freedreno/a6xx: convert const emit to OUT_PKT()

This is another hot packet.  This splits out each of the four cases
(geom vs frag, and indirect vs inline) intentionally, to avoid some
parity bit calc.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4813>
This commit is contained in:
Rob Clark 2020-04-29 10:09:28 -07:00 committed by Marge Bot
parent 710537b19c
commit 373e9ab27c
2 changed files with 71 additions and 36 deletions

View file

@ -23,6 +23,7 @@
*/
#include "fd6_const.h"
#include "fd6_pack.h"
#include "ir3_const.h"
@ -35,43 +36,71 @@ fd6_emit_const(struct fd_ringbuffer *ring, gl_shader_stage type,
uint32_t regid, uint32_t offset, uint32_t sizedwords,
const uint32_t *dwords, struct pipe_resource *prsc)
{
uint32_t i, sz, align_sz;
enum a6xx_state_src src;
debug_assert((regid % 4) == 0);
if (prsc) {
sz = 0;
src = SS6_INDIRECT;
} else {
sz = sizedwords;
src = SS6_DIRECT;
}
align_sz = align(sz, 4);
OUT_PKT7(ring, fd6_stage2opcode(type), 3 + align_sz);
OUT_RING(ring, CP_LOAD_STATE6_0_DST_OFF(regid/4) |
CP_LOAD_STATE6_0_STATE_TYPE(ST6_CONSTANTS) |
CP_LOAD_STATE6_0_STATE_SRC(src) |
CP_LOAD_STATE6_0_STATE_BLOCK(fd6_stage2shadersb(type)) |
CP_LOAD_STATE6_0_NUM_UNIT(DIV_ROUND_UP(sizedwords, 4)));
if (prsc) {
struct fd_bo *bo = fd_resource(prsc)->bo;
OUT_RELOC(ring, bo, offset, 0, 0);
if (fd6_geom_stage(type)) {
OUT_PKT(ring, CP_LOAD_STATE6_GEOM,
CP_LOAD_STATE6_0(
.dst_off = regid/4,
.state_type = ST6_CONSTANTS,
.state_src = SS6_INDIRECT,
.state_block = fd6_stage2shadersb(type),
.num_unit = DIV_ROUND_UP(sizedwords, 4)
),
CP_LOAD_STATE6_EXT_SRC_ADDR(
.bo = bo,
.bo_offset = offset
)
);
} else {
OUT_PKT(ring, CP_LOAD_STATE6_FRAG,
CP_LOAD_STATE6_0(
.dst_off = regid/4,
.state_type = ST6_CONSTANTS,
.state_src = SS6_INDIRECT,
.state_block = fd6_stage2shadersb(type),
.num_unit = DIV_ROUND_UP(sizedwords, 4)
),
CP_LOAD_STATE6_EXT_SRC_ADDR(
.bo = bo,
.bo_offset = offset
)
);
}
} else {
OUT_RING(ring, CP_LOAD_STATE6_1_EXT_SRC_ADDR(0));
OUT_RING(ring, CP_LOAD_STATE6_2_EXT_SRC_ADDR_HI(0));
/* NOTE we cheat a bit here, since we know mesa is aligning
* the size of the user buffer to 16 bytes. And we want to
* cut cycles in a hot path.
*/
uint32_t align_sz = align(sizedwords, 4);
dwords = (uint32_t *)&((uint8_t *)dwords)[offset];
}
for (i = 0; i < sz; i++) {
OUT_RING(ring, dwords[i]);
}
/* Zero-pad to multiple of 4 dwords */
for (i = sz; i < align_sz; i++) {
OUT_RING(ring, 0);
if (fd6_geom_stage(type)) {
OUT_PKTBUF(ring, CP_LOAD_STATE6_GEOM, dwords, align_sz,
CP_LOAD_STATE6_0(
.dst_off = regid/4,
.state_type = ST6_CONSTANTS,
.state_src = SS6_DIRECT,
.state_block = fd6_stage2shadersb(type),
.num_unit = DIV_ROUND_UP(sizedwords, 4)
),
CP_LOAD_STATE6_1(),
CP_LOAD_STATE6_2()
);
} else {
OUT_PKTBUF(ring, CP_LOAD_STATE6_FRAG, dwords, align_sz,
CP_LOAD_STATE6_0(
.dst_off = regid/4,
.state_type = ST6_CONSTANTS,
.state_src = SS6_DIRECT,
.state_block = fd6_stage2shadersb(type),
.num_unit = DIV_ROUND_UP(sizedwords, 4)
),
CP_LOAD_STATE6_1(),
CP_LOAD_STATE6_2()
);
}
}
}

View file

@ -210,24 +210,30 @@ fd6_emit_lrz_flush(struct fd_ringbuffer *ring)
OUT_RING(ring, LRZ_FLUSH);
}
static inline uint32_t
fd6_stage2opcode(gl_shader_stage type)
static inline bool
fd6_geom_stage(gl_shader_stage type)
{
switch (type) {
case MESA_SHADER_VERTEX:
case MESA_SHADER_TESS_CTRL:
case MESA_SHADER_TESS_EVAL:
case MESA_SHADER_GEOMETRY:
return CP_LOAD_STATE6_GEOM;
return true;
case MESA_SHADER_FRAGMENT:
case MESA_SHADER_COMPUTE:
case MESA_SHADER_KERNEL:
return CP_LOAD_STATE6_FRAG;
return false;
default:
unreachable("bad shader type");
}
}
static inline uint32_t
fd6_stage2opcode(gl_shader_stage type)
{
return fd6_geom_stage(type) ? CP_LOAD_STATE6_GEOM : CP_LOAD_STATE6_FRAG;
}
static inline enum a6xx_state_block
fd6_stage2shadersb(gl_shader_stage type)
{