context_pipe_state_set(rctx, &rctx->vs_shader->rstate);
- }
-}
-
-void r600_delete_ps_shader(struct pipe_context *ctx, void *state)
-{
- struct r600_context *rctx = (struct r600_context *)ctx;
- struct si_pipe_shader *shader = (struct si_pipe_shader *)state;
-
- if (rctx->ps_shader == shader) {
- rctx->ps_shader = NULL;
- }
-
- free(shader->tokens);
- si_pipe_shader_destroy(ctx, shader);
- free(shader);
-}
-
-void r600_delete_vs_shader(struct pipe_context *ctx, void *state)
-{
- struct r600_context *rctx = (struct r600_context *)ctx;
- struct si_pipe_shader *shader = (struct si_pipe_shader *)state;
-
- if (rctx->vs_shader == shader) {
- rctx->vs_shader = NULL;
- }
-
- free(shader->tokens);
- si_pipe_shader_destroy(ctx, shader);
- free(shader);
-}
-
static void r600_update_alpha_ref(struct r600_context *rctx)
{
#if 0
@@ -504,14 +435,10 @@ static void si_update_derived_state(struct r600_context *rctx)
if (!rctx->vs_shader->bo) {
si_pipe_shader_vs(ctx, rctx->vs_shader);
-
- r600_context_pipe_state_set(rctx, &rctx->vs_shader->rstate);
}
if (!rctx->ps_shader->bo) {
si_pipe_shader_ps(ctx, rctx->ps_shader);
-
- r600_context_pipe_state_set(rctx, &rctx->ps_shader->rstate);
}
if (rctx->shader_dirty) {
diff --git a/src/gallium/drivers/radeonsi/radeonsi_pipe.h b/src/gallium/drivers/radeonsi/radeonsi_pipe.h
index 972ca0166da..6adc7a4aea8 100644
--- a/src/gallium/drivers/radeonsi/radeonsi_pipe.h
+++ b/src/gallium/drivers/radeonsi/radeonsi_pipe.h
@@ -116,46 +116,6 @@ struct si_pipe_sampler_state {
uint32_t val[4];
};
-struct r600_vertex_element
-{
- unsigned count;
- struct pipe_vertex_element elements[PIPE_MAX_ATTRIBS];
-};
-
-struct r600_shader_io {
- unsigned name;
- unsigned gpr;
- unsigned done;
- int sid;
- unsigned param_offset;
- unsigned interpolate;
- boolean centroid;
-};
-
-struct r600_shader {
- unsigned ninput;
- unsigned noutput;
- struct r600_shader_io input[32];
- struct r600_shader_io output[32];
- boolean uses_kill;
- boolean fs_write_all;
- unsigned nr_cbufs;
-};
-
-struct si_pipe_shader {
- struct r600_shader shader;
- struct r600_pipe_state rstate;
- struct r600_resource *bo;
- struct r600_vertex_element vertex_elements;
- struct tgsi_token *tokens;
- unsigned num_sgprs;
- unsigned num_vgprs;
- unsigned spi_ps_input_ena;
- unsigned sprite_coord_enable;
- struct pipe_stream_output_info so;
- unsigned so_strides[4];
-};
-
/* needed for blitter save */
#define NUM_TEX_UNITS 16
@@ -196,7 +156,7 @@ struct r600_context {
struct r600_screen *screen;
struct radeon_winsys *ws;
struct r600_pipe_state *states[R600_PIPE_NSTATES];
- struct r600_vertex_element *vertex_elements;
+ struct si_vertex_element *vertex_elements;
struct pipe_framebuffer_state framebuffer;
unsigned pa_sc_line_stipple;
unsigned pa_su_sc_mode_cntl;
@@ -368,8 +328,6 @@ void r600_sampler_view_destroy(struct pipe_context *ctx,
struct pipe_sampler_view *state);
void r600_delete_state(struct pipe_context *ctx, void *state);
void r600_bind_vertex_elements(struct pipe_context *ctx, void *state);
-void *si_create_shader_state(struct pipe_context *ctx,
- const struct pipe_shader_state *state);
void r600_bind_ps_shader(struct pipe_context *ctx, void *state);
void r600_bind_vs_shader(struct pipe_context *ctx, void *state);
void r600_delete_ps_shader(struct pipe_context *ctx, void *state);
diff --git a/src/gallium/drivers/radeonsi/radeonsi_shader.c b/src/gallium/drivers/radeonsi/radeonsi_shader.c
index 81e4699d85d..fa86b87520e 100644
--- a/src/gallium/drivers/radeonsi/radeonsi_shader.c
+++ b/src/gallium/drivers/radeonsi/radeonsi_shader.c
@@ -347,7 +347,7 @@ static LLVMValueRef fetch_constant(
static void si_llvm_emit_epilogue(struct lp_build_tgsi_context * bld_base)
{
struct si_shader_context * si_shader_ctx = si_shader_context(bld_base);
- struct r600_shader * shader = &si_shader_ctx->shader->shader;
+ struct si_shader * shader = &si_shader_ctx->shader->shader;
struct lp_build_context * base = &bld_base->base;
struct lp_build_context * uint =
&si_shader_ctx->radeon_bld.soa.bld_base.uint_bld;
@@ -634,5 +634,5 @@ void si_pipe_shader_destroy(struct pipe_context *ctx, struct si_pipe_shader *sha
{
pipe_resource_reference((struct pipe_resource**)&shader->bo, NULL);
- memset(&shader->shader,0,sizeof(struct r600_shader));
+ memset(&shader->shader,0,sizeof(struct si_shader));
}
diff --git a/src/gallium/drivers/radeonsi/si_state.c b/src/gallium/drivers/radeonsi/si_state.c
index d2b17e0dea2..159e9890f82 100644
--- a/src/gallium/drivers/radeonsi/si_state.c
+++ b/src/gallium/drivers/radeonsi/si_state.c
@@ -26,6 +26,7 @@
#include "util/u_memory.h"
#include "util/u_framebuffer.h"
+#include "tgsi/tgsi_parse.h"
#include "radeonsi_pipe.h"
#include "si_state.h"
#include "sid.h"
@@ -1290,6 +1291,244 @@ static void si_set_framebuffer_state(struct pipe_context *ctx,
si_update_fb_blend_state(rctx);
}
+/*
+ * shaders
+ */
+
+static void *si_create_shader_state(struct pipe_context *ctx,
+ const struct pipe_shader_state *state)
+{
+ struct si_pipe_shader *shader = CALLOC_STRUCT(si_pipe_shader);
+
+ shader->tokens = tgsi_dup_tokens(state->tokens);
+ shader->so = state->stream_output;
+
+ return shader;
+}
+
+static void si_bind_vs_shader(struct pipe_context *ctx, void *state)
+{
+ struct r600_context *rctx = (struct r600_context *)ctx;
+ struct si_pipe_shader *shader = state;
+
+ if (rctx->vs_shader == state)
+ return;
+
+ rctx->shader_dirty = true;
+ rctx->vs_shader = shader;
+ si_pm4_bind_state(rctx, vs, shader->pm4);
+}
+
+static void si_bind_ps_shader(struct pipe_context *ctx, void *state)
+{
+ struct r600_context *rctx = (struct r600_context *)ctx;
+ struct si_pipe_shader *shader = state;
+
+ if (rctx->ps_shader == state)
+ return;
+
+ rctx->shader_dirty = true;
+ rctx->ps_shader = shader;
+ si_pm4_bind_state(rctx, ps, shader->pm4);
+}
+
+static void si_delete_vs_shader(struct pipe_context *ctx, void *state)
+{
+ struct r600_context *rctx = (struct r600_context *)ctx;
+ struct si_pipe_shader *shader = (struct si_pipe_shader *)state;
+
+ if (rctx->vs_shader == shader) {
+ rctx->vs_shader = NULL;
+ }
+
+ si_pm4_delete_state(rctx, vs, shader->pm4);
+ free(shader->tokens);
+ si_pipe_shader_destroy(ctx, shader);
+ free(shader);
+}
+
+static void si_delete_ps_shader(struct pipe_context *ctx, void *state)
+{
+ struct r600_context *rctx = (struct r600_context *)ctx;
+ struct si_pipe_shader *shader = (struct si_pipe_shader *)state;
+
+ if (rctx->ps_shader == shader) {
+ rctx->ps_shader = NULL;
+ }
+
+ si_pm4_delete_state(rctx, ps, shader->pm4);
+ free(shader->tokens);
+ si_pipe_shader_destroy(ctx, shader);
+ free(shader);
+}
+
+void si_pipe_shader_vs(struct pipe_context *ctx, struct si_pipe_shader *shader)
+{
+ struct r600_context *rctx = (struct r600_context *)ctx;
+ struct si_pm4_state *pm4;
+ unsigned num_sgprs, num_user_sgprs;
+ unsigned nparams, i;
+ uint64_t va;
+
+ if (si_pipe_shader_create(ctx, shader))
+ return;
+
+ si_pm4_delete_state(rctx, vs, shader->pm4);
+ pm4 = shader->pm4 = CALLOC_STRUCT(si_pm4_state);
+
+ si_pm4_inval_shader_cache(pm4);
+
+ /* Certain attributes (position, psize, etc.) don't count as params.
+ * VS is required to export at least one param and r600_shader_from_tgsi()
+ * takes care of adding a dummy export.
+ */
+ for (nparams = 0, i = 0 ; i < shader->shader.noutput; i++) {
+ if (shader->shader.output[i].name != TGSI_SEMANTIC_POSITION)
+ nparams++;
+ }
+ if (nparams < 1)
+ nparams = 1;
+
+ si_pm4_set_reg(pm4, R_0286C4_SPI_VS_OUT_CONFIG,
+ S_0286C4_VS_EXPORT_COUNT(nparams - 1));
+
+ si_pm4_set_reg(pm4, R_02870C_SPI_SHADER_POS_FORMAT,
+ S_02870C_POS0_EXPORT_FORMAT(V_02870C_SPI_SHADER_4COMP) |
+ S_02870C_POS1_EXPORT_FORMAT(V_02870C_SPI_SHADER_NONE) |
+ S_02870C_POS2_EXPORT_FORMAT(V_02870C_SPI_SHADER_NONE) |
+ S_02870C_POS3_EXPORT_FORMAT(V_02870C_SPI_SHADER_NONE));
+
+ va = r600_resource_va(ctx->screen, (void *)shader->bo);
+ si_pm4_add_bo(pm4, shader->bo, RADEON_USAGE_READ);
+ si_pm4_set_reg(pm4, R_00B120_SPI_SHADER_PGM_LO_VS, va >> 8);
+ si_pm4_set_reg(pm4, R_00B124_SPI_SHADER_PGM_HI_VS, va >> 40);
+
+ num_user_sgprs = 8;
+ num_sgprs = shader->num_sgprs;
+ if (num_user_sgprs > num_sgprs)
+ num_sgprs = num_user_sgprs;
+ /* Last 2 reserved SGPRs are used for VCC */
+ num_sgprs += 2;
+ assert(num_sgprs <= 104);
+
+ si_pm4_set_reg(pm4, R_00B128_SPI_SHADER_PGM_RSRC1_VS,
+ S_00B128_VGPRS((shader->num_vgprs - 1) / 4) |
+ S_00B128_SGPRS((num_sgprs - 1) / 8));
+ si_pm4_set_reg(pm4, R_00B12C_SPI_SHADER_PGM_RSRC2_VS,
+ S_00B12C_USER_SGPR(num_user_sgprs));
+
+ si_pm4_bind_state(rctx, vs, shader->pm4);
+}
+
+void si_pipe_shader_ps(struct pipe_context *ctx, struct si_pipe_shader *shader)
+{
+ struct r600_context *rctx = (struct r600_context *)ctx;
+ struct si_pm4_state *pm4;
+ unsigned i, exports_ps, num_cout, spi_ps_in_control, db_shader_control;
+ unsigned num_sgprs, num_user_sgprs;
+ int ninterp = 0;
+ boolean have_linear = FALSE, have_centroid = FALSE, have_perspective = FALSE;
+ unsigned spi_baryc_cntl;
+ uint64_t va;
+
+ if (si_pipe_shader_create(ctx, shader))
+ return;
+
+ si_pm4_delete_state(rctx, ps, shader->pm4);
+ pm4 = shader->pm4 = CALLOC_STRUCT(si_pm4_state);
+
+ si_pm4_inval_shader_cache(pm4);
+
+ db_shader_control = S_02880C_Z_ORDER(V_02880C_EARLY_Z_THEN_LATE_Z);
+ for (i = 0; i < shader->shader.ninput; i++) {
+ ninterp++;
+ /* XXX: Flat shading hangs the GPU */
+ if (shader->shader.input[i].interpolate == TGSI_INTERPOLATE_CONSTANT ||
+ (shader->shader.input[i].interpolate == TGSI_INTERPOLATE_COLOR &&
+ rctx->queued.named.rasterizer->flatshade))
+ have_linear = TRUE;
+ if (shader->shader.input[i].interpolate == TGSI_INTERPOLATE_LINEAR)
+ have_linear = TRUE;
+ if (shader->shader.input[i].interpolate == TGSI_INTERPOLATE_PERSPECTIVE)
+ have_perspective = TRUE;
+ if (shader->shader.input[i].centroid)
+ have_centroid = TRUE;
+ }
+
+ for (i = 0; i < shader->shader.noutput; i++) {
+ if (shader->shader.output[i].name == TGSI_SEMANTIC_POSITION)
+ db_shader_control |= S_02880C_Z_EXPORT_ENABLE(1);
+ if (shader->shader.output[i].name == TGSI_SEMANTIC_STENCIL)
+ db_shader_control |= 0; // XXX OP_VAL or TEST_VAL?
+ }
+ if (shader->shader.uses_kill)
+ db_shader_control |= S_02880C_KILL_ENABLE(1);
+
+ exports_ps = 0;
+ num_cout = 0;
+ for (i = 0; i < shader->shader.noutput; i++) {
+ if (shader->shader.output[i].name == TGSI_SEMANTIC_POSITION ||
+ shader->shader.output[i].name == TGSI_SEMANTIC_STENCIL)
+ exports_ps |= 1;
+ else if (shader->shader.output[i].name == TGSI_SEMANTIC_COLOR) {
+ if (shader->shader.fs_write_all)
+ num_cout = shader->shader.nr_cbufs;
+ else
+ num_cout++;
+ }
+ }
+ if (!exports_ps) {
+ /* always at least export 1 component per pixel */
+ exports_ps = 2;
+ }
+
+ spi_ps_in_control = S_0286D8_NUM_INTERP(ninterp);
+
+ spi_baryc_cntl = 0;
+ if (have_perspective)
+ spi_baryc_cntl |= have_centroid ?
+ S_0286E0_PERSP_CENTROID_CNTL(1) : S_0286E0_PERSP_CENTER_CNTL(1);
+ if (have_linear)
+ spi_baryc_cntl |= have_centroid ?
+ S_0286E0_LINEAR_CENTROID_CNTL(1) : S_0286E0_LINEAR_CENTER_CNTL(1);
+
+ si_pm4_set_reg(pm4, R_0286E0_SPI_BARYC_CNTL, spi_baryc_cntl);
+ si_pm4_set_reg(pm4, R_0286CC_SPI_PS_INPUT_ENA, shader->spi_ps_input_ena);
+ si_pm4_set_reg(pm4, R_0286D0_SPI_PS_INPUT_ADDR, shader->spi_ps_input_ena);
+ si_pm4_set_reg(pm4, R_0286D8_SPI_PS_IN_CONTROL, spi_ps_in_control);
+
+ /* XXX: Depends on Z buffer format? */
+ si_pm4_set_reg(pm4, R_028710_SPI_SHADER_Z_FORMAT, 0);
+
+ /* XXX: Depends on color buffer format? */
+ si_pm4_set_reg(pm4, R_028714_SPI_SHADER_COL_FORMAT,
+ S_028714_COL0_EXPORT_FORMAT(V_028714_SPI_SHADER_32_ABGR));
+
+ va = r600_resource_va(ctx->screen, (void *)shader->bo);
+ si_pm4_add_bo(pm4, shader->bo, RADEON_USAGE_READ);
+ si_pm4_set_reg(pm4, R_00B020_SPI_SHADER_PGM_LO_PS, va >> 8);
+ si_pm4_set_reg(pm4, R_00B024_SPI_SHADER_PGM_HI_PS, va >> 40);
+
+ num_user_sgprs = 6;
+ num_sgprs = shader->num_sgprs;
+ if (num_user_sgprs > num_sgprs)
+ num_sgprs = num_user_sgprs;
+ /* Last 2 reserved SGPRs are used for VCC */
+ num_sgprs += 2;
+ assert(num_sgprs <= 104);
+
+ si_pm4_set_reg(pm4, R_00B028_SPI_SHADER_PGM_RSRC1_PS,
+ S_00B028_VGPRS((shader->num_vgprs - 1) / 4) |
+ S_00B028_SGPRS((num_sgprs - 1) / 8));
+ si_pm4_set_reg(pm4, R_00B02C_SPI_SHADER_PGM_RSRC2_PS,
+ S_00B02C_USER_SGPR(num_user_sgprs));
+
+ si_pm4_set_reg(pm4, R_02880C_DB_SHADER_CONTROL, db_shader_control);
+
+ shader->sprite_coord_enable = rctx->sprite_coord_enable;
+ si_pm4_bind_state(rctx, ps, shader->pm4);
+}
+
void si_init_state_functions(struct r600_context *rctx)
{
rctx->context.create_blend_state = si_create_blend_state;
@@ -1312,6 +1551,13 @@ void si_init_state_functions(struct r600_context *rctx)
rctx->context.set_stencil_ref = si_set_pipe_stencil_ref;
rctx->context.set_framebuffer_state = si_set_framebuffer_state;
+
+ rctx->context.create_vs_state = si_create_shader_state;
+ rctx->context.create_fs_state = si_create_shader_state;
+ rctx->context.bind_vs_state = si_bind_vs_shader;
+ rctx->context.bind_fs_state = si_bind_ps_shader;
+ rctx->context.delete_vs_state = si_delete_vs_shader;
+ rctx->context.delete_fs_state = si_delete_ps_shader;
}
void si_init_config(struct r600_context *rctx)
@@ -1434,8 +1680,8 @@ bool si_update_draw_info_state(struct r600_context *rctx,
void si_update_spi_map(struct r600_context *rctx)
{
- struct r600_shader *ps = &rctx->ps_shader->shader;
- struct r600_shader *vs = &rctx->vs_shader->shader;
+ struct si_shader *ps = &rctx->ps_shader->shader;
+ struct si_shader *vs = &rctx->vs_shader->shader;
struct si_pm4_state *pm4 = CALLOC_STRUCT(si_pm4_state);
unsigned i, j, tmp;
diff --git a/src/gallium/drivers/radeonsi/si_state.h b/src/gallium/drivers/radeonsi/si_state.h
index d8f03392a4c..cd2b9626a92 100644
--- a/src/gallium/drivers/radeonsi/si_state.h
+++ b/src/gallium/drivers/radeonsi/si_state.h
@@ -61,6 +61,46 @@ struct si_state_dsa {
uint8_t writemask[2];
};
+struct si_vertex_element
+{
+ unsigned count;
+ struct pipe_vertex_element elements[PIPE_MAX_ATTRIBS];
+};
+
+struct si_shader_io {
+ unsigned name;
+ unsigned gpr;
+ unsigned done;
+ int sid;
+ unsigned param_offset;
+ unsigned interpolate;
+ bool centroid;
+};
+
+struct si_shader {
+ unsigned ninput;
+ unsigned noutput;
+ struct si_shader_io input[32];
+ struct si_shader_io output[32];
+ bool uses_kill;
+ bool fs_write_all;
+ unsigned nr_cbufs;
+};
+
+struct si_pipe_shader {
+ struct si_shader shader;
+ struct si_pm4_state *pm4;
+ struct r600_resource *bo;
+ struct si_vertex_element vertex_elements;
+ struct tgsi_token *tokens;
+ unsigned num_sgprs;
+ unsigned num_vgprs;
+ unsigned spi_ps_input_ena;
+ unsigned sprite_coord_enable;
+ struct pipe_stream_output_info so;
+ unsigned so_strides[4];
+};
+
union si_state {
struct {
struct si_pm4_state *init;
@@ -75,6 +115,8 @@ union si_state {
struct si_pm4_state *fb_rs;
struct si_pm4_state *fb_blend;
struct si_pm4_state *dsa_stencil_ref;
+ struct si_pm4_state *vs;
+ struct si_pm4_state *ps;
struct si_pm4_state *spi;
struct si_pm4_state *draw_info;
} named;