[Mesa-dev] [PATCH 2/4] tgsi: add support for buffer/atomic operations to tgsi_exec.
Dave Airlie
airlied at gmail.com
Mon Apr 11 03:20:20 UTC 2016
From: Dave Airlie <airlied at redhat.com>
This adds support for doing load/store/atomic operations on
buffer objects.
Signed-off-by: Dave Airlie <airlied at 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 | 217 ++++++++++++++++++++++++++++--
src/gallium/auxiliary/tgsi/tgsi_exec.h | 34 ++++-
src/gallium/drivers/softpipe/sp_fs_exec.c | 4 +-
5 files changed, 245 insertions(+), 14 deletions(-)
diff --git a/src/gallium/auxiliary/draw/draw_gs.c b/src/gallium/auxiliary/draw/draw_gs.c
index 14db2d6..a555b26 100644
--- a/src/gallium/auxiliary/draw/draw_gs.c
+++ b/src/gallium/auxiliary/draw/draw_gs.c
@@ -681,7 +681,7 @@ 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.image);
+ draw->gs.tgsi.sampler, draw->gs.tgsi.image, NULL);
}
}
diff --git a/src/gallium/auxiliary/draw/draw_vs_exec.c b/src/gallium/auxiliary/draw/draw_vs_exec.c
index 4cd755e..feb222b8 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.image);
+ draw->vs.tgsi.sampler, draw->vs.tgsi.image, NULL);
}
}
diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.c b/src/gallium/auxiliary/tgsi/tgsi_exec.c
index a595bbb..fb51051 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_exec.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_exec.c
@@ -854,7 +854,8 @@ tgsi_exec_machine_bind_shader(
struct tgsi_exec_machine *mach,
const struct tgsi_token *tokens,
struct tgsi_sampler *sampler,
- struct tgsi_image *image)
+ struct tgsi_image *image,
+ struct tgsi_buffer *buffer)
{
uint k;
struct tgsi_parse_context parse;
@@ -873,6 +874,7 @@ tgsi_exec_machine_bind_shader(
mach->Tokens = tokens;
mach->Sampler = sampler;
mach->Image = image;
+ mach->Buffer = buffer;
if (!tokens) {
/* unbind and free all */
@@ -3758,8 +3760,8 @@ get_image_coord_sample(unsigned tgsi_tex)
}
static void
-exec_load(struct tgsi_exec_machine *mach,
- const struct tgsi_full_instruction *inst)
+exec_load_img(struct tgsi_exec_machine *mach,
+ const struct tgsi_full_instruction *inst)
{
union tgsi_exec_channel r[4], sample_r;
uint unit;
@@ -3805,8 +3807,51 @@ exec_load(struct tgsi_exec_machine *mach,
}
static void
-exec_store(struct tgsi_exec_machine *mach,
- const struct tgsi_full_instruction *inst)
+exec_load_buf(struct tgsi_exec_machine *mach,
+ const struct tgsi_full_instruction *inst)
+{
+ union tgsi_exec_channel r[4];
+ uint unit;
+ int j;
+ uint chan;
+ float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE];
+ struct tgsi_buffer_params params;
+ int kilmask = mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0];
+
+ unit = fetch_sampler_unit(mach, inst, 0);
+
+ params.execmask = mach->ExecMask & mach->NonHelperMask & ~kilmask;
+ params.unit = unit;
+ IFETCH(&r[0], 1, TGSI_CHAN_X);
+
+ mach->Buffer->load(mach->Buffer, ¶ms,
+ r[0].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_load(struct tgsi_exec_machine *mach,
+ const struct tgsi_full_instruction *inst)
+{
+ if (inst->Src[0].Register.File == TGSI_FILE_IMAGE)
+ exec_load_img(mach, inst);
+ else
+ exec_load_buf(mach, inst);
+}
+
+static void
+exec_store_img(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];
@@ -3850,8 +3895,53 @@ exec_store(struct tgsi_exec_machine *mach,
}
static void
-exec_atomop(struct tgsi_exec_machine *mach,
- const struct tgsi_full_instruction *inst)
+exec_store_buf(struct tgsi_exec_machine *mach,
+ const struct tgsi_full_instruction *inst)
+{
+ union tgsi_exec_channel r[3];
+ union tgsi_exec_channel value[4];
+ float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE];
+ struct tgsi_buffer_params params;
+ int i, j;
+ uint unit;
+ int kilmask = mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0];
+
+ unit = inst->Dst[0].Register.Index;
+
+ params.execmask = mach->ExecMask & mach->NonHelperMask & ~kilmask;
+ params.unit = unit;
+ params.writemask = inst->Dst[0].Register.WriteMask;
+
+ IFETCH(&r[0], 0, TGSI_CHAN_X);
+ for (i = 0; i < 4; i++) {
+ FETCH(&value[i], 1, TGSI_CHAN_X + i);
+ }
+
+ 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->Buffer->store(mach->Buffer, ¶ms,
+ r[0].i,
+ rgba);
+}
+
+static void
+exec_store(struct tgsi_exec_machine *mach,
+ const struct tgsi_full_instruction *inst)
+{
+ if (inst->Dst[0].Register.File == TGSI_FILE_IMAGE)
+ exec_store_img(mach, inst);
+ else
+ exec_store_buf(mach, inst);
+}
+
+static void
+exec_atomop_img(struct tgsi_exec_machine *mach,
+ const struct tgsi_full_instruction *inst)
{
union tgsi_exec_channel r[4], sample_r;
union tgsi_exec_channel value[4], value2[4];
@@ -3918,8 +4008,77 @@ exec_atomop(struct tgsi_exec_machine *mach,
}
static void
-exec_resq(struct tgsi_exec_machine *mach,
- const struct tgsi_full_instruction *inst)
+exec_atomop_buf(struct tgsi_exec_machine *mach,
+ const struct tgsi_full_instruction *inst)
+{
+ union tgsi_exec_channel r[3];
+ 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_buffer_params params;
+ 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);
+
+ params.execmask = mach->ExecMask & mach->NonHelperMask & ~kilmask;
+ params.unit = unit;
+ params.writemask = inst->Dst[0].Register.WriteMask;
+
+ IFETCH(&r[0], 1, TGSI_CHAN_X);
+
+ 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);
+ }
+
+ 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->Buffer->op(mach->Buffer, ¶ms, inst->Instruction.Opcode,
+ r[0].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
+exec_atomop(struct tgsi_exec_machine *mach,
+ const struct tgsi_full_instruction *inst)
+{
+ if (inst->Src[0].Register.File == TGSI_FILE_IMAGE)
+ exec_atomop_img(mach, inst);
+ else
+ exec_atomop_buf(mach, inst);
+}
+
+static void
+exec_resq_img(struct tgsi_exec_machine *mach,
+ const struct tgsi_full_instruction *inst)
{
int result[4];
union tgsi_exec_channel r[4];
@@ -3952,6 +4111,46 @@ exec_resq(struct tgsi_exec_machine *mach,
}
static void
+exec_resq_buf(struct tgsi_exec_machine *mach,
+ const struct tgsi_full_instruction *inst)
+{
+ int result;
+ union tgsi_exec_channel r[4];
+ uint unit;
+ int i, chan;
+ struct tgsi_buffer_params params;
+ int kilmask = mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0];
+
+ unit = fetch_sampler_unit(mach, inst, 0);
+
+ params.execmask = mach->ExecMask & mach->NonHelperMask & ~kilmask;
+ params.unit = unit;
+
+ mach->Buffer->get_dims(mach->Buffer, ¶ms, &result);
+
+ for (i = 0; i < TGSI_QUAD_SIZE; i++) {
+ r[0].i[i] = result;
+ }
+
+ 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_INT);
+ }
+ }
+}
+
+static void
+exec_resq(struct tgsi_exec_machine *mach,
+ const struct tgsi_full_instruction *inst)
+{
+ if (inst->Src[0].Register.File == TGSI_FILE_IMAGE)
+ exec_resq_img(mach, inst);
+ else
+ exec_resq_buf(mach, inst);
+}
+
+static void
micro_i2f(union tgsi_exec_channel *dst,
const union tgsi_exec_channel *src)
{
diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.h b/src/gallium/auxiliary/tgsi/tgsi_exec.h
index 45fb8d4..257c58f 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_exec.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_exec.h
@@ -138,6 +138,36 @@ struct tgsi_image {
int dims[4]);
};
+struct tgsi_buffer_params {
+ unsigned unit;
+ unsigned execmask;
+ unsigned writemask;
+};
+
+struct tgsi_buffer {
+ /* buffer interfaces */
+ void (*load)(const struct tgsi_buffer *buffer,
+ const struct tgsi_buffer_params *params,
+ const int s[TGSI_QUAD_SIZE],
+ float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE]);
+
+ void (*store)(const struct tgsi_buffer *buffer,
+ const struct tgsi_buffer_params *params,
+ const int s[TGSI_QUAD_SIZE],
+ float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE]);
+
+ void (*op)(const struct tgsi_buffer *buffer,
+ const struct tgsi_buffer_params *params,
+ unsigned opcode,
+ const int s[TGSI_QUAD_SIZE],
+ float rgba[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE],
+ float rgba2[TGSI_NUM_CHANNELS][TGSI_QUAD_SIZE]);
+
+ void (*get_dims)(const struct tgsi_buffer *buffer,
+ const struct tgsi_buffer_params *params,
+ int *dim);
+};
+
/**
* Information for sampling textures, which must be implemented
* by code outside the TGSI executor.
@@ -334,6 +364,7 @@ struct tgsi_exec_machine
struct tgsi_sampler *Sampler;
struct tgsi_image *Image;
+ struct tgsi_buffer *Buffer;
unsigned ImmLimit;
const void *Consts[PIPE_MAX_CONSTANT_BUFFERS];
@@ -424,7 +455,8 @@ tgsi_exec_machine_bind_shader(
struct tgsi_exec_machine *mach,
const struct tgsi_token *tokens,
struct tgsi_sampler *sampler,
- struct tgsi_image *image);
+ struct tgsi_image *image,
+ struct tgsi_buffer *buffer);
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 bfd9a4b..6a5f7ac 100644
--- a/src/gallium/drivers/softpipe/sp_fs_exec.c
+++ b/src/gallium/drivers/softpipe/sp_fs_exec.c
@@ -70,7 +70,7 @@ exec_prepare( const struct sp_fragment_shader_variant *var,
*/
tgsi_exec_machine_bind_shader(machine,
var->tokens,
- sampler, image);
+ sampler, image, NULL);
}
@@ -186,7 +186,7 @@ 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, NULL);
+ tgsi_exec_machine_bind_shader(machine, NULL, NULL, NULL, NULL);
}
FREE( (void *) var->tokens );
--
2.5.5
More information about the mesa-dev
mailing list