From: Dave Airlie <airl...@redhat.com>
This adds support for load/store/atomic operations on images
along with image tracking support.
Signed-off-by: Dave Airlie <airl...@redhat.com>
---
src/gallium/auxiliary/draw/draw_gs.c | 2 +-
src/gallium/auxiliary/draw/draw_vs_exec.c | 2 +-
src/gallium/auxiliary/tgsi/tgsi_exec.c | 229 +++++++++++++++++++++++++++++-
src/gallium/auxiliary/tgsi/tgsi_exec.h | 40 +++++-
src/gallium/drivers/softpipe/sp_fs_exec.c | 4 +-
5 files changed, 271 insertions(+), 6 deletions(-)
diff --git a/src/gallium/auxiliary/draw/draw_gs.c
b/src/gallium/auxiliary/draw/draw_gs.c
index 6b33341..c4ced9f 100644
--- a/src/gallium/auxiliary/draw/draw_gs.c
+++ b/src/gallium/auxiliary/draw/draw_gs.c
@@ -687,7 +687,7 @@ void draw_geometry_shader_prepare(struct
draw_geometry_shader *shader,
if (!use_llvm && shader && shader->machine->Tokens !=
shader->state.tokens) {
tgsi_exec_machine_bind_shader(shader->machine,
shader->state.tokens,
- draw->gs.tgsi.sampler);
+ draw->gs.tgsi.sampler, NULL);
}
}
diff --git a/src/gallium/auxiliary/draw/draw_vs_exec.c
b/src/gallium/auxiliary/draw/draw_vs_exec.c
index abd64f5..8c759d4 100644
--- a/src/gallium/auxiliary/draw/draw_vs_exec.c
+++ b/src/gallium/auxiliary/draw/draw_vs_exec.c
@@ -70,7 +70,7 @@ vs_exec_prepare( struct draw_vertex_shader *shader,
if (evs->machine->Tokens != shader->state.tokens) {
tgsi_exec_machine_bind_shader(evs->machine,
shader->state.tokens,
- draw->vs.tgsi.sampler);
+ draw->vs.tgsi.sampler, NULL);
}
}
diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.c
b/src/gallium/auxiliary/tgsi/tgsi_exec.c
index fa1c916..fe82a95 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_exec.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_exec.c
@@ -853,7 +853,8 @@ void
tgsi_exec_machine_bind_shader(
struct tgsi_exec_machine *mach,
const struct tgsi_token *tokens,
- struct tgsi_sampler *sampler)
+ struct tgsi_sampler *sampler,
+ struct tgsi_image *image)
{
uint k;
struct tgsi_parse_context parse;
@@ -871,6 +872,7 @@ tgsi_exec_machine_bind_shader(
mach->Tokens = tokens;
mach->Sampler = sampler;
+ mach->Image = image;
if (!tokens) {
/* unbind and free all */
@@ -3706,6 +3708,206 @@ exec_dfracexp(struct tgsi_exec_machine *mach,
}
}
+static int
+get_image_coord_dim(int tgsi_tex, int *sample)
+{
+ int dim;
+ switch (tgsi_tex) {
+ case TGSI_TEXTURE_BUFFER:
+ case TGSI_TEXTURE_1D:
+ dim = 1;
+ break;
+ case TGSI_TEXTURE_2D:
+ case TGSI_TEXTURE_RECT:
+ case TGSI_TEXTURE_1D_ARRAY:
+ case TGSI_TEXTURE_2D_MSAA:
+ dim = 2;
+ break;
+ case TGSI_TEXTURE_3D:
+ case TGSI_TEXTURE_CUBE:
+ case TGSI_TEXTURE_2D_ARRAY:
+ case TGSI_TEXTURE_2D_ARRAY_MSAA:
+ case TGSI_TEXTURE_CUBE_ARRAY:
+ dim = 3;
+ break;
+ default:
+ assert(!"unknown texture target");
+ dim = 0;
+ break;
+ }
+
+ if (sample) {
+ switch (tgsi_tex) {
+ case TGSI_TEXTURE_2D_MSAA:
+ *sample = 3;
+ break;
+ case TGSI_TEXTURE_2D_ARRAY_MSAA:
+ *sample = 4;
+ break;
+ default:
+ *sample = 0;
+ break;
+ }
+ }
+ return dim;
+}
+
+static void
+exec_load(struct tgsi_exec_machine *mach,
+ const struct tgsi_full_instruction *inst)
+{
+ union tgsi_exec_channel r[4], sample_r;
+ uint unit;
+ int sample;
+ int i, j;
+ int dim;
+ uint chan;
+ float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE];
+ struct tgsi_image_params params;
+ int kilmask = mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0];
+
+ unit = fetch_sampler_unit(mach, inst, 0);
+ dim = get_image_coord_dim(inst->Memory.Texture, &sample);
+ assert(dim <= 3);
+
+ params.execmask = mach->ExecMask & mach->NonHelperMask & ~kilmask;
+ params.unit = unit;
+ params.tgsi_tex_instr = inst->Memory.Texture;
+ params.format = inst->Memory.Format;
+
+ for (i = 0; i < dim; i++) {
+ IFETCH(&r[i], 1, TGSI_CHAN_X + i);
+ }
+
+ if (sample)
+ IFETCH(&sample_r, 1, TGSI_CHAN_X + sample);
+
+ mach->Image->load(mach->Image, ¶ms,
+ r[0].i, r[1].i, r[2].i, sample_r.i,
+ rgba);
+ for (j = 0; j < TGSI_QUAD_SIZE; j++) {
+ r[0].f[j] = rgba[0][j];
+ r[1].f[j] = rgba[1][j];
+ r[2].f[j] = rgba[2][j];
+ r[3].f[j] = rgba[3][j];
+ }
+ for (chan = 0; chan < TGSI_NUM_CHANNELS; chan++) {
+ if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
+ store_dest(mach, &r[chan], &inst->Dst[0], inst, chan,
TGSI_EXEC_DATA_FLOAT);
+ }
+ }
+}
+
+static void
+exec_store(struct tgsi_exec_machine *mach,
+ const struct tgsi_full_instruction *inst)
+{
+ union tgsi_exec_channel r[3], sample_r;
+ union tgsi_exec_channel value[4];
+ float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE];
+ struct tgsi_image_params params;
+ int dim;
+ int sample;
+ int i, j;
+ uint unit;
+ int kilmask = mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0];
+ unit = inst->Dst[0].Register.Index;
+ dim = get_image_coord_dim(inst->Memory.Texture, &sample);
+ assert(dim <= 3);
+
+ params.execmask = mach->ExecMask & mach->NonHelperMask & ~kilmask;
+ params.unit = unit;
+ params.tgsi_tex_instr = inst->Memory.Texture;
+ params.format = inst->Memory.Format;
+
+ for (i = 0; i < dim; i++) {
+ IFETCH(&r[i], 0, TGSI_CHAN_X + i);
+ }
+
+ for (i = 0; i < 4; i++) {
+ FETCH(&value[i], 1, TGSI_CHAN_X + i);
+ }
+ if (sample)
+ IFETCH(&sample_r, 0, TGSI_CHAN_X + sample);
+
+ for (j = 0; j < TGSI_QUAD_SIZE; j++) {
+ rgba[0][j] = value[0].f[j];
+ rgba[1][j] = value[1].f[j];
+ rgba[2][j] = value[2].f[j];
+ rgba[3][j] = value[3].f[j];
+ }
+
+ mach->Image->store(mach->Image, ¶ms,
+ r[0].i, r[1].i, r[2].i, sample_r.i,
+ rgba);
+}
+
+static void
+exec_atomop(struct tgsi_exec_machine *mach,
+ const struct tgsi_full_instruction *inst)
+{
+ union tgsi_exec_channel r[3], sample_r;
+ union tgsi_exec_channel value[4], value2[4];
+ float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE];
+ float rgba2[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE];
+ struct tgsi_image_params params;
+ int dim;
+ int sample;
+ int i, j;
+ uint unit, chan;
+ int kilmask = mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0];
+ unit = fetch_sampler_unit(mach, inst, 0);
+ dim = get_image_coord_dim(inst->Memory.Texture, &sample);
+ assert(dim <= 3);
+
+ params.execmask = mach->ExecMask & mach->NonHelperMask & ~kilmask;
+ params.unit = unit;
+ params.tgsi_tex_instr = inst->Memory.Texture;
+ params.format = inst->Memory.Format;
+
+ for (i = 0; i < dim; i++) {
+ IFETCH(&r[i], 1, TGSI_CHAN_X + i);
+ }
+
+ for (i = 0; i < 4; i++) {
+ FETCH(&value[i], 2, TGSI_CHAN_X + i);
+ if (inst->Instruction.Opcode == TGSI_OPCODE_ATOMCAS)
+ FETCH(&value2[i], 3, TGSI_CHAN_X + i);
+ }
+ if (sample)
+ IFETCH(&sample_r, 1, TGSI_CHAN_X + sample);
+
+ for (j = 0; j < TGSI_QUAD_SIZE; j++) {
+ rgba[0][j] = value[0].f[j];
+ rgba[1][j] = value[1].f[j];
+ rgba[2][j] = value[2].f[j];
+ rgba[3][j] = value[3].f[j];
+ }
+ if (inst->Instruction.Opcode == TGSI_OPCODE_ATOMCAS) {
+ for (j = 0; j < TGSI_QUAD_SIZE; j++) {
+ rgba2[0][j] = value2[0].f[j];
+ rgba2[1][j] = value2[1].f[j];
+ rgba2[2][j] = value2[2].f[j];
+ rgba2[3][j] = value2[3].f[j];
+ }
+ }
+
+ mach->Image->op(mach->Image, ¶ms, inst->Instruction.Opcode,
+ r[0].i, r[1].i, r[2].i, sample_r.i,
+ rgba, rgba2);
+
+ for (j = 0; j < TGSI_QUAD_SIZE; j++) {
+ r[0].f[j] = rgba[0][j];
+ r[1].f[j] = rgba[1][j];
+ r[2].f[j] = rgba[2][j];
+ r[3].f[j] = rgba[3][j];
+ }
+ for (chan = 0; chan < TGSI_NUM_CHANNELS; chan++) {
+ if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
+ store_dest(mach, &r[chan], &inst->Dst[0], inst, chan,
TGSI_EXEC_DATA_FLOAT);
+ }
+ }
+}
static void
micro_i2f(union tgsi_exec_channel *dst,
@@ -5172,6 +5374,31 @@ exec_instruction(
case TGSI_OPCODE_D2U:
exec_d2u(mach, inst);
break;
+
+ case TGSI_OPCODE_LOAD:
+ exec_load(mach, inst);
+ break;
+
+ case TGSI_OPCODE_STORE:
+ exec_store(mach, inst);
+ break;
+
+ case TGSI_OPCODE_ATOMUADD:
+ case TGSI_OPCODE_ATOMXCHG:
+ case TGSI_OPCODE_ATOMCAS:
+ case TGSI_OPCODE_ATOMAND:
+ case TGSI_OPCODE_ATOMOR:
+ case TGSI_OPCODE_ATOMXOR:
+ case TGSI_OPCODE_ATOMUMIN:
+ case TGSI_OPCODE_ATOMUMAX:
+ case TGSI_OPCODE_ATOMIMIN:
+ case TGSI_OPCODE_ATOMIMAX:
+ exec_atomop(mach, inst);
+ break;
+
+ case TGSI_OPCODE_BARRIER:
+ case TGSI_OPCODE_MEMBAR:
+ break;
default:
assert( 0 );
}
diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.h
b/src/gallium/auxiliary/tgsi/tgsi_exec.h
index 05ae388..9ff8a72 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_exec.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_exec.h
@@ -98,6 +98,42 @@ enum tgsi_sampler_control
TGSI_SAMPLER_GATHER,
};
+struct tgsi_image_params {
+ unsigned unit;
+ unsigned tgsi_tex_instr;
+ enum pipe_format format;
+ unsigned execmask;
+};
+
+struct tgsi_image {
+ /* image interfaces */
+ void (*load)(struct tgsi_image *image,
+ struct tgsi_image_params *params,
+ const int s[TGSI_QUAD_SIZE],
+ const int t[TGSI_QUAD_SIZE],
+ const int r[TGSI_QUAD_SIZE],
+ const int sample[TGSI_QUAD_SIZE],
+ float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE]);
+
+ void (*store)(struct tgsi_image *image,
+ struct tgsi_image_params *params,
+ const int s[TGSI_QUAD_SIZE],
+ const int t[TGSI_QUAD_SIZE],
+ const int r[TGSI_QUAD_SIZE],
+ const int sample[TGSI_QUAD_SIZE],
+ float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE]);
+
+ void (*op)(struct tgsi_image *image,
+ struct tgsi_image_params *params,
+ unsigned opcode,
+ const int s[TGSI_QUAD_SIZE],
+ const int t[TGSI_QUAD_SIZE],
+ const int r[TGSI_QUAD_SIZE],
+ const int sample[TGSI_QUAD_SIZE],
+ float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE],
+ float rgba2[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE]);
+};
+
/**
* Information for sampling textures, which must be implemented
* by code outside the TGSI executor.
@@ -298,6 +334,7 @@ struct tgsi_exec_machine
struct tgsi_sampler *Sampler;
+ struct tgsi_image *Image;
unsigned ImmLimit;
const void *Consts[PIPE_MAX_CONSTANT_BUFFERS];
@@ -386,7 +423,8 @@ void
tgsi_exec_machine_bind_shader(
struct tgsi_exec_machine *mach,
const struct tgsi_token *tokens,
- struct tgsi_sampler *sampler);
+ struct tgsi_sampler *sampler,
+ struct tgsi_image *image);
uint
tgsi_exec_machine_run(
diff --git a/src/gallium/drivers/softpipe/sp_fs_exec.c
b/src/gallium/drivers/softpipe/sp_fs_exec.c
index e2d527d..2c5bf7e 100644
--- a/src/gallium/drivers/softpipe/sp_fs_exec.c
+++ b/src/gallium/drivers/softpipe/sp_fs_exec.c
@@ -69,7 +69,7 @@ exec_prepare( const struct sp_fragment_shader_variant *var,
*/
tgsi_exec_machine_bind_shader(machine,
var->tokens,
- sampler);
+ sampler, NULL);
}
@@ -184,7 +184,7 @@ exec_delete(struct sp_fragment_shader_variant *var,
struct tgsi_exec_machine *machine)
{
if (machine->Tokens == var->tokens) {
- tgsi_exec_machine_bind_shader(machine, NULL, NULL);
+ tgsi_exec_machine_bind_shader(machine, NULL, NULL, NULL);
}
FREE( (void *) var->tokens );
