#include "util/u_hash_table.h"
#include "util/u_hash.h"
#include "util/u_memory.h"
+#include "util/u_pack_color.h"
+#include "util/format_srgb.h"
#include "util/ralloc.h"
#include "tgsi/tgsi_dump.h"
#include "tgsi/tgsi_info.h"
+#include "tgsi/tgsi_lowering.h"
#include "vc4_context.h"
#include "vc4_qpu.h"
#endif
struct vc4_key {
- struct pipe_shader_state *shader_state;
+ struct vc4_uncompiled_shader *shader_state;
struct {
enum pipe_format format;
unsigned compare_mode:1;
unsigned compare_func:3;
+ unsigned wrap_s:3;
+ unsigned wrap_t:3;
uint8_t swizzle[4];
} tex[VC4_MAX_TEXTURE_SAMPLERS];
};
bool stencil_full_writemasks;
bool is_points;
bool is_lines;
+ bool alpha_test;
+ bool point_coord_upper_left;
+ bool light_twoside;
+ uint8_t alpha_test_func;
+ uint32_t point_sprite_mask;
struct pipe_rt_blend_state blend;
};
struct vc4_vs_key {
struct vc4_key base;
enum pipe_format attr_formats[8];
+ bool per_vertex_point_size;
};
static void
uint32_t uniform = c->num_uniforms++;
struct qreg u = { QFILE_UNIF, uniform };
+ if (uniform >= c->uniform_array_size) {
+ c->uniform_array_size = MAX2(MAX2(16, uniform + 1),
+ c->uniform_array_size * 2);
+
+ c->uniform_data = reralloc(c, c->uniform_data,
+ uint32_t,
+ c->uniform_array_size);
+ c->uniform_contents = reralloc(c, c->uniform_contents,
+ enum quniform_contents,
+ c->uniform_array_size);
+ }
+
c->uniform_contents[uniform] = contents;
c->uniform_data[uniform] = data;
get_temp_for_uniform(struct vc4_compile *c, enum quniform_contents contents,
uint32_t data)
{
- for (int i = 0; i < c->num_uniforms; i++) {
- if (c->uniform_contents[i] == contents &&
- c->uniform_data[i] == data)
- return c->uniforms[i];
- }
-
struct qreg u = add_uniform(c, contents, data);
struct qreg t = qir_MOV(c, u);
-
- resize_qreg_array(c, &c->uniforms, &c->uniforms_array_size,
- u.index + 1);
-
- c->uniforms[u.index] = t;
return t;
}
return dst;
}
+static struct qreg
+tgsi_to_qir_scalar(struct vc4_compile *c,
+ struct tgsi_full_instruction *tgsi_inst,
+ enum qop op, struct qreg *src, int i)
+{
+ struct qreg dst = qir_get_temp(c);
+ qir_emit(c, qir_inst(op, dst,
+ src[0 * 4 + 0],
+ c->undef));
+ return dst;
+}
+
+static struct qreg
+qir_srgb_decode(struct vc4_compile *c, struct qreg srgb)
+{
+ struct qreg low = qir_FMUL(c, srgb, qir_uniform_f(c, 1.0 / 12.92));
+ struct qreg high = qir_POW(c,
+ qir_FMUL(c,
+ qir_FADD(c,
+ srgb,
+ qir_uniform_f(c, 0.055)),
+ qir_uniform_f(c, 1.0 / 1.055)),
+ qir_uniform_f(c, 2.4));
+
+ qir_SF(c, qir_FSUB(c, srgb, qir_uniform_f(c, 0.04045)));
+ return qir_SEL_X_Y_NS(c, low, high);
+}
+
+static struct qreg
+qir_srgb_encode(struct vc4_compile *c, struct qreg linear)
+{
+ struct qreg low = qir_FMUL(c, linear, qir_uniform_f(c, 12.92));
+ struct qreg high = qir_FSUB(c,
+ qir_FMUL(c,
+ qir_uniform_f(c, 1.055),
+ qir_POW(c,
+ linear,
+ qir_uniform_f(c, 0.41666))),
+ qir_uniform_f(c, 0.055));
+
+ qir_SF(c, qir_FSUB(c, linear, qir_uniform_f(c, 0.0031308)));
+ return qir_SEL_X_Y_NS(c, low, high);
+}
+
static struct qreg
tgsi_to_qir_umul(struct vc4_compile *c,
struct tgsi_full_instruction *tgsi_inst,
src[2 * 4 + i]);
}
-static struct qreg
-tgsi_to_qir_lit(struct vc4_compile *c,
- struct tgsi_full_instruction *tgsi_inst,
- enum qop op, struct qreg *src, int i)
-{
- struct qreg x = src[0 * 4 + 0];
- struct qreg y = src[0 * 4 + 1];
- struct qreg w = src[0 * 4 + 3];
-
- switch (i) {
- case 0:
- case 3:
- return qir_uniform_f(c, 1.0);
- case 1:
- return qir_FMAX(c, src[0 * 4 + 0], qir_uniform_f(c, 0.0));
- case 2: {
- struct qreg zero = qir_uniform_f(c, 0.0);
-
- qir_SF(c, x);
- /* XXX: Clamp w to -128..128 */
- return qir_SEL_X_0_NC(c,
- qir_EXP2(c, qir_FMUL(c,
- w,
- qir_LOG2(c,
- qir_FMAX(c,
- y,
- zero)))));
- }
- default:
- assert(!"not reached");
- return c->undef;
- }
-}
-
static struct qreg
tgsi_to_qir_lrp(struct vc4_compile *c,
struct tgsi_full_instruction *tgsi_inst,
struct qreg s = src[0 * 4 + 0];
struct qreg t = src[0 * 4 + 1];
+ struct qreg r = src[0 * 4 + 2];
uint32_t unit = tgsi_inst->Src[1].Register.Index;
struct qreg proj = c->undef;
t = qir_FMUL(c, t, proj);
}
+ struct qreg texture_u[] = {
+ add_uniform(c, QUNIFORM_TEXTURE_CONFIG_P0, unit),
+ add_uniform(c, QUNIFORM_TEXTURE_CONFIG_P1, unit),
+ add_uniform(c, QUNIFORM_CONSTANT, 0),
+ add_uniform(c, QUNIFORM_CONSTANT, 0),
+ };
+ uint32_t next_texture_u = 0;
+
/* There is no native support for GL texture rectangle coordinates, so
* we have to rescale from ([0, width], [0, height]) to ([0, 1], [0,
* 1]).
unit));
}
- qir_TEX_T(c, t, add_uniform(c, QUNIFORM_TEXTURE_CONFIG_P0, unit));
+ if (tgsi_inst->Texture.Texture == TGSI_TEXTURE_CUBE ||
+ tgsi_inst->Texture.Texture == TGSI_TEXTURE_SHADOWCUBE) {
+ struct qreg ma = qir_FMAXABS(c, qir_FMAXABS(c, s, t), r);
+ struct qreg rcp_ma = qir_RCP(c, ma);
+ s = qir_FMUL(c, s, rcp_ma);
+ t = qir_FMUL(c, t, rcp_ma);
+ r = qir_FMUL(c, r, rcp_ma);
+
+ texture_u[2] = add_uniform(c, QUNIFORM_TEXTURE_CONFIG_P2, unit);
+
+ qir_TEX_R(c, r, texture_u[next_texture_u++]);
+ } else if (c->key->tex[unit].wrap_s == PIPE_TEX_WRAP_CLAMP_TO_BORDER ||
+ c->key->tex[unit].wrap_s == PIPE_TEX_WRAP_CLAMP ||
+ c->key->tex[unit].wrap_t == PIPE_TEX_WRAP_CLAMP_TO_BORDER ||
+ c->key->tex[unit].wrap_t == PIPE_TEX_WRAP_CLAMP) {
+ qir_TEX_R(c, get_temp_for_uniform(c, QUNIFORM_TEXTURE_BORDER_COLOR, unit),
+ texture_u[next_texture_u++]);
+ }
- struct qreg sampler_p1 = add_uniform(c, QUNIFORM_TEXTURE_CONFIG_P1,
- unit);
- if (tgsi_inst->Instruction.Opcode == TGSI_OPCODE_TXB) {
- qir_TEX_B(c, src[0 * 4 + 3], sampler_p1);
- qir_TEX_S(c, s, add_uniform(c, QUNIFORM_CONSTANT, 0));
- } else {
- qir_TEX_S(c, s, sampler_p1);
+ if (c->key->tex[unit].wrap_s == PIPE_TEX_WRAP_CLAMP) {
+ s = qir_FMIN(c, qir_FMAX(c, s, qir_uniform_f(c, 0.0)),
+ qir_uniform_f(c, 1.0));
+ }
+
+ if (c->key->tex[unit].wrap_t == PIPE_TEX_WRAP_CLAMP) {
+ t = qir_FMIN(c, qir_FMAX(c, t, qir_uniform_f(c, 0.0)),
+ qir_uniform_f(c, 1.0));
}
+ qir_TEX_T(c, t, texture_u[next_texture_u++]);
+
+ if (tgsi_inst->Instruction.Opcode == TGSI_OPCODE_TXB)
+ qir_TEX_B(c, src[0 * 4 + 3], texture_u[next_texture_u++]);
+
+ qir_TEX_S(c, s, texture_u[next_texture_u++]);
+
c->num_texture_samples++;
struct qreg r4 = qir_TEX_RESULT(c);
struct qreg depth_output;
- struct qreg compare = src[0 * 4 + 2];
-
- if (tgsi_inst->Instruction.Opcode == TGSI_OPCODE_TXP)
- compare = qir_FMUL(c, compare, proj);
-
struct qreg one = qir_uniform_f(c, 1.0f);
if (c->key->tex[unit].compare_mode) {
+ struct qreg compare = src[0 * 4 + 2];
+
+ if (tgsi_inst->Instruction.Opcode == TGSI_OPCODE_TXP)
+ compare = qir_FMUL(c, compare, proj);
+
switch (c->key->tex[unit].compare_func) {
case PIPE_FUNC_NEVER:
depth_output = qir_uniform_f(c, 0.0f);
}
const uint8_t *format_swiz = vc4_get_format_swizzle(format);
- uint8_t swiz[4];
- util_format_compose_swizzles(format_swiz, c->key->tex[unit].swizzle, swiz);
+ struct qreg texture_output[4];
+ for (int i = 0; i < 4; i++) {
+ texture_output[i] = get_swizzled_channel(c, unpacked,
+ format_swiz[i]);
+ }
+
+ if (util_format_is_srgb(format)) {
+ for (int i = 0; i < 3; i++)
+ texture_output[i] = qir_srgb_decode(c,
+ texture_output[i]);
+ }
+
for (int i = 0; i < 4; i++) {
if (!(tgsi_inst->Dst[0].Register.WriteMask & (1 << i)))
continue;
update_dst(c, tgsi_inst, i,
- get_swizzled_channel(c, unpacked, swiz[i]));
+ get_swizzled_channel(c, texture_output,
+ c->key->tex[unit].swizzle[i]));
}
}
-static struct qreg
-tgsi_to_qir_pow(struct vc4_compile *c,
- struct tgsi_full_instruction *tgsi_inst,
- enum qop op, struct qreg *src, int i)
-{
- /* Note that this instruction replicates its result from the x channel
- */
- return qir_EXP2(c, qir_FMUL(c,
- src[1 * 4 + 0],
- qir_LOG2(c, src[0 * 4 + 0])));
-}
-
static struct qreg
tgsi_to_qir_trunc(struct vc4_compile *c,
struct tgsi_full_instruction *tgsi_inst,
trunc);
}
-static struct qreg
-tgsi_to_qir_dp(struct vc4_compile *c,
- struct tgsi_full_instruction *tgsi_inst,
- int num, struct qreg *src, int i)
-{
- struct qreg sum = qir_FMUL(c, src[0 * 4 + 0], src[1 * 4 + 0]);
- for (int j = 1; j < num; j++) {
- sum = qir_FADD(c, sum, qir_FMUL(c,
- src[0 * 4 + j],
- src[1 * 4 + j]));
- }
- return sum;
-}
-
-static struct qreg
-tgsi_to_qir_dp2(struct vc4_compile *c,
- struct tgsi_full_instruction *tgsi_inst,
- enum qop op, struct qreg *src, int i)
-{
- return tgsi_to_qir_dp(c, tgsi_inst, 2, src, i);
-}
-
-static struct qreg
-tgsi_to_qir_dp3(struct vc4_compile *c,
- struct tgsi_full_instruction *tgsi_inst,
- enum qop op, struct qreg *src, int i)
-{
- return tgsi_to_qir_dp(c, tgsi_inst, 3, src, i);
-}
-
-static struct qreg
-tgsi_to_qir_dp4(struct vc4_compile *c,
- struct tgsi_full_instruction *tgsi_inst,
- enum qop op, struct qreg *src, int i)
-{
- return tgsi_to_qir_dp(c, tgsi_inst, 4, src, i);
-}
-
static struct qreg
tgsi_to_qir_abs(struct vc4_compile *c,
struct tgsi_full_instruction *tgsi_inst,
return sum;
}
+static struct qreg
+tgsi_to_qir_clamp(struct vc4_compile *c,
+ struct tgsi_full_instruction *tgsi_inst,
+ enum qop op, struct qreg *src, int i)
+{
+ return qir_FMAX(c, qir_FMIN(c,
+ src[0 * 4 + i],
+ src[2 * 4 + i]),
+ src[1 * 4 + i]);
+}
+
static void
emit_vertex_input(struct vc4_compile *c, int attr)
{
c->inputs[attr * 4 + 3] = qir_RCP(c, qir_FRAG_W(c));
}
+static void
+emit_point_coord_input(struct vc4_compile *c, int attr)
+{
+ if (c->point_x.file == QFILE_NULL) {
+ c->point_x = qir_uniform_f(c, 0.0);
+ c->point_y = qir_uniform_f(c, 0.0);
+ }
+
+ c->inputs[attr * 4 + 0] = c->point_x;
+ if (c->fs_key->point_coord_upper_left) {
+ c->inputs[attr * 4 + 1] = qir_FSUB(c,
+ qir_uniform_f(c, 1.0),
+ c->point_y);
+ } else {
+ c->inputs[attr * 4 + 1] = c->point_y;
+ }
+ c->inputs[attr * 4 + 2] = qir_uniform_f(c, 0.0);
+ c->inputs[attr * 4 + 3] = qir_uniform_f(c, 1.0);
+}
+
static struct qreg
emit_fragment_varying(struct vc4_compile *c, int index)
{
}
static void
-emit_fragment_input(struct vc4_compile *c, int attr)
+emit_fragment_input(struct vc4_compile *c, int attr,
+ struct tgsi_full_declaration *decl)
{
for (int i = 0; i < 4; i++) {
c->inputs[attr * 4 + i] =
emit_fragment_varying(c, attr * 4 + i);
c->num_inputs++;
+
+ if (decl->Semantic.Name == TGSI_SEMANTIC_COLOR ||
+ decl->Semantic.Name == TGSI_SEMANTIC_BCOLOR)
+ c->color_inputs |= 1 << i;
}
}
+static void
+emit_face_input(struct vc4_compile *c, int attr)
+{
+ c->inputs[attr * 4 + 0] = qir_FSUB(c,
+ qir_uniform_f(c, 1.0),
+ qir_FMUL(c,
+ qir_ITOF(c, qir_FRAG_REV_FLAG(c)),
+ qir_uniform_f(c, 2.0)));
+ c->inputs[attr * 4 + 1] = qir_uniform_f(c, 0.0);
+ c->inputs[attr * 4 + 2] = qir_uniform_f(c, 0.0);
+ c->inputs[attr * 4 + 3] = qir_uniform_f(c, 1.0);
+}
+
static void
emit_tgsi_declaration(struct vc4_compile *c,
struct tgsi_full_declaration *decl)
{
switch (decl->Declaration.File) {
- case TGSI_FILE_TEMPORARY:
+ case TGSI_FILE_TEMPORARY: {
+ uint32_t old_size = c->temps_array_size;
resize_qreg_array(c, &c->temps, &c->temps_array_size,
(decl->Range.Last + 1) * 4);
+
+ for (int i = old_size; i < c->temps_array_size; i++)
+ c->temps[i] = qir_uniform_ui(c, 0);
break;
+ }
case TGSI_FILE_INPUT:
resize_qreg_array(c, &c->inputs, &c->inputs_array_size,
if (decl->Semantic.Name ==
TGSI_SEMANTIC_POSITION) {
emit_fragcoord_input(c, i);
+ } else if (decl->Semantic.Name == TGSI_SEMANTIC_FACE) {
+ emit_face_input(c, i);
+ } else if (decl->Semantic.Name == TGSI_SEMANTIC_GENERIC &&
+ (c->fs_key->point_sprite_mask &
+ (1 << decl->Semantic.Index))) {
+ emit_point_coord_input(c, i);
} else {
- emit_fragment_input(c, i);
+ emit_fragment_input(c, i, decl);
}
} else {
emit_vertex_input(c, i);
case TGSI_SEMANTIC_COLOR:
c->output_color_index = decl->Range.First * 4;
break;
+ case TGSI_SEMANTIC_PSIZE:
+ c->output_point_size_index = decl->Range.First * 4;
+ break;
}
break;
[TGSI_OPCODE_CMP] = { 0, tgsi_to_qir_cmp },
[TGSI_OPCODE_MAD] = { 0, tgsi_to_qir_mad },
- [TGSI_OPCODE_DP2] = { 0, tgsi_to_qir_dp2 },
- [TGSI_OPCODE_DP3] = { 0, tgsi_to_qir_dp3 },
- [TGSI_OPCODE_DP4] = { 0, tgsi_to_qir_dp4 },
- [TGSI_OPCODE_RCP] = { QOP_RCP, tgsi_to_qir_alu },
- [TGSI_OPCODE_RSQ] = { QOP_RSQ, tgsi_to_qir_alu },
- [TGSI_OPCODE_EX2] = { QOP_EXP2, tgsi_to_qir_alu },
- [TGSI_OPCODE_LG2] = { QOP_LOG2, tgsi_to_qir_alu },
- [TGSI_OPCODE_LIT] = { 0, tgsi_to_qir_lit },
+ [TGSI_OPCODE_RCP] = { QOP_RCP, tgsi_to_qir_scalar },
+ [TGSI_OPCODE_RSQ] = { QOP_RSQ, tgsi_to_qir_scalar },
+ [TGSI_OPCODE_EX2] = { QOP_EXP2, tgsi_to_qir_scalar },
+ [TGSI_OPCODE_LG2] = { QOP_LOG2, tgsi_to_qir_scalar },
[TGSI_OPCODE_LRP] = { 0, tgsi_to_qir_lrp },
- [TGSI_OPCODE_POW] = { 0, tgsi_to_qir_pow },
[TGSI_OPCODE_TRUNC] = { 0, tgsi_to_qir_trunc },
[TGSI_OPCODE_FRC] = { 0, tgsi_to_qir_frc },
[TGSI_OPCODE_FLR] = { 0, tgsi_to_qir_flr },
[TGSI_OPCODE_SIN] = { 0, tgsi_to_qir_sin },
[TGSI_OPCODE_COS] = { 0, tgsi_to_qir_cos },
+ [TGSI_OPCODE_CLAMP] = { 0, tgsi_to_qir_clamp },
};
static int asdf = 0;
uint32_t tgsi_op = tgsi_inst->Instruction.Opcode;
blend->alpha_func);
}
+static void
+alpha_test_discard(struct vc4_compile *c)
+{
+ struct qreg src_alpha;
+ struct qreg alpha_ref = get_temp_for_uniform(c, QUNIFORM_ALPHA_REF, 0);
+
+ if (!c->fs_key->alpha_test)
+ return;
+
+ if (c->output_color_index != -1)
+ src_alpha = c->outputs[c->output_color_index + 3];
+ else
+ src_alpha = qir_uniform_f(c, 1.0);
+
+ if (c->discard.file == QFILE_NULL)
+ c->discard = qir_uniform_f(c, 0.0);
+
+ switch (c->fs_key->alpha_test_func) {
+ case PIPE_FUNC_NEVER:
+ c->discard = qir_uniform_f(c, 1.0);
+ break;
+ case PIPE_FUNC_ALWAYS:
+ break;
+ case PIPE_FUNC_EQUAL:
+ qir_SF(c, qir_FSUB(c, src_alpha, alpha_ref));
+ c->discard = qir_SEL_X_Y_ZS(c, c->discard,
+ qir_uniform_f(c, 1.0));
+ break;
+ case PIPE_FUNC_NOTEQUAL:
+ qir_SF(c, qir_FSUB(c, src_alpha, alpha_ref));
+ c->discard = qir_SEL_X_Y_ZC(c, c->discard,
+ qir_uniform_f(c, 1.0));
+ break;
+ case PIPE_FUNC_GREATER:
+ qir_SF(c, qir_FSUB(c, src_alpha, alpha_ref));
+ c->discard = qir_SEL_X_Y_NC(c, c->discard,
+ qir_uniform_f(c, 1.0));
+ break;
+ case PIPE_FUNC_GEQUAL:
+ qir_SF(c, qir_FSUB(c, alpha_ref, src_alpha));
+ c->discard = qir_SEL_X_Y_NS(c, c->discard,
+ qir_uniform_f(c, 1.0));
+ break;
+ case PIPE_FUNC_LESS:
+ qir_SF(c, qir_FSUB(c, src_alpha, alpha_ref));
+ c->discard = qir_SEL_X_Y_NS(c, c->discard,
+ qir_uniform_f(c, 1.0));
+ break;
+ case PIPE_FUNC_LEQUAL:
+ qir_SF(c, qir_FSUB(c, alpha_ref, src_alpha));
+ c->discard = qir_SEL_X_Y_NC(c, c->discard,
+ qir_uniform_f(c, 1.0));
+ break;
+ }
+}
+
static void
emit_frag_end(struct vc4_compile *c)
{
+ alpha_test_discard(c);
+
enum pipe_format color_format = c->fs_key->color_format;
const uint8_t *format_swiz = vc4_get_format_swizzle(color_format);
struct qreg tlb_read_color[4] = { c->undef, c->undef, c->undef, c->undef };
struct qreg dst_color[4] = { c->undef, c->undef, c->undef, c->undef };
+ struct qreg linear_dst_color[4] = { c->undef, c->undef, c->undef, c->undef };
if (c->fs_key->blend.blend_enable ||
c->fs_key->blend.colormask != 0xf) {
struct qreg r4 = qir_TLB_COLOR_READ(c);
for (int i = 0; i < 4; i++)
tlb_read_color[i] = qir_R4_UNPACK(c, r4, i);
- for (int i = 0; i < 4; i++)
+ for (int i = 0; i < 4; i++) {
dst_color[i] = get_swizzled_channel(c,
tlb_read_color,
format_swiz[i]);
+ if (util_format_is_srgb(color_format) && i != 3) {
+ linear_dst_color[i] =
+ qir_srgb_decode(c, dst_color[i]);
+ } else {
+ linear_dst_color[i] = dst_color[i];
+ }
+ }
}
struct qreg blend_color[4];
struct qreg undef_array[4] = {
c->undef, c->undef, c->undef, c->undef
};
- vc4_blend(c, blend_color, dst_color,
+ vc4_blend(c, blend_color, linear_dst_color,
(c->output_color_index != -1 ?
c->outputs + c->output_color_index :
undef_array));
+ if (util_format_is_srgb(color_format)) {
+ for (int i = 0; i < 3; i++)
+ blend_color[i] = qir_srgb_encode(c, blend_color[i]);
+ }
+
/* If the bit isn't set in the color mask, then just return the
* original dst color, instead.
*/
qir_VPM_WRITE(c, rcp_w);
}
+static void
+emit_point_size_write(struct vc4_compile *c)
+{
+ struct qreg point_size;
+
+ if (c->output_point_size_index)
+ point_size = c->outputs[c->output_point_size_index + 3];
+ else
+ point_size = qir_uniform_f(c, 1.0);
+
+ /* Workaround: HW-2726 PTB does not handle zero-size points (BCM2835,
+ * BCM21553).
+ */
+ point_size = qir_FMAX(c, point_size, qir_uniform_f(c, .125));
+
+ qir_VPM_WRITE(c, point_size);
+}
+
static void
emit_vert_end(struct vc4_compile *c)
{
emit_scaled_viewport_write(c, rcp_w);
emit_zs_write(c, rcp_w);
emit_rcp_wc_write(c, rcp_w);
+ if (c->vs_key->per_vertex_point_size)
+ emit_point_size_write(c);
for (int i = 4; i < c->num_outputs; i++) {
qir_VPM_WRITE(c, c->outputs[i]);
emit_scaled_viewport_write(c, rcp_w);
emit_zs_write(c, rcp_w);
emit_rcp_wc_write(c, rcp_w);
+ if (c->vs_key->per_vertex_point_size)
+ emit_point_size_write(c);
}
static struct vc4_compile *
int ret;
c->stage = stage;
-
- c->uniform_data = ralloc_array(c, uint32_t, 1024);
- c->uniform_contents = ralloc_array(c, enum quniform_contents, 1024);
-
- c->shader_state = key->shader_state;
- ret = tgsi_parse_init(&c->parser, c->shader_state->tokens);
- assert(ret == TGSI_PARSE_OK);
-
- if (vc4_debug & VC4_DEBUG_TGSI) {
- fprintf(stderr, "TGSI:\n");
- tgsi_dump(c->shader_state->tokens, 0);
- }
+ c->shader_state = &key->shader_state->base;
c->key = key;
switch (stage) {
break;
}
+ const struct tgsi_token *tokens = key->shader_state->base.tokens;
+ if (c->fs_key && c->fs_key->light_twoside) {
+ if (!key->shader_state->twoside_tokens) {
+ const struct tgsi_lowering_config lowering_config = {
+ .color_two_side = true,
+ };
+ struct tgsi_shader_info info;
+ key->shader_state->twoside_tokens =
+ tgsi_transform_lowering(&lowering_config,
+ key->shader_state->base.tokens,
+ &info);
+
+ /* If no transformation occurred, then NULL is
+ * returned and we just use our original tokens.
+ */
+ if (!key->shader_state->twoside_tokens) {
+ key->shader_state->twoside_tokens =
+ key->shader_state->base.tokens;
+ }
+ }
+ tokens = key->shader_state->twoside_tokens;
+ }
+
+ ret = tgsi_parse_init(&c->parser, tokens);
+ assert(ret == TGSI_PARSE_OK);
+
+ if (vc4_debug & VC4_DEBUG_TGSI) {
+ fprintf(stderr, "TGSI:\n");
+ tgsi_dump(tokens, 0);
+ }
+
while (!tgsi_parse_end_of_tokens(&c->parser)) {
tgsi_parse_token(&c->parser);
vc4_shader_state_create(struct pipe_context *pctx,
const struct pipe_shader_state *cso)
{
- struct pipe_shader_state *so = CALLOC_STRUCT(pipe_shader_state);
+ struct vc4_uncompiled_shader *so = CALLOC_STRUCT(vc4_uncompiled_shader);
if (!so)
return NULL;
- so->tokens = tgsi_dup_tokens(cso->tokens);
+ const struct tgsi_lowering_config lowering_config = {
+ .lower_DST = true,
+ .lower_XPD = true,
+ .lower_SCS = true,
+ .lower_POW = true,
+ .lower_LIT = true,
+ .lower_EXP = true,
+ .lower_LOG = true,
+ .lower_DP4 = true,
+ .lower_DP3 = true,
+ .lower_DPH = true,
+ .lower_DP2 = true,
+ .lower_DP2A = true,
+ };
+
+ struct tgsi_shader_info info;
+ so->base.tokens = tgsi_transform_lowering(&lowering_config, cso->tokens, &info);
+ if (!so->base.tokens)
+ so->base.tokens = tgsi_dup_tokens(cso->tokens);
return so;
}
QSTAGE_FRAG,
&key->base);
shader->num_inputs = c->num_inputs;
+ shader->color_inputs = c->color_inputs;
copy_uniform_state_to_shader(shader, 0, c);
shader->bo = vc4_bo_alloc_mem(vc4->screen, c->qpu_insts,
c->qpu_inst_count * sizeof(uint64_t),
for (int i = 0; i < texstate->num_textures; i++) {
struct pipe_sampler_view *sampler = texstate->textures[i];
struct pipe_sampler_state *sampler_state =
-texstate->samplers[i];
+ texstate->samplers[i];
if (sampler) {
- struct pipe_resource *prsc = sampler->texture;
- key->tex[i].format = prsc->format;
+ key->tex[i].format = sampler->format;
key->tex[i].swizzle[0] = sampler->swizzle_r;
key->tex[i].swizzle[1] = sampler->swizzle_g;
key->tex[i].swizzle[2] = sampler->swizzle_b;
key->tex[i].swizzle[3] = sampler->swizzle_a;
key->tex[i].compare_mode = sampler_state->compare_mode;
key->tex[i].compare_func = sampler_state->compare_func;
+ key->tex[i].wrap_s = sampler_state->wrap_s;
+ key->tex[i].wrap_t = sampler_state->wrap_t;
}
}
}
key->stencil_full_writemasks = vc4->zsa->stencil_uniforms[2] != 0;
key->depth_enabled = (vc4->zsa->base.depth.enabled ||
key->stencil_enabled);
+ if (vc4->zsa->base.alpha.enabled) {
+ key->alpha_test = true;
+ key->alpha_test_func = vc4->zsa->base.alpha.func;
+ }
+
+ if (key->is_points) {
+ key->point_sprite_mask =
+ vc4->rasterizer->base.sprite_coord_enable;
+ key->point_coord_upper_left =
+ (vc4->rasterizer->base.sprite_coord_mode ==
+ PIPE_SPRITE_COORD_UPPER_LEFT);
+ }
+
+ key->light_twoside = vc4->rasterizer->base.light_twoside;
vc4->prog.fs = util_hash_table_get(vc4->fs_cache, key);
if (vc4->prog.fs)
vc4_fs_compile(vc4, shader, key);
util_hash_table_set(vc4->fs_cache, key, shader);
+ if (vc4->rasterizer->base.flatshade &&
+ vc4->prog.fs &&
+ vc4->prog.fs->color_inputs != shader->color_inputs) {
+ vc4->dirty |= VC4_DIRTY_FLAT_SHADE_FLAGS;
+ }
+
vc4->prog.fs = shader;
}
static void
-vc4_update_compiled_vs(struct vc4_context *vc4)
+vc4_update_compiled_vs(struct vc4_context *vc4, uint8_t prim_mode)
{
struct vc4_vs_key local_key;
struct vc4_vs_key *key = &local_key;
for (int i = 0; i < ARRAY_SIZE(key->attr_formats); i++)
key->attr_formats[i] = vc4->vtx->pipe[i].src_format;
+ key->per_vertex_point_size =
+ (prim_mode == PIPE_PRIM_POINTS &&
+ vc4->rasterizer->base.point_size_per_vertex);
+
vc4->prog.vs = util_hash_table_get(vc4->vs_cache, key);
if (vc4->prog.vs)
return;
vc4_update_compiled_shaders(struct vc4_context *vc4, uint8_t prim_mode)
{
vc4_update_compiled_fs(vc4, prim_mode);
- vc4_update_compiled_vs(vc4);
+ vc4_update_compiled_vs(vc4, prim_mode);
}
static unsigned
struct delete_state {
struct vc4_context *vc4;
- struct pipe_shader_state *shader_state;
+ struct vc4_uncompiled_shader *shader_state;
};
static enum pipe_error
vc4_shader_state_delete(struct pipe_context *pctx, void *hwcso)
{
struct vc4_context *vc4 = vc4_context(pctx);
- struct pipe_shader_state *so = hwcso;
+ struct vc4_uncompiled_shader *so = hwcso;
struct delete_state del;
del.vc4 = vc4;
util_hash_table_foreach(vc4->fs_cache, fs_delete_from_cache, &del);
util_hash_table_foreach(vc4->vs_cache, vs_delete_from_cache, &del);
- free((void *)so->tokens);
+ if (so->twoside_tokens != so->base.tokens)
+ free((void *)so->twoside_tokens);
+ free((void *)so->base.tokens);
free(so);
}
-static uint32_t translate_wrap(uint32_t p_wrap)
+static uint32_t translate_wrap(uint32_t p_wrap, bool using_nearest)
{
switch (p_wrap) {
case PIPE_TEX_WRAP_REPEAT:
return 0;
- case PIPE_TEX_WRAP_CLAMP:
case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
return 1;
case PIPE_TEX_WRAP_MIRROR_REPEAT:
return 2;
case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
return 3;
+ case PIPE_TEX_WRAP_CLAMP:
+ return (using_nearest ? 1 : 3);
default:
fprintf(stderr, "Unknown wrap mode %d\n", p_wrap);
assert(!"not reached");
struct vc4_resource *rsc = vc4_resource(texture->texture);
cl_reloc(vc4, &vc4->uniforms, rsc->bo,
- rsc->slices[0].offset | texture->u.tex.last_level |
- ((rsc->vc4_format & 7) << 4));
+ VC4_SET_FIELD(rsc->slices[0].offset >> 12, VC4_TEX_P0_OFFSET) |
+ VC4_SET_FIELD(texture->u.tex.last_level, VC4_TEX_P0_MIPLVLS) |
+ VC4_SET_FIELD(texture->target == PIPE_TEXTURE_CUBE,
+ VC4_TEX_P0_CMMODE) |
+ VC4_SET_FIELD(rsc->vc4_format & 7, VC4_TEX_P0_TYPE));
}
static void
struct pipe_sampler_view *texture = texstate->textures[unit];
struct vc4_resource *rsc = vc4_resource(texture->texture);
struct pipe_sampler_state *sampler = texstate->samplers[unit];
- static const uint32_t mipfilter_map[] = {
- [PIPE_TEX_MIPFILTER_NEAREST] = 2,
- [PIPE_TEX_MIPFILTER_LINEAR] = 4,
- [PIPE_TEX_MIPFILTER_NONE] = 0
+ static const uint8_t minfilter_map[6] = {
+ VC4_TEX_P1_MINFILT_NEAR_MIP_NEAR,
+ VC4_TEX_P1_MINFILT_LIN_MIP_NEAR,
+ VC4_TEX_P1_MINFILT_NEAR_MIP_LIN,
+ VC4_TEX_P1_MINFILT_LIN_MIP_LIN,
+ VC4_TEX_P1_MINFILT_NEAREST,
+ VC4_TEX_P1_MINFILT_LINEAR,
};
- static const uint32_t imgfilter_map[] = {
- [PIPE_TEX_FILTER_NEAREST] = 1,
- [PIPE_TEX_FILTER_LINEAR] = 0,
+ static const uint32_t magfilter_map[] = {
+ [PIPE_TEX_FILTER_NEAREST] = VC4_TEX_P1_MAGFILT_NEAREST,
+ [PIPE_TEX_FILTER_LINEAR] = VC4_TEX_P1_MAGFILT_LINEAR,
};
+ bool either_nearest =
+ (sampler->mag_img_filter == PIPE_TEX_MIPFILTER_NEAREST ||
+ sampler->min_img_filter == PIPE_TEX_MIPFILTER_NEAREST);
+
+ cl_u32(&vc4->uniforms,
+ VC4_SET_FIELD(rsc->vc4_format >> 4, VC4_TEX_P1_TYPE4) |
+ VC4_SET_FIELD(texture->texture->height0 & 2047,
+ VC4_TEX_P1_HEIGHT) |
+ VC4_SET_FIELD(texture->texture->width0 & 2047,
+ VC4_TEX_P1_WIDTH) |
+ VC4_SET_FIELD(magfilter_map[sampler->mag_img_filter],
+ VC4_TEX_P1_MAGFILT) |
+ VC4_SET_FIELD(minfilter_map[sampler->min_mip_filter * 2 +
+ sampler->min_img_filter],
+ VC4_TEX_P1_MINFILT) |
+ VC4_SET_FIELD(translate_wrap(sampler->wrap_s, either_nearest),
+ VC4_TEX_P1_WRAP_S) |
+ VC4_SET_FIELD(translate_wrap(sampler->wrap_t, either_nearest),
+ VC4_TEX_P1_WRAP_T));
+}
+
+static void
+write_texture_p2(struct vc4_context *vc4,
+ struct vc4_texture_stateobj *texstate,
+ uint32_t unit)
+{
+ struct pipe_sampler_view *texture = texstate->textures[unit];
+ struct vc4_resource *rsc = vc4_resource(texture->texture);
+
cl_u32(&vc4->uniforms,
- ((rsc->vc4_format >> 4) << 31) |
- (texture->texture->height0 << 20) |
- (texture->texture->width0 << 8) |
- (imgfilter_map[sampler->mag_img_filter] << 7) |
- ((imgfilter_map[sampler->min_img_filter] +
- mipfilter_map[sampler->min_mip_filter]) << 4) |
- (translate_wrap(sampler->wrap_t) << 2) |
- (translate_wrap(sampler->wrap_s) << 0));
+ VC4_SET_FIELD(VC4_TEX_P2_PTYPE_CUBE_MAP_STRIDE,
+ VC4_TEX_P2_PTYPE) |
+ VC4_SET_FIELD(rsc->cube_map_stride >> 12, VC4_TEX_P2_CMST));
+}
+
+
+#define SWIZ(x,y,z,w) { \
+ UTIL_FORMAT_SWIZZLE_##x, \
+ UTIL_FORMAT_SWIZZLE_##y, \
+ UTIL_FORMAT_SWIZZLE_##z, \
+ UTIL_FORMAT_SWIZZLE_##w \
+}
+
+static void
+write_texture_border_color(struct vc4_context *vc4,
+ struct vc4_texture_stateobj *texstate,
+ uint32_t unit)
+{
+ struct pipe_sampler_state *sampler = texstate->samplers[unit];
+ struct pipe_sampler_view *texture = texstate->textures[unit];
+ struct vc4_resource *rsc = vc4_resource(texture->texture);
+ union util_color uc;
+
+ const struct util_format_description *tex_format_desc =
+ util_format_description(texture->format);
+
+ float border_color[4];
+ for (int i = 0; i < 4; i++)
+ border_color[i] = sampler->border_color.f[i];
+ if (util_format_is_srgb(texture->format)) {
+ for (int i = 0; i < 3; i++)
+ border_color[i] =
+ util_format_linear_to_srgb_float(border_color[i]);
+ }
+
+ /* Turn the border color into the layout of channels that it would
+ * have when stored as texture contents.
+ */
+ float storage_color[4];
+ util_format_unswizzle_4f(storage_color,
+ border_color,
+ tex_format_desc->swizzle);
+
+ /* Now, pack so that when the vc4_format-sampled texture contents are
+ * replaced with our border color, the vc4_get_format_swizzle()
+ * swizzling will get the right channels.
+ */
+ if (util_format_is_depth_or_stencil(texture->format)) {
+ uc.ui[0] = util_pack_z(PIPE_FORMAT_Z24X8_UNORM,
+ sampler->border_color.f[0]) << 8;
+ } else {
+ switch (rsc->vc4_format) {
+ default:
+ case VC4_TEXTURE_TYPE_RGBA8888:
+ util_pack_color(storage_color,
+ PIPE_FORMAT_R8G8B8A8_UNORM, &uc);
+ break;
+ case VC4_TEXTURE_TYPE_RGBA4444:
+ util_pack_color(storage_color,
+ PIPE_FORMAT_A8B8G8R8_UNORM, &uc);
+ break;
+ case VC4_TEXTURE_TYPE_RGB565:
+ util_pack_color(storage_color,
+ PIPE_FORMAT_B8G8R8A8_UNORM, &uc);
+ break;
+ case VC4_TEXTURE_TYPE_ALPHA:
+ uc.ui[0] = float_to_ubyte(storage_color[0]) << 24;
+ break;
+ case VC4_TEXTURE_TYPE_LUMALPHA:
+ uc.ui[0] = ((float_to_ubyte(storage_color[1]) << 24) |
+ (float_to_ubyte(storage_color[0]) << 0));
+ break;
+ }
+ }
+
+ cl_u32(&vc4->uniforms, uc.ui[0]);
}
static uint32_t
write_texture_p1(vc4, texstate, uinfo->data[i]);
break;
+ case QUNIFORM_TEXTURE_CONFIG_P2:
+ write_texture_p2(vc4, texstate, uinfo->data[i]);
+ break;
+
+ case QUNIFORM_TEXTURE_BORDER_COLOR:
+ write_texture_border_color(vc4, texstate, uinfo->data[i]);
+ break;
+
case QUNIFORM_TEXRECT_SCALE_X:
case QUNIFORM_TEXRECT_SCALE_Y:
cl_u32(&vc4->uniforms,
(vc4->stencil_ref.ref_value[uinfo->data[i]] << 8) :
0));
break;
+
+ case QUNIFORM_ALPHA_REF:
+ cl_f(&vc4->uniforms, vc4->zsa->base.alpha.ref_value);
+ break;
}
#if 0
uint32_t written_val = *(uint32_t *)(vc4->uniforms.next - 4);