/**
* Emit a PIPE_CONTROL with various flushing flags.
*
* The caller is responsible for deciding what flags are appropriate for the
* given generation.
*/
void
brw_emit_pipe_control_flush(struct brw_context *brw, uint32_t flags)
{
if (brw->gen >= 8) {
if (brw->gen == 8)
gen8_add_cs_stall_workaround_bits(&flags);
BEGIN_BATCH(6);
OUT_BATCH(_3DSTATE_PIPE_CONTROL | (6 - 2));
OUT_BATCH(flags);
OUT_BATCH(0);
OUT_BATCH(0);
OUT_BATCH(0);
OUT_BATCH(0);
ADVANCE_BATCH();
} else if (brw->gen >= 6) {
flags |= gen7_cs_stall_every_four_pipe_controls(brw, flags);
BEGIN_BATCH(5);
OUT_BATCH(_3DSTATE_PIPE_CONTROL | (5 - 2));
OUT_BATCH(flags);
OUT_BATCH(0);
OUT_BATCH(0);
OUT_BATCH(0);
ADVANCE_BATCH();
} else {
BEGIN_BATCH(4);
OUT_BATCH(_3DSTATE_PIPE_CONTROL | flags | (4 - 2));
OUT_BATCH(0);
OUT_BATCH(0);
OUT_BATCH(0);
ADVANCE_BATCH();
}
}
/**
* Emit a PIPE_CONTROL that writes to a buffer object.
*
* \p flags should contain one of the following items:
* - PIPE_CONTROL_WRITE_IMMEDIATE
* - PIPE_CONTROL_WRITE_TIMESTAMP
* - PIPE_CONTROL_WRITE_DEPTH_COUNT
*/
void
brw_emit_pipe_control_write(struct brw_context *brw, uint32_t flags,
drm_intel_bo *bo, uint32_t offset,
uint32_t imm_lower, uint32_t imm_upper)
{
if (brw->gen >= 8) {
if (brw->gen == 8)
gen8_add_cs_stall_workaround_bits(&flags);
BEGIN_BATCH(6);
OUT_BATCH(_3DSTATE_PIPE_CONTROL | (6 - 2));
OUT_BATCH(flags);
OUT_RELOC64(bo, I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
offset);
OUT_BATCH(imm_lower);
OUT_BATCH(imm_upper);
ADVANCE_BATCH();
} else if (brw->gen >= 6) {
flags |= gen7_cs_stall_every_four_pipe_controls(brw, flags);
/* PPGTT/GGTT is selected by DW2 bit 2 on Sandybridge, but DW1 bit 24
* on later platforms. We always use PPGTT on Gen7+.
*/
unsigned gen6_gtt = brw->gen == 6 ? PIPE_CONTROL_GLOBAL_GTT_WRITE : 0;
BEGIN_BATCH(5);
OUT_BATCH(_3DSTATE_PIPE_CONTROL | (5 - 2));
OUT_BATCH(flags);
OUT_RELOC(bo, I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
gen6_gtt | offset);
OUT_BATCH(imm_lower);
OUT_BATCH(imm_upper);
ADVANCE_BATCH();
} else {
BEGIN_BATCH(4);
OUT_BATCH(_3DSTATE_PIPE_CONTROL | flags | (4 - 2));
OUT_RELOC(bo, I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
PIPE_CONTROL_GLOBAL_GTT_WRITE | offset);
OUT_BATCH(imm_lower);
OUT_BATCH(imm_upper);
ADVANCE_BATCH();
}
}
/**
* Emit a PIPE_CONTROL with various flushing flags.
*
* The caller is responsible for deciding what flags are appropriate for the
* given generation.
*/
void
brw_emit_pipe_control_flush(struct brw_context *brw, uint32_t flags)
{
if (brw->gen >= 6 &&
(flags & PIPE_CONTROL_CACHE_FLUSH_BITS) &&
(flags & PIPE_CONTROL_CACHE_INVALIDATE_BITS)) {
/* A pipe control command with flush and invalidate bits set
* simultaneously is an inherently racy operation on Gen6+ if the
* contents of the flushed caches were intended to become visible from
* any of the invalidated caches. Split it in two PIPE_CONTROLs, the
* first one should stall the pipeline to make sure that the flushed R/W
* caches are coherent with memory once the specified R/O caches are
* invalidated. On pre-Gen6 hardware the (implicit) R/O cache
* invalidation seems to happen at the bottom of the pipeline together
* with any write cache flush, so this shouldn't be a concern.
*/
brw_emit_pipe_control_flush(brw, (flags & PIPE_CONTROL_CACHE_FLUSH_BITS) |
PIPE_CONTROL_CS_STALL);
flags &= ~(PIPE_CONTROL_CACHE_FLUSH_BITS | PIPE_CONTROL_CS_STALL);
}
if (brw->gen >= 8) {
if (brw->gen == 8)
gen8_add_cs_stall_workaround_bits(&flags);
if (brw->gen == 9 &&
(flags & PIPE_CONTROL_VF_CACHE_INVALIDATE)) {
/* Hardware workaround: SKL
*
* Emit Pipe Control with all bits set to zero before emitting
* a Pipe Control with VF Cache Invalidate set.
*/
brw_emit_pipe_control_flush(brw, 0);
}
BEGIN_BATCH(6);
OUT_BATCH(_3DSTATE_PIPE_CONTROL | (6 - 2));
OUT_BATCH(flags);
OUT_BATCH(0);
OUT_BATCH(0);
OUT_BATCH(0);
OUT_BATCH(0);
ADVANCE_BATCH();
} else if (brw->gen >= 6) {
if (brw->gen == 6 &&
(flags & PIPE_CONTROL_RENDER_TARGET_FLUSH)) {
/* Hardware workaround: SNB B-Spec says:
*
* [Dev-SNB{W/A}]: Before a PIPE_CONTROL with Write Cache Flush
* Enable = 1, a PIPE_CONTROL with any non-zero post-sync-op is
* required.
*/
brw_emit_post_sync_nonzero_flush(brw);
}
flags |= gen7_cs_stall_every_four_pipe_controls(brw, flags);
BEGIN_BATCH(5);
OUT_BATCH(_3DSTATE_PIPE_CONTROL | (5 - 2));
OUT_BATCH(flags);
OUT_BATCH(0);
OUT_BATCH(0);
OUT_BATCH(0);
ADVANCE_BATCH();
} else {
BEGIN_BATCH(4);
OUT_BATCH(_3DSTATE_PIPE_CONTROL | flags | (4 - 2));
OUT_BATCH(0);
OUT_BATCH(0);
OUT_BATCH(0);
ADVANCE_BATCH();
}
}
