fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-05-02 20:58:04 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

turnip: use global bo for clear blit shaders

Browse source 
Fill the global bo will all possible shaders for 3D clear/blit. Note the
global bo size is still <4k (so this doesn't cost any extra memory), this
saves having to allocate shaders in sub_cs everytime the 3D path is used.

Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5776>
This commit is contained in:
Jonathan Marek 2020-07-06 11:48:54 -04:00 • committed by Marge Bot
parent 10020b8d17
commit 14c554a391
3 changed files with 103 additions and 94 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

181
src/freedreno/vulkan/tu_clear_blit.c
View file

@ -318,6 +318,87 @@ r2d_run(struct tu_cmd_buffer *cmd, struct tu_cs *cs)
/* r3d_ = shader path operations */
void
tu_init_clear_blit_shaders(struct tu6_global *global)
{
#define MOV(args...) { .cat1 = { .opc_cat = 1, .src_type = TYPE_S32, .dst_type = TYPE_S32, args } }
#define CAT2(op, args...) { .cat2 = { .opc_cat = 2, .opc = (op) & 63, .full = 1, args } }
#define CAT3(op, args...) { .cat3 = { .opc_cat = 3, .opc = (op) & 63, args } }
static const instr_t vs_code[] = {
/* r0.xyz = r0.w ? c1.xyz : c0.xyz
* r1.xy = r0.w ? c1.zw : c0.zw
* r0.w = 1.0f
*/
CAT3(OPC_SEL_B32, .repeat = 2, .dst = 0,
.c1 = {.src1_c = 1, .src1 = 4}, .src1_r = 1,
.src2 = 3,
.c2 = {.src3_c = 1, .dummy = 1, .src3 = 0}),
CAT3(OPC_SEL_B32, .repeat = 1, .dst = 4,
.c1 = {.src1_c = 1, .src1 = 6}, .src1_r = 1,
.src2 = 3,
.c2 = {.src3_c = 1, .dummy = 1, .src3 = 2}),
MOV(.dst = 3, .src_im = 1, .fim_val = 1.0f ),
{ .cat0 = { .opc = OPC_END } },
};
static const instr_t vs_layered[] = {
{ .cat0 = { .opc = OPC_CHMASK } },
{ .cat0 = { .opc = OPC_CHSH } },
};
static const instr_t gs_code[] = {
/* (sy)(ss)(nop3)shr.b r0.w, r0.x, 16 (extract local_id) */
CAT2(OPC_SHR_B, .dst = 3, .src1 = 0, .src2_im = 1, .src2 = 16,
.src1_r = 1, .src2_r = 1, .ss = 1, .sync = 1),
/* x = (local_id & 1) ? c1.x : c0.x */
CAT2(OPC_AND_B, .dst = 0, .src1 = 3, .src2_im = 1, .src2 = 1),
/* y = (local_id & 2) ? c1.y : c0.y */
CAT2(OPC_AND_B, .dst = 1, .src1 = 3, .src2_im = 1, .src2 = 2),
/* pred = (local_id >= 4), used by OPC_KILL */
CAT2(OPC_CMPS_S, .dst = REG_P0 * 4, .cond = IR3_COND_GE, .src1 = 3, .src2_im = 1, .src2 = 4),
/* vertex_flags_out = (local_id == 0) ? 4 : 0 - first vertex flag */
CAT2(OPC_CMPS_S, .dst = 4, .cond = IR3_COND_EQ, .src1 = 3, .src2_im = 1, .src2 = 0),
MOV(.dst = 2, .src_c = 1, .src = 2), /* depth clear value from c0.z */
MOV(.dst = 3, .src_im = 1, .fim_val = 1.0f),
MOV(.dst = 5, .src_c = 1, .src = 3), /* layer id from c0.w */
/* (rpt1)sel.b32 r0.x, (r)c1.x, (r)r0.x, (r)c0.x */
CAT3(OPC_SEL_B32, .repeat = 1, .dst = 0,
.c1 = {.src1_c = 1, .src1 = 4, .dummy = 4}, .src1_r = 1,
.src2 = 0,
.c2 = {.src3_c = 1, .dummy = 1, .src3 = 0}),
CAT2(OPC_SHL_B, .dst = 4, .src1 = 4, .src2_im = 1, .src2 = 2),
{ .cat0 = { .opc = OPC_KILL } },
{ .cat0 = { .opc = OPC_END, .ss = 1, .sync = 1 } },
};
static const instr_t fs_blit[] = {
/* " bary.f (ei)r63.x, 0, r0.x" note the blob doesn't have this in its
* blit path (its not clear what allows it to not have it)
*/
CAT2(OPC_BARY_F, .ei = 1, .full = 1, .dst = 63 * 4, .src1_im = 1),
{ .cat0 = { .opc = OPC_END } },
};
memcpy(&global->shaders[GLOBAL_SH_VS], vs_code, sizeof(vs_code));
memcpy(&global->shaders[GLOBAL_SH_VS_LAYER], vs_layered, sizeof(vs_layered));
memcpy(&global->shaders[GLOBAL_SH_GS_LAYER], gs_code, sizeof(gs_code));
memcpy(&global->shaders[GLOBAL_SH_FS_BLIT], fs_blit, sizeof(fs_blit));
for (uint32_t num_rts = 0; num_rts <= MAX_RTS; num_rts++) {
instr_t *code = global->shaders[GLOBAL_SH_FS_CLEAR0 + num_rts];
for (uint32_t i = 0; i < num_rts; i++) {
/* (rpt3)mov.s32s32 r0.x, (r)c[i].x */
*code++ = (instr_t) MOV(.repeat = 3, .dst = i * 4, .src_c = 1, .src_r = 1, .src = i * 4);
}
*code++ = (instr_t) { .cat0 = { .opc = OPC_END } };
}
}
static void
r3d_common(struct tu_cmd_buffer *cmd, struct tu_cs *cs, bool blit, uint32_t num_rts,
bool layered_clear)
@ -415,105 +496,17 @@ r3d_common(struct tu_cmd_buffer *cmd, struct tu_cs *cs, bool blit, uint32_t num_
.const_state = &dummy_const_state,
}, *gs = layered_clear ? &gs_shader : NULL;
#define MOV(args...) { .cat1 = { .opc_cat = 1, .src_type = TYPE_F32, .dst_type = TYPE_F32, args } }
#define CAT2(op, args...) { .cat2 = { .opc_cat = 2, .opc = (op) & 63, .full = 1, args } }
#define CAT3(op, args...) { .cat3 = { .opc_cat = 3, .opc = (op) & 63, args } }
static const instr_t vs_code[] = {
/* r0.xyz = r0.w ? c1.xyz : c0.xyz
* r1.xy = r0.w ? c1.zw : c0.zw
* r0.w = 1.0f
*/
CAT3(OPC_SEL_B32, .repeat = 2, .dst = 0,
.c1 = {.src1_c = 1, .src1 = 4}, .src1_r = 1,
.src2 = 3,
.c2 = {.src3_c = 1, .dummy = 1, .src3 = 0}),
CAT3(OPC_SEL_B32, .repeat = 1, .dst = 4,
.c1 = {.src1_c = 1, .src1 = 6}, .src1_r = 1,
.src2 = 3,
.c2 = {.src3_c = 1, .dummy = 1, .src3 = 2}),
MOV(.dst = 3, .src_im = 1, .fim_val = 1.0f ),
{ .cat0 = { .opc = OPC_END } },
};
static const instr_t vs_layered[] = {
{ .cat0 = { .opc = OPC_CHMASK } },
{ .cat0 = { .opc = OPC_CHSH } },
};
static const instr_t gs_code[16] = {
/* (sy)(ss)(nop3)shr.b r0.w, r0.x, 16 (extract local_id) */
CAT2(OPC_SHR_B, .dst = 3, .src1 = 0, .src2_im = 1, .src2 = 16,
.src1_r = 1, .src2_r = 1, .ss = 1, .sync = 1),
/* x = (local_id & 1) ? c1.x : c0.x */
CAT2(OPC_AND_B, .dst = 0, .src1 = 3, .src2_im = 1, .src2 = 1),
/* y = (local_id & 2) ? c1.y : c0.y */
CAT2(OPC_AND_B, .dst = 1, .src1 = 3, .src2_im = 1, .src2 = 2),
/* pred = (local_id >= 4), used by OPC_KILL */
CAT2(OPC_CMPS_S, .dst = REG_P0 * 4, .cond = IR3_COND_GE, .src1 = 3, .src2_im = 1, .src2 = 4),
/* vertex_flags_out = (local_id == 0) ? 4 : 0 - first vertex flag */
CAT2(OPC_CMPS_S, .dst = 4, .cond = IR3_COND_EQ, .src1 = 3, .src2_im = 1, .src2 = 0),
MOV(.dst = 2, .src_c = 1, .src = 2), /* depth clear value from c0.z */
MOV(.dst = 3, .src_im = 1, .fim_val = 1.0f),
MOV(.dst = 5, .src_c = 1, .src = 3), /* layer id from c0.w */
/* (rpt1)sel.b32 r0.x, (r)c1.x, (r)r0.x, (r)c0.x */
CAT3(OPC_SEL_B32, .repeat = 1, .dst = 0,
.c1 = {.src1_c = 1, .src1 = 4, .dummy = 4}, .src1_r = 1,
.src2 = 0,
.c2 = {.src3_c = 1, .dummy = 1, .src3 = 0}),
CAT2(OPC_SHL_B, .dst = 4, .src1 = 4, .src2_im = 1, .src2 = 2),
{ .cat0 = { .opc = OPC_KILL } },
{ .cat0 = { .opc = OPC_END, .ss = 1, .sync = 1 } },
};
#define FS_OFFSET (16 * sizeof(instr_t))
#define GS_OFFSET (32 * sizeof(instr_t))
/* shaders */
struct tu_cs_memory shaders = { };
VkResult result = tu_cs_alloc(&cmd->sub_cs, 2 + layered_clear,
16 * sizeof(instr_t), &shaders);
assert(result == VK_SUCCESS);
if (layered_clear) {
memcpy(shaders.map, vs_layered, sizeof(vs_layered));
memcpy((uint8_t*) shaders.map + GS_OFFSET, gs_code, sizeof(gs_code));
} else {
memcpy(shaders.map, vs_code, sizeof(vs_code));
}
instr_t *fs_code = (instr_t*) ((uint8_t*) shaders.map + FS_OFFSET);
for (uint32_t i = 0; i < num_rts; i++) {
/* (rpt3)mov.s32s32 r0.x, (r)c[i].x */
*fs_code++ = (instr_t) { .cat1 = {
.opc_cat = 1, .src_type = TYPE_S32, .dst_type = TYPE_S32,
.repeat = 3, .dst = i * 4, .src_c = 1, .src_r = 1, .src = i * 4
} };
}
/* " bary.f (ei)r63.x, 0, r0.x" note the blob doesn't have this in its
* blit path (its not clear what allows it to not have it)
*/
if (blit) {
*fs_code++ = (instr_t) { .cat2 = {
.opc_cat = 2, .opc = OPC_BARY_F & 63, .ei = 1, .full = 1,
.dst = regid(63, 0), .src1_im = 1
} };
}
*fs_code++ = (instr_t) { .cat0 = { .opc = OPC_END } };
/* note: assumed <= 16 instructions (MAX_RTS is 8) */
tu_cs_emit_regs(cs, A6XX_HLSQ_UPDATE_CNTL(0x7ffff));
tu6_emit_xs_config(cs, MESA_SHADER_VERTEX, &vs, shaders.iova);
tu6_emit_xs_config(cs, MESA_SHADER_VERTEX, &vs,
global_iova(cmd, shaders[gs ? GLOBAL_SH_VS_LAYER : GLOBAL_SH_VS]));
tu6_emit_xs_config(cs, MESA_SHADER_TESS_CTRL, NULL, 0);
tu6_emit_xs_config(cs, MESA_SHADER_TESS_EVAL, NULL, 0);
tu6_emit_xs_config(cs, MESA_SHADER_GEOMETRY, gs, shaders.iova + GS_OFFSET);
tu6_emit_xs_config(cs, MESA_SHADER_FRAGMENT, &fs, shaders.iova + FS_OFFSET);
tu6_emit_xs_config(cs, MESA_SHADER_GEOMETRY, gs,
global_iova(cmd, shaders[GLOBAL_SH_GS_LAYER]));
tu6_emit_xs_config(cs, MESA_SHADER_FRAGMENT, &fs,
global_iova(cmd, shaders[blit ? GLOBAL_SH_FS_BLIT : (GLOBAL_SH_FS_CLEAR0 + num_rts)]));
tu_cs_emit_regs(cs, A6XX_PC_PRIMITIVE_CNTL_0());
tu_cs_emit_regs(cs, A6XX_VFD_CONTROL_0());

1
src/freedreno/vulkan/tu_device.c
View file

@ -1278,6 +1278,7 @@ tu_CreateDevice(VkPhysicalDevice physicalDevice,
goto fail_global_bo_map;
memcpy(device->global_bo.map + gb_offset(border_color), border_color, sizeof(border_color));
tu_init_clear_blit_shaders(device->global_bo.map);
VkPipelineCacheCreateInfo ci;
ci.sType = VK_STRUCTURE_TYPE_PIPELINE_CACHE_CREATE_INFO;

15
src/freedreno/vulkan/tu_private.h
View file

@ -339,12 +339,25 @@ struct tu_bo
void *map;
};
enum global_shader {
GLOBAL_SH_VS,
GLOBAL_SH_VS_LAYER,
GLOBAL_SH_GS_LAYER,
GLOBAL_SH_FS_BLIT,
GLOBAL_SH_FS_CLEAR0,
GLOBAL_SH_FS_CLEAR_MAX = GLOBAL_SH_FS_CLEAR0 + MAX_RTS,
GLOBAL_SH_COUNT,
};
/* This struct defines the layout of the global_bo */
struct tu6_global
{
/* 6 bcolor_entry entries, one for each VK_BORDER_COLOR */
uint8_t border_color[128 * 6];
/* clear/blit shaders, all <= 16 instrs (16 instr = 1 instrlen unit) */
instr_t shaders[GLOBAL_SH_COUNT][16];
uint32_t seqno_dummy; /* dummy seqno for CP_EVENT_WRITE */
uint32_t _pad0;
volatile uint32_t vsc_draw_overflow;
@ -361,6 +374,8 @@ struct tu6_global
#define gb_offset(member) offsetof(struct tu6_global, member)
#define global_iova(cmd, member) ((cmd)->device->global_bo.iova + gb_offset(member))
void tu_init_clear_blit_shaders(struct tu6_global *global);
/* extra space in vsc draw/prim streams */
#define VSC_PAD 0x40