#include "util/u_memory.h"
+static void r600_set_streamout_enable(struct r600_common_context *rctx, bool enable);
+
static struct pipe_stream_output_target *
r600_create_so_target(struct pipe_context *ctx,
struct pipe_resource *buffer,
void r600_streamout_buffers_dirty(struct r600_common_context *rctx)
{
+ struct r600_atom *begin = &rctx->streamout.begin_atom;
+ unsigned num_bufs = util_bitcount(rctx->streamout.enabled_mask);
+ unsigned num_bufs_appended = util_bitcount(rctx->streamout.enabled_mask &
+ rctx->streamout.append_bitmask);
+
+ if (!num_bufs)
+ return;
+
rctx->streamout.num_dw_for_end =
12 + /* flush_vgt_streamout */
- util_bitcount(rctx->streamout.enabled_mask) * 8 + /* STRMOUT_BUFFER_UPDATE */
- 3 /* set_streamout_enable(0) */;
+ num_bufs * 11; /* STRMOUT_BUFFER_UPDATE, BUFFER_SIZE */
- rctx->streamout.begin_atom.num_dw =
- 12 + /* flush_vgt_streamout */
- 6 + /* set_streamout_enable */
- util_bitcount(rctx->streamout.enabled_mask) * 7 + /* SET_CONTEXT_REG */
- (rctx->family >= CHIP_RS780 &&
- rctx->family <= CHIP_RV740 ? util_bitcount(rctx->streamout.enabled_mask) * 5 : 0) + /* STRMOUT_BASE_UPDATE */
- util_bitcount(rctx->streamout.enabled_mask & rctx->streamout.append_bitmask) * 8 + /* STRMOUT_BUFFER_UPDATE */
- util_bitcount(rctx->streamout.enabled_mask & ~rctx->streamout.append_bitmask) * 6 + /* STRMOUT_BUFFER_UPDATE */
- (rctx->family > CHIP_R600 && rctx->family < CHIP_RS780 ? 2 : 0) + /* SURFACE_BASE_UPDATE */
- rctx->streamout.num_dw_for_end;
-
- rctx->streamout.begin_atom.dirty = true;
+ begin->num_dw = 12; /* flush_vgt_streamout */
+
+ if (rctx->chip_class >= SI) {
+ begin->num_dw += num_bufs * 4; /* SET_CONTEXT_REG */
+ } else {
+ begin->num_dw += num_bufs * 7; /* SET_CONTEXT_REG */
+
+ if (rctx->family >= CHIP_RS780 && rctx->family <= CHIP_RV740)
+ begin->num_dw += num_bufs * 5; /* STRMOUT_BASE_UPDATE */
+ }
+
+ begin->num_dw +=
+ num_bufs_appended * 8 + /* STRMOUT_BUFFER_UPDATE */
+ (num_bufs - num_bufs_appended) * 6 + /* STRMOUT_BUFFER_UPDATE */
+ (rctx->family > CHIP_R600 && rctx->family < CHIP_RS780 ? 2 : 0); /* SURFACE_BASE_UPDATE */
+
+ rctx->set_atom_dirty(rctx, begin, true);
+
+ r600_set_streamout_enable(rctx, true);
}
void r600_set_streamout_targets(struct pipe_context *ctx,
unsigned num_targets,
struct pipe_stream_output_target **targets,
- unsigned append_bitmask)
+ const unsigned *offsets)
{
struct r600_common_context *rctx = (struct r600_common_context *)ctx;
unsigned i;
+ unsigned append_bitmask = 0;
/* Stop streamout. */
if (rctx->streamout.num_targets && rctx->streamout.begin_emitted) {
for (i = 0; i < num_targets; i++) {
pipe_so_target_reference((struct pipe_stream_output_target**)&rctx->streamout.targets[i], targets[i]);
r600_context_add_resource_size(ctx, targets[i]->buffer);
+ if (offsets[i] == ((unsigned)-1))
+ append_bitmask |= 1 << i;
}
for (; i < rctx->streamout.num_targets; i++) {
pipe_so_target_reference((struct pipe_stream_output_target**)&rctx->streamout.targets[i], NULL);
if (num_targets) {
r600_streamout_buffers_dirty(rctx);
+ } else {
+ rctx->set_atom_dirty(rctx, &rctx->streamout.begin_atom, false);
+ r600_set_streamout_enable(rctx, false);
}
}
static void r600_flush_vgt_streamout(struct r600_common_context *rctx)
{
- struct radeon_winsys_cs *cs = rctx->rings.gfx.cs;
+ struct radeon_winsys_cs *cs = rctx->gfx.cs;
+ unsigned reg_strmout_cntl;
+
+ /* The register is at different places on different ASICs. */
+ if (rctx->chip_class >= CIK) {
+ reg_strmout_cntl = R_0300FC_CP_STRMOUT_CNTL;
+ } else if (rctx->chip_class >= EVERGREEN) {
+ reg_strmout_cntl = R_0084FC_CP_STRMOUT_CNTL;
+ } else {
+ reg_strmout_cntl = R_008490_CP_STRMOUT_CNTL;
+ }
- r600_write_config_reg(cs, R_008490_CP_STRMOUT_CNTL, 0);
+ if (rctx->chip_class >= CIK) {
+ radeon_set_uconfig_reg(cs, reg_strmout_cntl, 0);
+ } else {
+ radeon_set_config_reg(cs, reg_strmout_cntl, 0);
+ }
radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, 0));
radeon_emit(cs, EVENT_TYPE(EVENT_TYPE_SO_VGTSTREAMOUT_FLUSH) | EVENT_INDEX(0));
radeon_emit(cs, PKT3(PKT3_WAIT_REG_MEM, 5, 0));
radeon_emit(cs, WAIT_REG_MEM_EQUAL); /* wait until the register is equal to the reference value */
- radeon_emit(cs, R_008490_CP_STRMOUT_CNTL >> 2); /* register */
+ radeon_emit(cs, reg_strmout_cntl >> 2); /* register */
radeon_emit(cs, 0);
radeon_emit(cs, S_008490_OFFSET_UPDATE_DONE(1)); /* reference value */
radeon_emit(cs, S_008490_OFFSET_UPDATE_DONE(1)); /* mask */
radeon_emit(cs, 4); /* poll interval */
}
-static void evergreen_flush_vgt_streamout(struct r600_common_context *rctx)
-{
- struct radeon_winsys_cs *cs = rctx->rings.gfx.cs;
-
- r600_write_config_reg(cs, R_0084FC_CP_STRMOUT_CNTL, 0);
-
- radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, 0));
- radeon_emit(cs, EVENT_TYPE(EVENT_TYPE_SO_VGTSTREAMOUT_FLUSH) | EVENT_INDEX(0));
-
- radeon_emit(cs, PKT3(PKT3_WAIT_REG_MEM, 5, 0));
- radeon_emit(cs, WAIT_REG_MEM_EQUAL); /* wait until the register is equal to the reference value */
- radeon_emit(cs, R_0084FC_CP_STRMOUT_CNTL >> 2); /* register */
- radeon_emit(cs, 0);
- radeon_emit(cs, S_0084FC_OFFSET_UPDATE_DONE(1)); /* reference value */
- radeon_emit(cs, S_0084FC_OFFSET_UPDATE_DONE(1)); /* mask */
- radeon_emit(cs, 4); /* poll interval */
-}
-
-static void r600_set_streamout_enable(struct r600_common_context *rctx, unsigned buffer_enable_bit)
-{
- struct radeon_winsys_cs *cs = rctx->rings.gfx.cs;
-
- if (buffer_enable_bit) {
- r600_write_context_reg(cs, R_028AB0_VGT_STRMOUT_EN, S_028AB0_STREAMOUT(1));
- r600_write_context_reg(cs, R_028B20_VGT_STRMOUT_BUFFER_EN, buffer_enable_bit);
- } else {
- r600_write_context_reg(cs, R_028AB0_VGT_STRMOUT_EN, S_028AB0_STREAMOUT(0));
- }
-}
-
-static void evergreen_set_streamout_enable(struct r600_common_context *rctx, unsigned buffer_enable_bit)
-{
- struct radeon_winsys_cs *cs = rctx->rings.gfx.cs;
-
- if (buffer_enable_bit) {
- r600_write_context_reg_seq(cs, R_028B94_VGT_STRMOUT_CONFIG, 2);
- radeon_emit(cs, S_028B94_STREAMOUT_0_EN(1)); /* R_028B94_VGT_STRMOUT_CONFIG */
- radeon_emit(cs, S_028B98_STREAM_0_BUFFER_EN(buffer_enable_bit)); /* R_028B98_VGT_STRMOUT_BUFFER_CONFIG */
- } else {
- r600_write_context_reg(cs, R_028B94_VGT_STRMOUT_CONFIG, S_028B94_STREAMOUT_0_EN(0));
- }
-}
-
-static void r600_emit_reloc(struct r600_common_context *rctx,
- struct r600_ring *ring, struct r600_resource *rbo,
- enum radeon_bo_usage usage)
-{
- struct radeon_winsys_cs *cs = ring->cs;
- bool has_vm = ((struct r600_common_screen*)rctx->b.screen)->info.r600_virtual_address;
- unsigned reloc = r600_context_bo_reloc(rctx, ring, rbo, usage);
-
- if (!has_vm) {
- radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
- radeon_emit(cs, reloc);
- }
-}
-
static void r600_emit_streamout_begin(struct r600_common_context *rctx, struct r600_atom *atom)
{
- struct radeon_winsys_cs *cs = rctx->rings.gfx.cs;
+ struct radeon_winsys_cs *cs = rctx->gfx.cs;
struct r600_so_target **t = rctx->streamout.targets;
unsigned *stride_in_dw = rctx->streamout.stride_in_dw;
unsigned i, update_flags = 0;
- uint64_t va;
- if (rctx->chip_class >= EVERGREEN) {
- evergreen_flush_vgt_streamout(rctx);
- evergreen_set_streamout_enable(rctx, rctx->streamout.enabled_mask);
- } else {
- r600_flush_vgt_streamout(rctx);
- r600_set_streamout_enable(rctx, rctx->streamout.enabled_mask);
- }
+ r600_flush_vgt_streamout(rctx);
for (i = 0; i < rctx->streamout.num_targets; i++) {
if (!t[i])
t[i]->stride_in_dw = stride_in_dw[i];
- va = r600_resource_va(rctx->b.screen,
- (void*)t[i]->b.buffer);
+ if (rctx->chip_class >= SI) {
+ /* SI binds streamout buffers as shader resources.
+ * VGT only counts primitives and tells the shader
+ * through SGPRs what to do. */
+ radeon_set_context_reg_seq(cs, R_028AD0_VGT_STRMOUT_BUFFER_SIZE_0 + 16*i, 2);
+ radeon_emit(cs, (t[i]->b.buffer_offset +
+ t[i]->b.buffer_size) >> 2); /* BUFFER_SIZE (in DW) */
+ radeon_emit(cs, stride_in_dw[i]); /* VTX_STRIDE (in DW) */
+ } else {
+ uint64_t va = r600_resource(t[i]->b.buffer)->gpu_address;
- update_flags |= SURFACE_BASE_UPDATE_STRMOUT(i);
+ update_flags |= SURFACE_BASE_UPDATE_STRMOUT(i);
- r600_write_context_reg_seq(cs, R_028AD0_VGT_STRMOUT_BUFFER_SIZE_0 + 16*i, 3);
- radeon_emit(cs, (t[i]->b.buffer_offset +
- t[i]->b.buffer_size) >> 2); /* BUFFER_SIZE (in DW) */
- radeon_emit(cs, stride_in_dw[i]); /* VTX_STRIDE (in DW) */
- radeon_emit(cs, va >> 8); /* BUFFER_BASE */
+ radeon_set_context_reg_seq(cs, R_028AD0_VGT_STRMOUT_BUFFER_SIZE_0 + 16*i, 3);
+ radeon_emit(cs, (t[i]->b.buffer_offset +
+ t[i]->b.buffer_size) >> 2); /* BUFFER_SIZE (in DW) */
+ radeon_emit(cs, stride_in_dw[i]); /* VTX_STRIDE (in DW) */
+ radeon_emit(cs, va >> 8); /* BUFFER_BASE */
- r600_emit_reloc(rctx, &rctx->rings.gfx, r600_resource(t[i]->b.buffer),
- RADEON_USAGE_WRITE);
+ r600_emit_reloc(rctx, &rctx->gfx, r600_resource(t[i]->b.buffer),
+ RADEON_USAGE_WRITE, RADEON_PRIO_RINGS_STREAMOUT);
- /* R7xx requires this packet after updating BUFFER_BASE.
- * Without this, R7xx locks up. */
- if (rctx->family >= CHIP_RS780 && rctx->family <= CHIP_RV740) {
- radeon_emit(cs, PKT3(PKT3_STRMOUT_BASE_UPDATE, 1, 0));
- radeon_emit(cs, i);
- radeon_emit(cs, va >> 8);
+ /* R7xx requires this packet after updating BUFFER_BASE.
+ * Without this, R7xx locks up. */
+ if (rctx->family >= CHIP_RS780 && rctx->family <= CHIP_RV740) {
+ radeon_emit(cs, PKT3(PKT3_STRMOUT_BASE_UPDATE, 1, 0));
+ radeon_emit(cs, i);
+ radeon_emit(cs, va >> 8);
- r600_emit_reloc(rctx, &rctx->rings.gfx, r600_resource(t[i]->b.buffer),
- RADEON_USAGE_WRITE);
+ r600_emit_reloc(rctx, &rctx->gfx, r600_resource(t[i]->b.buffer),
+ RADEON_USAGE_WRITE, RADEON_PRIO_RINGS_STREAMOUT);
+ }
}
- if (rctx->streamout.append_bitmask & (1 << i)) {
- va = r600_resource_va(rctx->b.screen,
- (void*)t[i]->buf_filled_size) + t[i]->buf_filled_size_offset;
+ if (rctx->streamout.append_bitmask & (1 << i) && t[i]->buf_filled_size_valid) {
+ uint64_t va = t[i]->buf_filled_size->gpu_address +
+ t[i]->buf_filled_size_offset;
+
/* Append. */
radeon_emit(cs, PKT3(PKT3_STRMOUT_BUFFER_UPDATE, 4, 0));
radeon_emit(cs, STRMOUT_SELECT_BUFFER(i) |
radeon_emit(cs, va); /* src address lo */
radeon_emit(cs, va >> 32); /* src address hi */
- r600_emit_reloc(rctx, &rctx->rings.gfx, t[i]->buf_filled_size,
- RADEON_USAGE_READ);
+ r600_emit_reloc(rctx, &rctx->gfx, t[i]->buf_filled_size,
+ RADEON_USAGE_READ, RADEON_PRIO_SO_FILLED_SIZE);
} else {
/* Start from the beginning. */
radeon_emit(cs, PKT3(PKT3_STRMOUT_BUFFER_UPDATE, 4, 0));
void r600_emit_streamout_end(struct r600_common_context *rctx)
{
- struct radeon_winsys_cs *cs = rctx->rings.gfx.cs;
+ struct radeon_winsys_cs *cs = rctx->gfx.cs;
struct r600_so_target **t = rctx->streamout.targets;
unsigned i;
uint64_t va;
- if (rctx->chip_class >= EVERGREEN) {
- evergreen_flush_vgt_streamout(rctx);
- } else {
- r600_flush_vgt_streamout(rctx);
- }
+ r600_flush_vgt_streamout(rctx);
for (i = 0; i < rctx->streamout.num_targets; i++) {
if (!t[i])
continue;
- va = r600_resource_va(rctx->b.screen,
- (void*)t[i]->buf_filled_size) + t[i]->buf_filled_size_offset;
+ va = t[i]->buf_filled_size->gpu_address + t[i]->buf_filled_size_offset;
radeon_emit(cs, PKT3(PKT3_STRMOUT_BUFFER_UPDATE, 4, 0));
radeon_emit(cs, STRMOUT_SELECT_BUFFER(i) |
STRMOUT_OFFSET_SOURCE(STRMOUT_OFFSET_NONE) |
radeon_emit(cs, 0); /* unused */
radeon_emit(cs, 0); /* unused */
- r600_emit_reloc(rctx, &rctx->rings.gfx, t[i]->buf_filled_size,
- RADEON_USAGE_WRITE);
+ r600_emit_reloc(rctx, &rctx->gfx, t[i]->buf_filled_size,
+ RADEON_USAGE_WRITE, RADEON_PRIO_SO_FILLED_SIZE);
+
+ /* Zero the buffer size. The counters (primitives generated,
+ * primitives emitted) may be enabled even if there is not
+ * buffer bound. This ensures that the primitives-emitted query
+ * won't increment. */
+ radeon_set_context_reg(cs, R_028AD0_VGT_STRMOUT_BUFFER_SIZE_0 + 16*i, 0);
+
+ t[i]->buf_filled_size_valid = true;
}
+ rctx->streamout.begin_emitted = false;
+ rctx->flags |= R600_CONTEXT_STREAMOUT_FLUSH;
+}
+
+/* STREAMOUT CONFIG DERIVED STATE
+ *
+ * Streamout must be enabled for the PRIMITIVES_GENERATED query to work.
+ * The buffer mask is an independent state, so no writes occur if there
+ * are no buffers bound.
+ */
+
+static bool r600_get_strmout_en(struct r600_common_context *rctx)
+{
+ return rctx->streamout.streamout_enabled ||
+ rctx->streamout.prims_gen_query_enabled;
+}
+
+static void r600_emit_streamout_enable(struct r600_common_context *rctx,
+ struct r600_atom *atom)
+{
+ unsigned strmout_config_reg = R_028AB0_VGT_STRMOUT_EN;
+ unsigned strmout_config_val = S_028B94_STREAMOUT_0_EN(r600_get_strmout_en(rctx));
+ unsigned strmout_buffer_reg = R_028B20_VGT_STRMOUT_BUFFER_EN;
+ unsigned strmout_buffer_val = rctx->streamout.hw_enabled_mask &
+ rctx->streamout.enabled_stream_buffers_mask;
+
if (rctx->chip_class >= EVERGREEN) {
- evergreen_set_streamout_enable(rctx, 0);
- } else {
- r600_set_streamout_enable(rctx, 0);
+ strmout_buffer_reg = R_028B98_VGT_STRMOUT_BUFFER_CONFIG;
+
+ strmout_config_reg = R_028B94_VGT_STRMOUT_CONFIG;
+ strmout_config_val |=
+ S_028B94_RAST_STREAM(0) |
+ S_028B94_STREAMOUT_1_EN(r600_get_strmout_en(rctx)) |
+ S_028B94_STREAMOUT_2_EN(r600_get_strmout_en(rctx)) |
+ S_028B94_STREAMOUT_3_EN(r600_get_strmout_en(rctx));
}
+ radeon_set_context_reg(rctx->gfx.cs, strmout_buffer_reg, strmout_buffer_val);
+ radeon_set_context_reg(rctx->gfx.cs, strmout_config_reg, strmout_config_val);
+}
- rctx->streamout.begin_emitted = false;
+static void r600_set_streamout_enable(struct r600_common_context *rctx, bool enable)
+{
+ bool old_strmout_en = r600_get_strmout_en(rctx);
+ unsigned old_hw_enabled_mask = rctx->streamout.hw_enabled_mask;
- if (rctx->chip_class >= R700) {
- rctx->flags |= R600_CONTEXT_STREAMOUT_FLUSH;
- } else {
- rctx->flags |= R600_CONTEXT_FLUSH_AND_INV;
+ rctx->streamout.streamout_enabled = enable;
+
+ rctx->streamout.hw_enabled_mask = rctx->streamout.enabled_mask |
+ (rctx->streamout.enabled_mask << 4) |
+ (rctx->streamout.enabled_mask << 8) |
+ (rctx->streamout.enabled_mask << 12);
+
+ if ((old_strmout_en != r600_get_strmout_en(rctx)) ||
+ (old_hw_enabled_mask != rctx->streamout.hw_enabled_mask)) {
+ rctx->set_atom_dirty(rctx, &rctx->streamout.enable_atom, true);
+ }
+}
+
+void r600_update_prims_generated_query_state(struct r600_common_context *rctx,
+ unsigned type, int diff)
+{
+ if (type == PIPE_QUERY_PRIMITIVES_GENERATED) {
+ bool old_strmout_en = r600_get_strmout_en(rctx);
+
+ rctx->streamout.num_prims_gen_queries += diff;
+ assert(rctx->streamout.num_prims_gen_queries >= 0);
+
+ rctx->streamout.prims_gen_query_enabled =
+ rctx->streamout.num_prims_gen_queries != 0;
+
+ if (old_strmout_en != r600_get_strmout_en(rctx)) {
+ rctx->set_atom_dirty(rctx, &rctx->streamout.enable_atom, true);
+ }
}
}
rctx->b.create_stream_output_target = r600_create_so_target;
rctx->b.stream_output_target_destroy = r600_so_target_destroy;
rctx->streamout.begin_atom.emit = r600_emit_streamout_begin;
+ rctx->streamout.enable_atom.emit = r600_emit_streamout_enable;
+ rctx->streamout.enable_atom.num_dw = 6;
}