* IN THE SOFTWARE.
*/
-#include <stdio.h>
#include <inttypes.h>
#include "pipe/p_state.h"
#include "util/u_format.h"
#include "util/u_hash_table.h"
#include "util/u_hash.h"
#include "util/u_memory.h"
-#include "tgsi/tgsi_parse.h"
+#include "util/ralloc.h"
#include "tgsi/tgsi_dump.h"
+#include "tgsi/tgsi_info.h"
#include "vc4_context.h"
#include "vc4_qpu.h"
#include "simpenrose/simpenrose.h"
#endif
-struct tgsi_to_qir {
- struct tgsi_parse_context parser;
- struct qcompile *c;
- struct qreg *temps;
- struct qreg *inputs;
- struct qreg *outputs;
- struct qreg *uniforms;
- struct qreg *consts;
- struct qreg line_x, point_x, point_y;
-
- uint32_t num_consts;
-
- struct pipe_shader_state *shader_state;
- struct vc4_fs_key *fs_key;
- struct vc4_vs_key *vs_key;
-
- uint32_t *uniform_data;
- enum quniform_contents *uniform_contents;
- uint32_t num_uniforms;
- uint32_t num_outputs;
- uint32_t num_texture_samples;
-};
-
struct vc4_key {
struct pipe_shader_state *shader_state;
+ struct {
+ enum pipe_format format;
+ unsigned compare_mode:1;
+ unsigned compare_func:3;
+ uint8_t swizzle[4];
+ } tex[VC4_MAX_TEXTURE_SAMPLERS];
};
struct vc4_fs_key {
struct vc4_key base;
enum pipe_format color_format;
bool depth_enabled;
+ bool stencil_enabled;
+ bool stencil_twoside;
+ bool stencil_full_writemasks;
bool is_points;
bool is_lines;
+ bool alpha_test;
+ bool point_coord_upper_left;
+ 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
+resize_qreg_array(struct vc4_compile *c,
+ struct qreg **regs,
+ uint32_t *size,
+ uint32_t decl_size)
+{
+ if (*size >= decl_size)
+ return;
+
+ uint32_t old_size = *size;
+ *size = MAX2(*size * 2, decl_size);
+ *regs = reralloc(c, *regs, struct qreg, *size);
+ if (!*regs) {
+ fprintf(stderr, "Malloc failure\n");
+ abort();
+ }
+
+ for (uint32_t i = old_size; i < *size; i++)
+ (*regs)[i] = c->undef;
+}
+
static struct qreg
-add_uniform(struct tgsi_to_qir *trans,
+add_uniform(struct vc4_compile *c,
enum quniform_contents contents,
uint32_t data)
{
- uint32_t uniform = trans->num_uniforms++;
+ uint32_t uniform = c->num_uniforms++;
struct qreg u = { QFILE_UNIF, uniform };
- trans->uniform_contents[uniform] = contents;
- trans->uniform_data[uniform] = data;
+ c->uniform_contents[uniform] = contents;
+ c->uniform_data[uniform] = data;
return u;
}
static struct qreg
-get_temp_for_uniform(struct tgsi_to_qir *trans, enum quniform_contents contents,
+get_temp_for_uniform(struct vc4_compile *c, enum quniform_contents contents,
uint32_t data)
{
- struct qcompile *c = trans->c;
-
- for (int i = 0; i < trans->num_uniforms; i++) {
- if (trans->uniform_contents[i] == contents &&
- trans->uniform_data[i] == data)
- return trans->uniforms[i];
+ 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(trans, contents, data);
+ struct qreg u = add_uniform(c, contents, data);
struct qreg t = qir_MOV(c, u);
- trans->uniforms[u.index] = t;
+ resize_qreg_array(c, &c->uniforms, &c->uniforms_array_size,
+ u.index + 1);
+
+ c->uniforms[u.index] = t;
return t;
}
static struct qreg
-qir_uniform_ui(struct tgsi_to_qir *trans, uint32_t ui)
+qir_uniform_ui(struct vc4_compile *c, uint32_t ui)
{
- return get_temp_for_uniform(trans, QUNIFORM_CONSTANT, ui);
+ return get_temp_for_uniform(c, QUNIFORM_CONSTANT, ui);
}
static struct qreg
-qir_uniform_f(struct tgsi_to_qir *trans, float f)
+qir_uniform_f(struct vc4_compile *c, float f)
{
- return qir_uniform_ui(trans, fui(f));
+ return qir_uniform_ui(c, fui(f));
}
static struct qreg
-get_src(struct tgsi_to_qir *trans, struct tgsi_src_register *src, int i)
+get_src(struct vc4_compile *c, unsigned tgsi_op,
+ struct tgsi_src_register *src, int i)
{
- struct qcompile *c = trans->c;
struct qreg r = c->undef;
uint32_t s = i;
case TGSI_FILE_NULL:
return r;
case TGSI_FILE_TEMPORARY:
- r = trans->temps[src->Index * 4 + s];
+ r = c->temps[src->Index * 4 + s];
break;
case TGSI_FILE_IMMEDIATE:
- r = trans->consts[src->Index * 4 + s];
+ r = c->consts[src->Index * 4 + s];
break;
case TGSI_FILE_CONSTANT:
- r = get_temp_for_uniform(trans, QUNIFORM_UNIFORM,
+ r = get_temp_for_uniform(c, QUNIFORM_UNIFORM,
src->Index * 4 + s);
break;
case TGSI_FILE_INPUT:
- r = trans->inputs[src->Index * 4 + s];
+ r = c->inputs[src->Index * 4 + s];
break;
case TGSI_FILE_SAMPLER:
case TGSI_FILE_SAMPLER_VIEW:
if (src->Absolute)
r = qir_FMAXABS(c, r, r);
- if (src->Negate)
- r = qir_FSUB(c, qir_uniform_f(trans, 0), r);
+ if (src->Negate) {
+ switch (tgsi_opcode_infer_src_type(tgsi_op)) {
+ case TGSI_TYPE_SIGNED:
+ case TGSI_TYPE_UNSIGNED:
+ r = qir_SUB(c, qir_uniform_ui(c, 0), r);
+ break;
+ default:
+ r = qir_FSUB(c, qir_uniform_f(c, 0.0), r);
+ break;
+ }
+ }
return r;
};
static void
-update_dst(struct tgsi_to_qir *trans, struct tgsi_full_instruction *tgsi_inst,
+update_dst(struct vc4_compile *c, struct tgsi_full_instruction *tgsi_inst,
int i, struct qreg val)
{
struct tgsi_dst_register *tgsi_dst = &tgsi_inst->Dst[0].Register;
switch (tgsi_dst->File) {
case TGSI_FILE_TEMPORARY:
- trans->temps[tgsi_dst->Index * 4 + i] = val;
+ c->temps[tgsi_dst->Index * 4 + i] = val;
break;
case TGSI_FILE_OUTPUT:
- trans->outputs[tgsi_dst->Index * 4 + i] = val;
- trans->num_outputs = MAX2(trans->num_outputs,
- tgsi_dst->Index * 4 + i + 1);
+ c->outputs[tgsi_dst->Index * 4 + i] = val;
+ c->num_outputs = MAX2(c->num_outputs,
+ tgsi_dst->Index * 4 + i + 1);
break;
default:
fprintf(stderr, "unknown dst file %d\n", tgsi_dst->File);
};
static struct qreg
-tgsi_to_qir_alu(struct tgsi_to_qir *trans,
+get_swizzled_channel(struct vc4_compile *c,
+ struct qreg *srcs, int swiz)
+{
+ switch (swiz) {
+ default:
+ case UTIL_FORMAT_SWIZZLE_NONE:
+ fprintf(stderr, "warning: unknown swizzle\n");
+ /* FALLTHROUGH */
+ case UTIL_FORMAT_SWIZZLE_0:
+ return qir_uniform_f(c, 0.0);
+ case UTIL_FORMAT_SWIZZLE_1:
+ return qir_uniform_f(c, 1.0);
+ case UTIL_FORMAT_SWIZZLE_X:
+ case UTIL_FORMAT_SWIZZLE_Y:
+ case UTIL_FORMAT_SWIZZLE_Z:
+ case UTIL_FORMAT_SWIZZLE_W:
+ return srcs[swiz];
+ }
+}
+
+static struct qreg
+tgsi_to_qir_alu(struct vc4_compile *c,
struct tgsi_full_instruction *tgsi_inst,
enum qop op, struct qreg *src, int i)
{
- struct qcompile *c = trans->c;
struct qreg dst = qir_get_temp(c);
qir_emit(c, qir_inst4(op, dst,
src[0 * 4 + i],
}
static struct qreg
-tgsi_to_qir_mad(struct tgsi_to_qir *trans,
+tgsi_to_qir_umul(struct vc4_compile *c,
+ struct tgsi_full_instruction *tgsi_inst,
+ enum qop op, struct qreg *src, int i)
+{
+ struct qreg src0_hi = qir_SHR(c, src[0 * 4 + i],
+ qir_uniform_ui(c, 16));
+ struct qreg src0_lo = qir_AND(c, src[0 * 4 + i],
+ qir_uniform_ui(c, 0xffff));
+ struct qreg src1_hi = qir_SHR(c, src[1 * 4 + i],
+ qir_uniform_ui(c, 16));
+ struct qreg src1_lo = qir_AND(c, src[1 * 4 + i],
+ qir_uniform_ui(c, 0xffff));
+
+ struct qreg hilo = qir_MUL24(c, src0_hi, src1_lo);
+ struct qreg lohi = qir_MUL24(c, src0_lo, src1_hi);
+ struct qreg lolo = qir_MUL24(c, src0_lo, src1_lo);
+
+ return qir_ADD(c, lolo, qir_SHL(c,
+ qir_ADD(c, hilo, lohi),
+ qir_uniform_ui(c, 16)));
+}
+
+static struct qreg
+tgsi_to_qir_idiv(struct vc4_compile *c,
+ struct tgsi_full_instruction *tgsi_inst,
+ enum qop op, struct qreg *src, int i)
+{
+ return qir_FTOI(c, qir_FMUL(c,
+ qir_ITOF(c, src[0 * 4 + i]),
+ qir_RCP(c, qir_ITOF(c, src[1 * 4 + i]))));
+}
+
+static struct qreg
+tgsi_to_qir_ineg(struct vc4_compile *c,
+ struct tgsi_full_instruction *tgsi_inst,
+ enum qop op, struct qreg *src, int i)
+{
+ return qir_SUB(c, qir_uniform_ui(c, 0), src[0 * 4 + i]);
+}
+
+static struct qreg
+tgsi_to_qir_seq(struct vc4_compile *c,
+ struct tgsi_full_instruction *tgsi_inst,
+ enum qop op, struct qreg *src, int i)
+{
+ qir_SF(c, qir_FSUB(c, src[0 * 4 + i], src[1 * 4 + i]));
+ return qir_SEL_X_0_ZS(c, qir_uniform_f(c, 1.0));
+}
+
+static struct qreg
+tgsi_to_qir_sne(struct vc4_compile *c,
+ struct tgsi_full_instruction *tgsi_inst,
+ enum qop op, struct qreg *src, int i)
+{
+ qir_SF(c, qir_FSUB(c, src[0 * 4 + i], src[1 * 4 + i]));
+ return qir_SEL_X_0_ZC(c, qir_uniform_f(c, 1.0));
+}
+
+static struct qreg
+tgsi_to_qir_slt(struct vc4_compile *c,
+ struct tgsi_full_instruction *tgsi_inst,
+ enum qop op, struct qreg *src, int i)
+{
+ qir_SF(c, qir_FSUB(c, src[0 * 4 + i], src[1 * 4 + i]));
+ return qir_SEL_X_0_NS(c, qir_uniform_f(c, 1.0));
+}
+
+static struct qreg
+tgsi_to_qir_sge(struct vc4_compile *c,
+ struct tgsi_full_instruction *tgsi_inst,
+ enum qop op, struct qreg *src, int i)
+{
+ qir_SF(c, qir_FSUB(c, src[0 * 4 + i], src[1 * 4 + i]));
+ return qir_SEL_X_0_NC(c, qir_uniform_f(c, 1.0));
+}
+
+static struct qreg
+tgsi_to_qir_fseq(struct vc4_compile *c,
+ struct tgsi_full_instruction *tgsi_inst,
+ enum qop op, struct qreg *src, int i)
+{
+ qir_SF(c, qir_FSUB(c, src[0 * 4 + i], src[1 * 4 + i]));
+ return qir_SEL_X_0_ZS(c, qir_uniform_ui(c, ~0));
+}
+
+static struct qreg
+tgsi_to_qir_fsne(struct vc4_compile *c,
+ struct tgsi_full_instruction *tgsi_inst,
+ enum qop op, struct qreg *src, int i)
+{
+ qir_SF(c, qir_FSUB(c, src[0 * 4 + i], src[1 * 4 + i]));
+ return qir_SEL_X_0_ZC(c, qir_uniform_ui(c, ~0));
+}
+
+static struct qreg
+tgsi_to_qir_fslt(struct vc4_compile *c,
+ struct tgsi_full_instruction *tgsi_inst,
+ enum qop op, struct qreg *src, int i)
+{
+ qir_SF(c, qir_FSUB(c, src[0 * 4 + i], src[1 * 4 + i]));
+ return qir_SEL_X_0_NS(c, qir_uniform_ui(c, ~0));
+}
+
+static struct qreg
+tgsi_to_qir_fsge(struct vc4_compile *c,
+ struct tgsi_full_instruction *tgsi_inst,
+ enum qop op, struct qreg *src, int i)
+{
+ qir_SF(c, qir_FSUB(c, src[0 * 4 + i], src[1 * 4 + i]));
+ return qir_SEL_X_0_NC(c, qir_uniform_ui(c, ~0));
+}
+
+static struct qreg
+tgsi_to_qir_useq(struct vc4_compile *c,
+ struct tgsi_full_instruction *tgsi_inst,
+ enum qop op, struct qreg *src, int i)
+{
+ qir_SF(c, qir_SUB(c, src[0 * 4 + i], src[1 * 4 + i]));
+ return qir_SEL_X_0_ZS(c, qir_uniform_ui(c, ~0));
+}
+
+static struct qreg
+tgsi_to_qir_usne(struct vc4_compile *c,
+ struct tgsi_full_instruction *tgsi_inst,
+ enum qop op, struct qreg *src, int i)
+{
+ qir_SF(c, qir_SUB(c, src[0 * 4 + i], src[1 * 4 + i]));
+ return qir_SEL_X_0_ZC(c, qir_uniform_ui(c, ~0));
+}
+
+static struct qreg
+tgsi_to_qir_islt(struct vc4_compile *c,
+ struct tgsi_full_instruction *tgsi_inst,
+ enum qop op, struct qreg *src, int i)
+{
+ qir_SF(c, qir_SUB(c, src[0 * 4 + i], src[1 * 4 + i]));
+ return qir_SEL_X_0_NS(c, qir_uniform_ui(c, ~0));
+}
+
+static struct qreg
+tgsi_to_qir_isge(struct vc4_compile *c,
+ struct tgsi_full_instruction *tgsi_inst,
+ enum qop op, struct qreg *src, int i)
+{
+ qir_SF(c, qir_SUB(c, src[0 * 4 + i], src[1 * 4 + i]));
+ return qir_SEL_X_0_NC(c, qir_uniform_ui(c, ~0));
+}
+
+static struct qreg
+tgsi_to_qir_cmp(struct vc4_compile *c,
+ struct tgsi_full_instruction *tgsi_inst,
+ enum qop op, struct qreg *src, int i)
+{
+ qir_SF(c, src[0 * 4 + i]);
+ return qir_SEL_X_Y_NS(c,
+ src[1 * 4 + i],
+ src[2 * 4 + i]);
+}
+
+static struct qreg
+tgsi_to_qir_mad(struct vc4_compile *c,
struct tgsi_full_instruction *tgsi_inst,
enum qop op, struct qreg *src, int i)
{
- struct qcompile *c = trans->c;
return qir_FADD(c,
qir_FMUL(c,
src[0 * 4 + i],
}
static struct qreg
-tgsi_to_qir_lit(struct tgsi_to_qir *trans,
+tgsi_to_qir_lit(struct vc4_compile *c,
struct tgsi_full_instruction *tgsi_inst,
enum qop op, struct qreg *src, int i)
{
- struct qcompile *c = trans->c;
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(trans, 1.0);
+ return qir_uniform_f(c, 1.0);
case 1:
- return qir_FMAX(c, src[0 * 4 + 0], qir_uniform_f(trans, 0.0));
+ return qir_FMAX(c, src[0 * 4 + 0], qir_uniform_f(c, 0.0));
case 2: {
- struct qreg zero = qir_uniform_f(trans, 0.0);
+ struct qreg zero = qir_uniform_f(c, 0.0);
+ qir_SF(c, x);
/* XXX: Clamp w to -128..128 */
- return qir_CMP(c,
- x,
- zero,
- qir_EXP2(c, qir_FMUL(c,
- w,
- qir_LOG2(c,
- qir_FMAX(c,
- y,
- zero)))));
+ 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");
}
static struct qreg
-tgsi_to_qir_lrp(struct tgsi_to_qir *trans,
+tgsi_to_qir_lrp(struct vc4_compile *c,
struct tgsi_full_instruction *tgsi_inst,
enum qop op, struct qreg *src, int i)
{
- struct qcompile *c = trans->c;
struct qreg src0 = src[0 * 4 + i];
struct qreg src1 = src[1 * 4 + i];
struct qreg src2 = src[2 * 4 + i];
}
static void
-tgsi_to_qir_tex(struct tgsi_to_qir *trans,
+tgsi_to_qir_tex(struct vc4_compile *c,
struct tgsi_full_instruction *tgsi_inst,
enum qop op, struct qreg *src)
{
- struct qcompile *c = trans->c;
-
assert(!tgsi_inst->Instruction.Saturate);
struct qreg s = src[0 * 4 + 0];
struct qreg t = src[0 * 4 + 1];
+ uint32_t unit = tgsi_inst->Src[1].Register.Index;
+ struct qreg proj = c->undef;
if (tgsi_inst->Instruction.Opcode == TGSI_OPCODE_TXP) {
- struct qreg proj = qir_RCP(c, src[0 * 4 + 3]);
+ proj = qir_RCP(c, src[0 * 4 + 3]);
s = qir_FMUL(c, s, proj);
t = qir_FMUL(c, t, proj);
}
* we have to rescale from ([0, width], [0, height]) to ([0, 1], [0,
* 1]).
*/
- if (tgsi_inst->Texture.Texture == TGSI_TEXTURE_RECT) {
- uint32_t sampler = 0; /* XXX */
+ if (tgsi_inst->Texture.Texture == TGSI_TEXTURE_RECT ||
+ tgsi_inst->Texture.Texture == TGSI_TEXTURE_SHADOWRECT) {
s = qir_FMUL(c, s,
- get_temp_for_uniform(trans,
+ get_temp_for_uniform(c,
QUNIFORM_TEXRECT_SCALE_X,
- sampler));
+ unit));
t = qir_FMUL(c, t,
- get_temp_for_uniform(trans,
+ get_temp_for_uniform(c,
QUNIFORM_TEXRECT_SCALE_Y,
- sampler));
+ unit));
}
- uint32_t tex_and_sampler = 0; /* XXX */
- qir_TEX_T(c, t, add_uniform(trans, QUNIFORM_TEXTURE_CONFIG_P0,
- tex_and_sampler));
+ qir_TEX_T(c, t, add_uniform(c, QUNIFORM_TEXTURE_CONFIG_P0, unit));
- struct qreg sampler_p1 = add_uniform(trans, QUNIFORM_TEXTURE_CONFIG_P1,
- tex_and_sampler);
+ 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(trans, QUNIFORM_CONSTANT, 0));
+ qir_TEX_S(c, s, add_uniform(c, QUNIFORM_CONSTANT, 0));
} else {
qir_TEX_S(c, s, sampler_p1);
}
- trans->num_texture_samples++;
- qir_emit(c, qir_inst(QOP_TEX_RESULT, c->undef, c->undef, c->undef));
+ c->num_texture_samples++;
+ struct qreg r4 = qir_TEX_RESULT(c);
+
+ enum pipe_format format = c->key->tex[unit].format;
+
+ struct qreg unpacked[4];
+ if (util_format_is_depth_or_stencil(format)) {
+ struct qreg depthf = qir_ITOF(c, qir_SHR(c, r4,
+ qir_uniform_ui(c, 8)));
+ struct qreg normalized = qir_FMUL(c, depthf,
+ qir_uniform_f(c, 1.0f/0xffffff));
+
+ 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) {
+ switch (c->key->tex[unit].compare_func) {
+ case PIPE_FUNC_NEVER:
+ depth_output = qir_uniform_f(c, 0.0f);
+ break;
+ case PIPE_FUNC_ALWAYS:
+ depth_output = one;
+ break;
+ case PIPE_FUNC_EQUAL:
+ qir_SF(c, qir_FSUB(c, compare, normalized));
+ depth_output = qir_SEL_X_0_ZS(c, one);
+ break;
+ case PIPE_FUNC_NOTEQUAL:
+ qir_SF(c, qir_FSUB(c, compare, normalized));
+ depth_output = qir_SEL_X_0_ZC(c, one);
+ break;
+ case PIPE_FUNC_GREATER:
+ qir_SF(c, qir_FSUB(c, compare, normalized));
+ depth_output = qir_SEL_X_0_NC(c, one);
+ break;
+ case PIPE_FUNC_GEQUAL:
+ qir_SF(c, qir_FSUB(c, normalized, compare));
+ depth_output = qir_SEL_X_0_NS(c, one);
+ break;
+ case PIPE_FUNC_LESS:
+ qir_SF(c, qir_FSUB(c, compare, normalized));
+ depth_output = qir_SEL_X_0_NS(c, one);
+ break;
+ case PIPE_FUNC_LEQUAL:
+ qir_SF(c, qir_FSUB(c, normalized, compare));
+ depth_output = qir_SEL_X_0_NC(c, one);
+ break;
+ }
+ } else {
+ depth_output = normalized;
+ }
+ for (int i = 0; i < 4; i++)
+ unpacked[i] = depth_output;
+ } else {
+ for (int i = 0; i < 4; i++)
+ unpacked[i] = qir_R4_UNPACK(c, r4, i);
+ }
+
+ 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);
for (int i = 0; i < 4; i++) {
if (!(tgsi_inst->Dst[0].Register.WriteMask & (1 << i)))
continue;
- struct qreg dst = qir_get_temp(c);
- qir_emit(c, qir_inst(QOP_R4_UNPACK_A + i,
- dst,
- c->undef, c->undef));
- update_dst(trans, tgsi_inst, i, dst);
+ update_dst(c, tgsi_inst, i,
+ get_swizzled_channel(c, unpacked, swiz[i]));
}
}
static struct qreg
-tgsi_to_qir_pow(struct tgsi_to_qir *trans,
+tgsi_to_qir_pow(struct vc4_compile *c,
struct tgsi_full_instruction *tgsi_inst,
enum qop op, struct qreg *src, int i)
{
- struct qcompile *c = trans->c;
-
/* Note that this instruction replicates its result from the x channel
*/
return qir_EXP2(c, qir_FMUL(c,
}
static struct qreg
-tgsi_to_qir_trunc(struct tgsi_to_qir *trans,
+tgsi_to_qir_trunc(struct vc4_compile *c,
struct tgsi_full_instruction *tgsi_inst,
enum qop op, struct qreg *src, int i)
{
- struct qcompile *c = trans->c;
return qir_ITOF(c, qir_FTOI(c, src[0 * 4 + i]));
}
* to zero).
*/
static struct qreg
-tgsi_to_qir_frc(struct tgsi_to_qir *trans,
+tgsi_to_qir_frc(struct vc4_compile *c,
struct tgsi_full_instruction *tgsi_inst,
enum qop op, struct qreg *src, int i)
{
- struct qcompile *c = trans->c;
struct qreg trunc = qir_ITOF(c, qir_FTOI(c, src[0 * 4 + i]));
struct qreg diff = qir_FSUB(c, src[0 * 4 + i], trunc);
- return qir_CMP(c,
- diff,
- qir_FADD(c, diff, qir_uniform_f(trans, 1.0)),
- diff);
+ qir_SF(c, diff);
+ return qir_SEL_X_Y_NS(c,
+ qir_FADD(c, diff, qir_uniform_f(c, 1.0)),
+ diff);
}
/**
* zero).
*/
static struct qreg
-tgsi_to_qir_flr(struct tgsi_to_qir *trans,
+tgsi_to_qir_flr(struct vc4_compile *c,
struct tgsi_full_instruction *tgsi_inst,
enum qop op, struct qreg *src, int i)
{
- struct qcompile *c = trans->c;
struct qreg trunc = qir_ITOF(c, qir_FTOI(c, src[0 * 4 + i]));
- return qir_CMP(c,
- src[0 * 4 + i],
- qir_FSUB(c, trunc, qir_uniform_f(trans, 1.0)),
- trunc);
+
+ /* This will be < 0 if we truncated and the truncation was of a value
+ * that was < 0 in the first place.
+ */
+ qir_SF(c, qir_FSUB(c, src[0 * 4 + i], trunc));
+
+ return qir_SEL_X_Y_NS(c,
+ qir_FSUB(c, trunc, qir_uniform_f(c, 1.0)),
+ trunc);
}
static struct qreg
-tgsi_to_qir_dp(struct tgsi_to_qir *trans,
+tgsi_to_qir_dp(struct vc4_compile *c,
struct tgsi_full_instruction *tgsi_inst,
int num, struct qreg *src, int i)
{
- struct qcompile *c = trans->c;
-
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,
}
static struct qreg
-tgsi_to_qir_dp2(struct tgsi_to_qir *trans,
+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(trans, tgsi_inst, 2, src, i);
+ return tgsi_to_qir_dp(c, tgsi_inst, 2, src, i);
}
static struct qreg
-tgsi_to_qir_dp3(struct tgsi_to_qir *trans,
+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(trans, tgsi_inst, 3, src, i);
+ return tgsi_to_qir_dp(c, tgsi_inst, 3, src, i);
}
static struct qreg
-tgsi_to_qir_dp4(struct tgsi_to_qir *trans,
+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(trans, tgsi_inst, 4, src, i);
+ return tgsi_to_qir_dp(c, tgsi_inst, 4, src, i);
}
static struct qreg
-tgsi_to_qir_abs(struct tgsi_to_qir *trans,
+tgsi_to_qir_abs(struct vc4_compile *c,
struct tgsi_full_instruction *tgsi_inst,
enum qop op, struct qreg *src, int i)
{
- struct qcompile *c = trans->c;
struct qreg arg = src[0 * 4 + i];
return qir_FMAXABS(c, arg, arg);
}
/* Note that this instruction replicates its result from the x channel */
static struct qreg
-tgsi_to_qir_sin(struct tgsi_to_qir *trans,
+tgsi_to_qir_sin(struct vc4_compile *c,
struct tgsi_full_instruction *tgsi_inst,
enum qop op, struct qreg *src, int i)
{
- struct qcompile *c = trans->c;
float coeff[] = {
2.0 * M_PI,
-pow(2.0 * M_PI, 3) / (3 * 2 * 1),
struct qreg scaled_x =
qir_FMUL(c,
src[0 * 4 + 0],
- qir_uniform_f(trans, 1.0f / (M_PI * 2.0f)));
+ qir_uniform_f(c, 1.0f / (M_PI * 2.0f)));
- struct qreg x = tgsi_to_qir_frc(trans, NULL, 0, &scaled_x, 0);
+ struct qreg x = tgsi_to_qir_frc(c, NULL, 0, &scaled_x, 0);
struct qreg x2 = qir_FMUL(c, x, x);
- struct qreg sum = qir_FMUL(c, x, qir_uniform_f(trans, coeff[0]));
+ struct qreg sum = qir_FMUL(c, x, qir_uniform_f(c, coeff[0]));
for (int i = 1; i < ARRAY_SIZE(coeff); i++) {
x = qir_FMUL(c, x, x2);
sum = qir_FADD(c,
sum,
qir_FMUL(c,
x,
- qir_uniform_f(trans, coeff[i])));
+ qir_uniform_f(c, coeff[i])));
}
return sum;
}
/* Note that this instruction replicates its result from the x channel */
static struct qreg
-tgsi_to_qir_cos(struct tgsi_to_qir *trans,
+tgsi_to_qir_cos(struct vc4_compile *c,
struct tgsi_full_instruction *tgsi_inst,
enum qop op, struct qreg *src, int i)
{
- struct qcompile *c = trans->c;
float coeff[] = {
1.0f,
-pow(2.0 * M_PI, 2) / (2 * 1),
struct qreg scaled_x =
qir_FMUL(c, src[0 * 4 + 0],
- qir_uniform_f(trans, 1.0f / (M_PI * 2.0f)));
- struct qreg x_frac = tgsi_to_qir_frc(trans, NULL, 0, &scaled_x, 0);
+ qir_uniform_f(c, 1.0f / (M_PI * 2.0f)));
+ struct qreg x_frac = tgsi_to_qir_frc(c, NULL, 0, &scaled_x, 0);
- struct qreg sum = qir_uniform_f(trans, coeff[0]);
+ struct qreg sum = qir_uniform_f(c, coeff[0]);
struct qreg x2 = qir_FMUL(c, x_frac, x_frac);
struct qreg x = x2; /* Current x^2, x^4, or x^6 */
for (int i = 1; i < ARRAY_SIZE(coeff); i++) {
struct qreg mul = qir_FMUL(c,
x,
- qir_uniform_f(trans, coeff[i]));
+ qir_uniform_f(c, coeff[i]));
if (i == 0)
sum = mul;
else
}
static void
-emit_vertex_input(struct tgsi_to_qir *trans, int attr)
+emit_vertex_input(struct vc4_compile *c, int attr)
{
- enum pipe_format format = trans->vs_key->attr_formats[attr];
- struct qcompile *c = trans->c;
+ enum pipe_format format = c->vs_key->attr_formats[attr];
struct qreg vpm_reads[4];
/* Right now, we're setting the VPM offsets to be 16 bytes wide every
for (int i = 0; i < 4; i++) {
uint8_t swiz = desc->swizzle[i];
+ struct qreg result;
- switch (swiz) {
- case UTIL_FORMAT_SWIZZLE_NONE:
+ if (swiz > UTIL_FORMAT_SWIZZLE_W)
+ result = get_swizzled_channel(c, vpm_reads, swiz);
+ else if (desc->channel[swiz].size == 32 &&
+ desc->channel[swiz].type == UTIL_FORMAT_TYPE_FLOAT) {
+ result = get_swizzled_channel(c, vpm_reads, swiz);
+ } else if (desc->channel[swiz].size == 8 &&
+ (desc->channel[swiz].type == UTIL_FORMAT_TYPE_UNSIGNED ||
+ desc->channel[swiz].type == UTIL_FORMAT_TYPE_SIGNED) &&
+ desc->channel[swiz].normalized) {
+ struct qreg vpm = vpm_reads[0];
+ if (desc->channel[swiz].type == UTIL_FORMAT_TYPE_SIGNED)
+ vpm = qir_XOR(c, vpm, qir_uniform_ui(c, 0x80808080));
+ result = qir_UNPACK_8(c, vpm, swiz);
+ } else {
if (!format_warned) {
- fprintf(stderr,
- "vtx element %d NONE swizzle: %s\n",
- attr, util_format_name(format));
- format_warned = true;
- }
- /* FALLTHROUGH */
- case UTIL_FORMAT_SWIZZLE_0:
- trans->inputs[attr * 4 + i] = qir_uniform_ui(trans, 0);
- break;
- case UTIL_FORMAT_SWIZZLE_1:
- trans->inputs[attr * 4 + i] = qir_uniform_ui(trans,
- fui(1.0));
- break;
- default:
- if (!format_warned &&
- (desc->channel[swiz].type != UTIL_FORMAT_TYPE_FLOAT ||
- desc->channel[swiz].size != 32)) {
fprintf(stderr,
"vtx element %d unsupported type: %s\n",
attr, util_format_name(format));
format_warned = true;
}
+ result = qir_uniform_f(c, 0.0);
+ }
- trans->inputs[attr * 4 + i] = vpm_reads[swiz];
- break;
+ if (desc->channel[swiz].normalized &&
+ desc->channel[swiz].type == UTIL_FORMAT_TYPE_SIGNED) {
+ result = qir_FSUB(c,
+ qir_FMUL(c,
+ result,
+ qir_uniform_f(c, 2.0)),
+ qir_uniform_f(c, 1.0));
}
+
+ c->inputs[attr * 4 + i] = result;
}
}
static void
-emit_fragcoord_input(struct tgsi_to_qir *trans, int attr)
+tgsi_to_qir_kill_if(struct vc4_compile *c, struct qreg *src, int i)
{
- struct qcompile *c = trans->c;
+ if (c->discard.file == QFILE_NULL)
+ c->discard = qir_uniform_f(c, 0.0);
+ qir_SF(c, src[0 * 4 + i]);
+ c->discard = qir_SEL_X_Y_NS(c, qir_uniform_f(c, 1.0),
+ c->discard);
+}
- trans->inputs[attr * 4 + 0] = qir_FRAG_X(c);
- trans->inputs[attr * 4 + 1] = qir_FRAG_Y(c);
- trans->inputs[attr * 4 + 2] =
+static void
+emit_fragcoord_input(struct vc4_compile *c, int attr)
+{
+ c->inputs[attr * 4 + 0] = qir_FRAG_X(c);
+ c->inputs[attr * 4 + 1] = qir_FRAG_Y(c);
+ c->inputs[attr * 4 + 2] =
qir_FMUL(c,
- qir_FRAG_Z(c),
- qir_uniform_f(trans, 1.0 / 0xffffff));
- trans->inputs[attr * 4 + 3] = qir_FRAG_RCP_W(c);
+ qir_ITOF(c, qir_FRAG_Z(c)),
+ qir_uniform_f(c, 1.0 / 0xffffff));
+ c->inputs[attr * 4 + 3] = qir_RCP(c, qir_FRAG_W(c));
}
-static struct qreg
-emit_fragment_varying(struct tgsi_to_qir *trans, int index)
+static void
+emit_point_coord_input(struct vc4_compile *c, int attr)
{
- struct qcompile *c = trans->c;
+ 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)
+{
struct qreg vary = {
QFILE_VARY,
index
};
- /* XXX: multiply by W */
- return qir_VARY_ADD_C(c, qir_MOV(c, vary));
+ return qir_VARY_ADD_C(c, qir_FMUL(c, vary, qir_FRAG_W(c)));
}
static void
-emit_fragment_input(struct tgsi_to_qir *trans, int attr)
+emit_fragment_input(struct vc4_compile *c, int attr,
+ struct tgsi_full_declaration *decl)
{
- struct qcompile *c = trans->c;
-
for (int i = 0; i < 4; i++) {
- trans->inputs[attr * 4 + i] =
- emit_fragment_varying(trans, attr * 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_tgsi_declaration(struct tgsi_to_qir *trans,
+emit_tgsi_declaration(struct vc4_compile *c,
struct tgsi_full_declaration *decl)
{
- struct qcompile *c = trans->c;
-
switch (decl->Declaration.File) {
+ case TGSI_FILE_TEMPORARY:
+ resize_qreg_array(c, &c->temps, &c->temps_array_size,
+ (decl->Range.Last + 1) * 4);
+ break;
+
case TGSI_FILE_INPUT:
+ resize_qreg_array(c, &c->inputs, &c->inputs_array_size,
+ (decl->Range.Last + 1) * 4);
+
for (int i = decl->Range.First;
i <= decl->Range.Last;
i++) {
if (c->stage == QSTAGE_FRAG) {
if (decl->Semantic.Name ==
TGSI_SEMANTIC_POSITION) {
- emit_fragcoord_input(trans, i);
+ emit_fragcoord_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(trans, i);
+ emit_fragment_input(c, i, decl);
}
} else {
- emit_vertex_input(trans, i);
+ emit_vertex_input(c, i);
}
}
break;
+
+ case TGSI_FILE_OUTPUT:
+ resize_qreg_array(c, &c->outputs, &c->outputs_array_size,
+ (decl->Range.Last + 1) * 4);
+
+ switch (decl->Semantic.Name) {
+ case TGSI_SEMANTIC_POSITION:
+ c->output_position_index = decl->Range.First * 4;
+ break;
+ 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;
}
}
static void
-emit_tgsi_instruction(struct tgsi_to_qir *trans,
+emit_tgsi_instruction(struct vc4_compile *c,
struct tgsi_full_instruction *tgsi_inst)
{
- struct qcompile *c = trans->c;
struct {
enum qop op;
- struct qreg (*func)(struct tgsi_to_qir *trans,
+ struct qreg (*func)(struct vc4_compile *c,
struct tgsi_full_instruction *tgsi_inst,
enum qop op,
struct qreg *src, int i);
[TGSI_OPCODE_SUB] = { QOP_FSUB, tgsi_to_qir_alu },
[TGSI_OPCODE_MIN] = { QOP_FMIN, tgsi_to_qir_alu },
[TGSI_OPCODE_MAX] = { QOP_FMAX, tgsi_to_qir_alu },
+ [TGSI_OPCODE_F2I] = { QOP_FTOI, tgsi_to_qir_alu },
+ [TGSI_OPCODE_I2F] = { QOP_ITOF, tgsi_to_qir_alu },
+ [TGSI_OPCODE_UADD] = { QOP_ADD, tgsi_to_qir_alu },
+ [TGSI_OPCODE_USHR] = { QOP_SHR, tgsi_to_qir_alu },
+ [TGSI_OPCODE_ISHR] = { QOP_ASR, tgsi_to_qir_alu },
+ [TGSI_OPCODE_SHL] = { QOP_SHL, tgsi_to_qir_alu },
+ [TGSI_OPCODE_IMIN] = { QOP_MIN, tgsi_to_qir_alu },
+ [TGSI_OPCODE_IMAX] = { QOP_MAX, tgsi_to_qir_alu },
+ [TGSI_OPCODE_AND] = { QOP_AND, tgsi_to_qir_alu },
+ [TGSI_OPCODE_OR] = { QOP_OR, tgsi_to_qir_alu },
+ [TGSI_OPCODE_XOR] = { QOP_XOR, tgsi_to_qir_alu },
+ [TGSI_OPCODE_NOT] = { QOP_NOT, tgsi_to_qir_alu },
+
+ [TGSI_OPCODE_UMUL] = { 0, tgsi_to_qir_umul },
+ [TGSI_OPCODE_IDIV] = { 0, tgsi_to_qir_idiv },
+ [TGSI_OPCODE_INEG] = { 0, tgsi_to_qir_ineg },
+
[TGSI_OPCODE_RSQ] = { QOP_RSQ, tgsi_to_qir_alu },
- [TGSI_OPCODE_SEQ] = { QOP_SEQ, tgsi_to_qir_alu },
- [TGSI_OPCODE_SNE] = { QOP_SNE, tgsi_to_qir_alu },
- [TGSI_OPCODE_SGE] = { QOP_SGE, tgsi_to_qir_alu },
- [TGSI_OPCODE_SLT] = { QOP_SLT, tgsi_to_qir_alu },
- [TGSI_OPCODE_CMP] = { QOP_CMP, tgsi_to_qir_alu },
+ [TGSI_OPCODE_SEQ] = { 0, tgsi_to_qir_seq },
+ [TGSI_OPCODE_SNE] = { 0, tgsi_to_qir_sne },
+ [TGSI_OPCODE_SGE] = { 0, tgsi_to_qir_sge },
+ [TGSI_OPCODE_SLT] = { 0, tgsi_to_qir_slt },
+ [TGSI_OPCODE_FSEQ] = { 0, tgsi_to_qir_fseq },
+ [TGSI_OPCODE_FSNE] = { 0, tgsi_to_qir_fsne },
+ [TGSI_OPCODE_FSGE] = { 0, tgsi_to_qir_fsge },
+ [TGSI_OPCODE_FSLT] = { 0, tgsi_to_qir_fslt },
+ [TGSI_OPCODE_USEQ] = { 0, tgsi_to_qir_useq },
+ [TGSI_OPCODE_USNE] = { 0, tgsi_to_qir_usne },
+ [TGSI_OPCODE_ISGE] = { 0, tgsi_to_qir_isge },
+ [TGSI_OPCODE_ISLT] = { 0, tgsi_to_qir_islt },
+
+ [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 },
for (int s = 0; s < 3; s++) {
for (int i = 0; i < 4; i++) {
src_regs[4 * s + i] =
- get_src(trans, &tgsi_inst->Src[s].Register, i);
+ get_src(c, tgsi_inst->Instruction.Opcode,
+ &tgsi_inst->Src[s].Register, i);
}
}
case TGSI_OPCODE_TEX:
case TGSI_OPCODE_TXP:
case TGSI_OPCODE_TXB:
- tgsi_to_qir_tex(trans, tgsi_inst,
+ tgsi_to_qir_tex(c, tgsi_inst,
op_trans[tgsi_op].op, src_regs);
return;
+ case TGSI_OPCODE_KILL:
+ c->discard = qir_uniform_f(c, 1.0);
+ return;
+ case TGSI_OPCODE_KILL_IF:
+ for (int i = 0; i < 4; i++)
+ tgsi_to_qir_kill_if(c, src_regs, i);
+ return;
default:
break;
}
struct qreg result;
- result = op_trans[tgsi_op].func(trans, tgsi_inst,
+ result = op_trans[tgsi_op].func(c, tgsi_inst,
op_trans[tgsi_op].op,
src_regs, i);
result = qir_FMAX(c,
qir_FMIN(c,
result,
- qir_uniform_f(trans, 1.0)),
- qir_uniform_f(trans, low));
+ qir_uniform_f(c, 1.0)),
+ qir_uniform_f(c, low));
}
- update_dst(trans, tgsi_inst, i, result);
+ update_dst(c, tgsi_inst, i, result);
}
}
static void
-parse_tgsi_immediate(struct tgsi_to_qir *trans, struct tgsi_full_immediate *imm)
+parse_tgsi_immediate(struct vc4_compile *c, struct tgsi_full_immediate *imm)
{
for (int i = 0; i < 4; i++) {
- unsigned n = trans->num_consts++;
- trans->consts[n] = qir_uniform_ui(trans, imm->u[i].Uint);
+ unsigned n = c->num_consts++;
+ resize_qreg_array(c, &c->consts, &c->consts_array_size, n + 1);
+ c->consts[n] = qir_uniform_ui(c, imm->u[i].Uint);
+ }
+}
+
+static struct qreg
+vc4_blend_channel(struct vc4_compile *c,
+ struct qreg *dst,
+ struct qreg *src,
+ struct qreg val,
+ unsigned factor,
+ int channel)
+{
+ switch(factor) {
+ case PIPE_BLENDFACTOR_ONE:
+ return val;
+ case PIPE_BLENDFACTOR_SRC_COLOR:
+ return qir_FMUL(c, val, src[channel]);
+ case PIPE_BLENDFACTOR_SRC_ALPHA:
+ return qir_FMUL(c, val, src[3]);
+ case PIPE_BLENDFACTOR_DST_ALPHA:
+ return qir_FMUL(c, val, dst[3]);
+ case PIPE_BLENDFACTOR_DST_COLOR:
+ return qir_FMUL(c, val, dst[channel]);
+ case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
+ return qir_FMIN(c, src[3], qir_FSUB(c,
+ qir_uniform_f(c, 1.0),
+ dst[3]));
+ case PIPE_BLENDFACTOR_CONST_COLOR:
+ return qir_FMUL(c, val,
+ get_temp_for_uniform(c,
+ QUNIFORM_BLEND_CONST_COLOR,
+ channel));
+ case PIPE_BLENDFACTOR_CONST_ALPHA:
+ return qir_FMUL(c, val,
+ get_temp_for_uniform(c,
+ QUNIFORM_BLEND_CONST_COLOR,
+ 3));
+ case PIPE_BLENDFACTOR_ZERO:
+ return qir_uniform_f(c, 0.0);
+ case PIPE_BLENDFACTOR_INV_SRC_COLOR:
+ return qir_FMUL(c, val, qir_FSUB(c, qir_uniform_f(c, 1.0),
+ src[channel]));
+ case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
+ return qir_FMUL(c, val, qir_FSUB(c, qir_uniform_f(c, 1.0),
+ src[3]));
+ case PIPE_BLENDFACTOR_INV_DST_ALPHA:
+ return qir_FMUL(c, val, qir_FSUB(c, qir_uniform_f(c, 1.0),
+ dst[3]));
+ case PIPE_BLENDFACTOR_INV_DST_COLOR:
+ return qir_FMUL(c, val, qir_FSUB(c, qir_uniform_f(c, 1.0),
+ dst[channel]));
+ case PIPE_BLENDFACTOR_INV_CONST_COLOR:
+ return qir_FMUL(c, val,
+ qir_FSUB(c, qir_uniform_f(c, 1.0),
+ get_temp_for_uniform(c,
+ QUNIFORM_BLEND_CONST_COLOR,
+ channel)));
+ case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
+ return qir_FMUL(c, val,
+ qir_FSUB(c, qir_uniform_f(c, 1.0),
+ get_temp_for_uniform(c,
+ QUNIFORM_BLEND_CONST_COLOR,
+ 3)));
+
+ default:
+ case PIPE_BLENDFACTOR_SRC1_COLOR:
+ case PIPE_BLENDFACTOR_SRC1_ALPHA:
+ case PIPE_BLENDFACTOR_INV_SRC1_COLOR:
+ case PIPE_BLENDFACTOR_INV_SRC1_ALPHA:
+ /* Unsupported. */
+ fprintf(stderr, "Unknown blend factor %d\n", factor);
+ return val;
+ }
+}
+
+static struct qreg
+vc4_blend_func(struct vc4_compile *c,
+ struct qreg src, struct qreg dst,
+ unsigned func)
+{
+ switch (func) {
+ case PIPE_BLEND_ADD:
+ return qir_FADD(c, src, dst);
+ case PIPE_BLEND_SUBTRACT:
+ return qir_FSUB(c, src, dst);
+ case PIPE_BLEND_REVERSE_SUBTRACT:
+ return qir_FSUB(c, dst, src);
+ case PIPE_BLEND_MIN:
+ return qir_FMIN(c, src, dst);
+ case PIPE_BLEND_MAX:
+ return qir_FMAX(c, src, dst);
+
+ default:
+ /* Unsupported. */
+ fprintf(stderr, "Unknown blend func %d\n", func);
+ return src;
+
+ }
+}
+
+/**
+ * Implements fixed function blending in shader code.
+ *
+ * VC4 doesn't have any hardware support for blending. Instead, you read the
+ * current contents of the destination from the tile buffer after having
+ * waited for the scoreboard (which is handled by vc4_qpu_emit.c), then do
+ * math using your output color and that destination value, and update the
+ * output color appropriately.
+ */
+static void
+vc4_blend(struct vc4_compile *c, struct qreg *result,
+ struct qreg *dst_color, struct qreg *src_color)
+{
+ struct pipe_rt_blend_state *blend = &c->fs_key->blend;
+
+ if (!blend->blend_enable) {
+ for (int i = 0; i < 4; i++)
+ result[i] = src_color[i];
+ return;
+ }
+
+ struct qreg src_blend[4], dst_blend[4];
+ for (int i = 0; i < 3; i++) {
+ src_blend[i] = vc4_blend_channel(c,
+ dst_color, src_color,
+ src_color[i],
+ blend->rgb_src_factor, i);
+ dst_blend[i] = vc4_blend_channel(c,
+ dst_color, src_color,
+ dst_color[i],
+ blend->rgb_dst_factor, i);
+ }
+ src_blend[3] = vc4_blend_channel(c,
+ dst_color, src_color,
+ src_color[3],
+ blend->alpha_src_factor, 3);
+ dst_blend[3] = vc4_blend_channel(c,
+ dst_color, src_color,
+ dst_color[3],
+ blend->alpha_dst_factor, 3);
+
+ for (int i = 0; i < 3; i++) {
+ result[i] = vc4_blend_func(c,
+ src_blend[i], dst_blend[i],
+ blend->rgb_func);
+ }
+ result[3] = vc4_blend_func(c,
+ src_blend[3], dst_blend[3],
+ 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 tgsi_to_qir *trans)
+emit_frag_end(struct vc4_compile *c)
{
- struct qcompile *c = trans->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 };
+ 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++)
+ dst_color[i] = get_swizzled_channel(c,
+ tlb_read_color,
+ format_swiz[i]);
+ }
- struct qreg t = qir_get_temp(c);
+ 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,
+ (c->output_color_index != -1 ?
+ c->outputs + c->output_color_index :
+ undef_array));
- const struct util_format_description *format_desc =
- util_format_description(trans->fs_key->color_format);
+ /* If the bit isn't set in the color mask, then just return the
+ * original dst color, instead.
+ */
+ for (int i = 0; i < 4; i++) {
+ if (!(c->fs_key->blend.colormask & (1 << i))) {
+ blend_color[i] = dst_color[i];
+ }
+ }
/* Debug: Sometimes you're getting a black output and just want to see
* if the FS is getting executed at all. Spam magenta into the color
* output.
*/
if (0) {
- trans->outputs[format_desc->swizzle[0]] =
- qir_uniform_ui(trans, fui(1.0));
- trans->outputs[format_desc->swizzle[1]] =
- qir_uniform_ui(trans, fui(0.0));
- trans->outputs[format_desc->swizzle[2]] =
- qir_uniform_ui(trans, fui(1.0));
- trans->outputs[format_desc->swizzle[3]] =
- qir_uniform_ui(trans, fui(0.5));
+ blend_color[0] = qir_uniform_f(c, 1.0);
+ blend_color[1] = qir_uniform_f(c, 0.0);
+ blend_color[2] = qir_uniform_f(c, 1.0);
+ blend_color[3] = qir_uniform_f(c, 0.5);
}
- struct qreg swizzled_outputs[4] = {
- trans->outputs[format_desc->swizzle[0]],
- trans->outputs[format_desc->swizzle[1]],
- trans->outputs[format_desc->swizzle[2]],
- trans->outputs[format_desc->swizzle[3]],
- };
+ struct qreg swizzled_outputs[4];
+ for (int i = 0; i < 4; i++) {
+ swizzled_outputs[i] = get_swizzled_channel(c, blend_color,
+ format_swiz[i]);
+ }
+
+ if (c->discard.file != QFILE_NULL)
+ qir_TLB_DISCARD_SETUP(c, c->discard);
+
+ if (c->fs_key->stencil_enabled) {
+ qir_TLB_STENCIL_SETUP(c, add_uniform(c, QUNIFORM_STENCIL, 0));
+ if (c->fs_key->stencil_twoside) {
+ qir_TLB_STENCIL_SETUP(c, add_uniform(c, QUNIFORM_STENCIL, 1));
+ }
+ if (c->fs_key->stencil_full_writemasks) {
+ qir_TLB_STENCIL_SETUP(c, add_uniform(c, QUNIFORM_STENCIL, 2));
+ }
+ }
- if (trans->fs_key->depth_enabled) {
- qir_emit(c, qir_inst(QOP_TLB_PASSTHROUGH_Z_WRITE, c->undef,
- c->undef, c->undef));
+ if (c->fs_key->depth_enabled) {
+ struct qreg z;
+ if (c->output_position_index != -1) {
+ z = qir_FTOI(c, qir_FMUL(c, c->outputs[c->output_position_index + 2],
+ qir_uniform_f(c, 0xffffff)));
+ } else {
+ z = qir_FRAG_Z(c);
+ }
+ qir_TLB_Z_WRITE(c, z);
+ }
+
+ bool color_written = false;
+ for (int i = 0; i < 4; i++) {
+ if (swizzled_outputs[i].file != QFILE_NULL)
+ color_written = true;
+ }
+
+ struct qreg packed_color;
+ if (color_written) {
+ /* Fill in any undefined colors. The simulator will assertion
+ * fail if we read something that wasn't written, and I don't
+ * know what hardware does.
+ */
+ for (int i = 0; i < 4; i++) {
+ if (swizzled_outputs[i].file == QFILE_NULL)
+ swizzled_outputs[i] = qir_uniform_f(c, 0.0);
+ }
+ packed_color = qir_get_temp(c);
+ qir_emit(c, qir_inst4(QOP_PACK_COLORS, packed_color,
+ swizzled_outputs[0],
+ swizzled_outputs[1],
+ swizzled_outputs[2],
+ swizzled_outputs[3]));
+ } else {
+ packed_color = qir_uniform_ui(c, 0);
}
- qir_emit(c, qir_inst4(QOP_PACK_COLORS, t,
- swizzled_outputs[0],
- swizzled_outputs[1],
- swizzled_outputs[2],
- swizzled_outputs[3]));
qir_emit(c, qir_inst(QOP_TLB_COLOR_WRITE, c->undef,
- t, c->undef));
+ packed_color, c->undef));
}
static void
-emit_scaled_viewport_write(struct tgsi_to_qir *trans, struct qreg rcp_w)
+emit_scaled_viewport_write(struct vc4_compile *c, struct qreg rcp_w)
{
- struct qcompile *c = trans->c;
struct qreg xyi[2];
for (int i = 0; i < 2; i++) {
struct qreg scale =
- add_uniform(trans, QUNIFORM_VIEWPORT_X_SCALE + i, 0);
+ add_uniform(c, QUNIFORM_VIEWPORT_X_SCALE + i, 0);
xyi[i] = qir_FTOI(c, qir_FMUL(c,
qir_FMUL(c,
- trans->outputs[i],
+ c->outputs[i],
scale),
rcp_w));
}
}
static void
-emit_zs_write(struct tgsi_to_qir *trans, struct qreg rcp_w)
+emit_zs_write(struct vc4_compile *c, struct qreg rcp_w)
{
- struct qcompile *c = trans->c;
-
- struct qreg zscale = add_uniform(trans, QUNIFORM_VIEWPORT_Z_SCALE, 0);
- struct qreg zoffset = add_uniform(trans, QUNIFORM_VIEWPORT_Z_OFFSET, 0);
+ struct qreg zscale = add_uniform(c, QUNIFORM_VIEWPORT_Z_SCALE, 0);
+ struct qreg zoffset = add_uniform(c, QUNIFORM_VIEWPORT_Z_OFFSET, 0);
qir_VPM_WRITE(c, qir_FMUL(c, qir_FADD(c, qir_FMUL(c,
- trans->outputs[2],
+ c->outputs[2],
zscale),
zoffset),
rcp_w));
}
static void
-emit_rcp_wc_write(struct tgsi_to_qir *trans, struct qreg rcp_w)
+emit_rcp_wc_write(struct vc4_compile *c, struct qreg rcp_w)
{
- struct qcompile *c = trans->c;
-
qir_VPM_WRITE(c, rcp_w);
}
static void
-emit_vert_end(struct tgsi_to_qir *trans)
+emit_point_size_write(struct vc4_compile *c)
{
- struct qcompile *c = trans->c;
+ struct qreg point_size;
- struct qreg rcp_w = qir_RCP(c, trans->outputs[3]);
+ 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);
- emit_scaled_viewport_write(trans, rcp_w);
- emit_zs_write(trans, rcp_w);
- emit_rcp_wc_write(trans, rcp_w);
+ /* Workaround: HW-2726 PTB does not handle zero-size points (BCM2835,
+ * BCM21553).
+ */
+ point_size = qir_FMAX(c, point_size, qir_uniform_f(c, .125));
- for (int i = 4; i < trans->num_outputs; i++) {
- qir_VPM_WRITE(c, trans->outputs[i]);
- }
+ qir_VPM_WRITE(c, point_size);
}
static void
-emit_coord_end(struct tgsi_to_qir *trans)
+emit_vert_end(struct vc4_compile *c)
{
- struct qcompile *c = trans->c;
+ struct qreg rcp_w = qir_RCP(c, c->outputs[3]);
+
+ 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]);
+ }
+}
- struct qreg rcp_w = qir_RCP(c, trans->outputs[3]);
+static void
+emit_coord_end(struct vc4_compile *c)
+{
+ struct qreg rcp_w = qir_RCP(c, c->outputs[3]);
for (int i = 0; i < 4; i++)
- qir_VPM_WRITE(c, trans->outputs[i]);
+ qir_VPM_WRITE(c, c->outputs[i]);
- emit_scaled_viewport_write(trans, rcp_w);
- emit_zs_write(trans, rcp_w);
- emit_rcp_wc_write(trans, rcp_w);
+ 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 tgsi_to_qir *
-vc4_shader_tgsi_to_qir(struct vc4_compiled_shader *shader, enum qstage stage,
+static struct vc4_compile *
+vc4_shader_tgsi_to_qir(struct vc4_context *vc4,
+ struct vc4_compiled_shader *shader, enum qstage stage,
struct vc4_key *key)
{
- struct tgsi_to_qir *trans = CALLOC_STRUCT(tgsi_to_qir);
- struct qcompile *c;
+ struct vc4_compile *c = qir_compile_init();
int ret;
- c = qir_compile_init();
c->stage = stage;
- memset(trans, 0, sizeof(*trans));
- /* XXX sizing */
- trans->temps = calloc(sizeof(struct qreg), 1024);
- trans->inputs = calloc(sizeof(struct qreg), 8 * 4);
- trans->outputs = calloc(sizeof(struct qreg), 1024);
- trans->uniforms = calloc(sizeof(struct qreg), 1024);
- trans->consts = calloc(sizeof(struct qreg), 1024);
+ c->uniform_data = ralloc_array(c, uint32_t, 1024);
+ c->uniform_contents = ralloc_array(c, enum quniform_contents, 1024);
- trans->uniform_data = calloc(sizeof(uint32_t), 1024);
- trans->uniform_contents = calloc(sizeof(enum quniform_contents), 1024);
-
- trans->shader_state = key->shader_state;
- trans->c = c;
- ret = tgsi_parse_init(&trans->parser, trans->shader_state->tokens);
+ 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(trans->shader_state->tokens, 0);
+ tgsi_dump(c->shader_state->tokens, 0);
}
+ c->key = key;
switch (stage) {
case QSTAGE_FRAG:
- trans->fs_key = (struct vc4_fs_key *)key;
- if (trans->fs_key->is_points) {
- trans->point_x = emit_fragment_varying(trans, 0);
- trans->point_y = emit_fragment_varying(trans, 0);
- } else if (trans->fs_key->is_lines) {
- trans->line_x = emit_fragment_varying(trans, 0);
+ c->fs_key = (struct vc4_fs_key *)key;
+ if (c->fs_key->is_points) {
+ c->point_x = emit_fragment_varying(c, 0);
+ c->point_y = emit_fragment_varying(c, 0);
+ } else if (c->fs_key->is_lines) {
+ c->line_x = emit_fragment_varying(c, 0);
}
break;
case QSTAGE_VERT:
- trans->vs_key = (struct vc4_vs_key *)key;
+ c->vs_key = (struct vc4_vs_key *)key;
break;
case QSTAGE_COORD:
- trans->vs_key = (struct vc4_vs_key *)key;
+ c->vs_key = (struct vc4_vs_key *)key;
break;
}
- while (!tgsi_parse_end_of_tokens(&trans->parser)) {
- tgsi_parse_token(&trans->parser);
+ while (!tgsi_parse_end_of_tokens(&c->parser)) {
+ tgsi_parse_token(&c->parser);
- switch (trans->parser.FullToken.Token.Type) {
+ switch (c->parser.FullToken.Token.Type) {
case TGSI_TOKEN_TYPE_DECLARATION:
- emit_tgsi_declaration(trans,
- &trans->parser.FullToken.FullDeclaration);
+ emit_tgsi_declaration(c,
+ &c->parser.FullToken.FullDeclaration);
break;
case TGSI_TOKEN_TYPE_INSTRUCTION:
- emit_tgsi_instruction(trans,
- &trans->parser.FullToken.FullInstruction);
+ emit_tgsi_instruction(c,
+ &c->parser.FullToken.FullInstruction);
break;
case TGSI_TOKEN_TYPE_IMMEDIATE:
- parse_tgsi_immediate(trans,
- &trans->parser.FullToken.FullImmediate);
+ parse_tgsi_immediate(c,
+ &c->parser.FullToken.FullImmediate);
break;
}
}
switch (stage) {
case QSTAGE_FRAG:
- emit_frag_end(trans);
+ emit_frag_end(c);
break;
case QSTAGE_VERT:
- emit_vert_end(trans);
+ emit_vert_end(c);
break;
case QSTAGE_COORD:
- emit_coord_end(trans);
+ emit_coord_end(c);
break;
}
- tgsi_parse_free(&trans->parser);
- free(trans->temps);
+ tgsi_parse_free(&c->parser);
qir_optimize(c);
fprintf(stderr, "QIR:\n");
qir_dump(c);
}
- vc4_generate_code(c);
+ qir_reorder_uniforms(c);
+ vc4_generate_code(vc4, c);
if (vc4_debug & VC4_DEBUG_SHADERDB) {
fprintf(stderr, "SHADER-DB: %s: %d instructions\n",
qir_get_stage_name(c->stage), c->qpu_inst_count);
fprintf(stderr, "SHADER-DB: %s: %d uniforms\n",
- qir_get_stage_name(c->stage), trans->num_uniforms);
+ qir_get_stage_name(c->stage), c->num_uniforms);
}
- return trans;
+ return c;
}
static void *
static void
copy_uniform_state_to_shader(struct vc4_compiled_shader *shader,
int shader_index,
- struct tgsi_to_qir *trans)
+ struct vc4_compile *c)
{
- int count = trans->num_uniforms;
+ int count = c->num_uniforms;
struct vc4_shader_uniform_info *uinfo = &shader->uniforms[shader_index];
uinfo->count = count;
uinfo->data = malloc(count * sizeof(*uinfo->data));
- memcpy(uinfo->data, trans->uniform_data,
+ memcpy(uinfo->data, c->uniform_data,
count * sizeof(*uinfo->data));
uinfo->contents = malloc(count * sizeof(*uinfo->contents));
- memcpy(uinfo->contents, trans->uniform_contents,
+ memcpy(uinfo->contents, c->uniform_contents,
count * sizeof(*uinfo->contents));
- uinfo->num_texture_samples = trans->num_texture_samples;
+ uinfo->num_texture_samples = c->num_texture_samples;
}
static void
vc4_fs_compile(struct vc4_context *vc4, struct vc4_compiled_shader *shader,
struct vc4_fs_key *key)
{
- struct tgsi_to_qir *trans = vc4_shader_tgsi_to_qir(shader, QSTAGE_FRAG,
- &key->base);
- shader->num_inputs = trans->c->num_inputs;
- copy_uniform_state_to_shader(shader, 0, trans);
- shader->bo = vc4_bo_alloc_mem(vc4->screen, trans->c->qpu_insts,
- trans->c->qpu_inst_count * sizeof(uint64_t),
+ struct vc4_compile *c = vc4_shader_tgsi_to_qir(vc4, shader,
+ 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),
"fs_code");
- qir_compile_destroy(trans->c);
- free(trans);
+ qir_compile_destroy(c);
}
static void
vc4_vs_compile(struct vc4_context *vc4, struct vc4_compiled_shader *shader,
struct vc4_vs_key *key)
{
- struct tgsi_to_qir *vs_trans = vc4_shader_tgsi_to_qir(shader,
- QSTAGE_VERT,
- &key->base);
- copy_uniform_state_to_shader(shader, 0, vs_trans);
-
- struct tgsi_to_qir *cs_trans = vc4_shader_tgsi_to_qir(shader,
- QSTAGE_COORD,
- &key->base);
- copy_uniform_state_to_shader(shader, 1, cs_trans);
-
- uint32_t vs_size = vs_trans->c->qpu_inst_count * sizeof(uint64_t);
- uint32_t cs_size = cs_trans->c->qpu_inst_count * sizeof(uint64_t);
+ struct vc4_compile *vs_c = vc4_shader_tgsi_to_qir(vc4, shader,
+ QSTAGE_VERT,
+ &key->base);
+ copy_uniform_state_to_shader(shader, 0, vs_c);
+
+ struct vc4_compile *cs_c = vc4_shader_tgsi_to_qir(vc4, shader,
+ QSTAGE_COORD,
+ &key->base);
+ copy_uniform_state_to_shader(shader, 1, cs_c);
+
+ uint32_t vs_size = vs_c->qpu_inst_count * sizeof(uint64_t);
+ uint32_t cs_size = cs_c->qpu_inst_count * sizeof(uint64_t);
shader->coord_shader_offset = vs_size; /* XXX: alignment? */
shader->bo = vc4_bo_alloc(vc4->screen,
shader->coord_shader_offset + cs_size,
"vs_code");
void *map = vc4_bo_map(shader->bo);
- memcpy(map, vs_trans->c->qpu_insts, vs_size);
+ memcpy(map, vs_c->qpu_insts, vs_size);
memcpy(map + shader->coord_shader_offset,
- cs_trans->c->qpu_insts, cs_size);
+ cs_c->qpu_insts, cs_size);
- qir_compile_destroy(vs_trans->c);
- qir_compile_destroy(cs_trans->c);
+ qir_compile_destroy(vs_c);
+ qir_compile_destroy(cs_c);
+}
+
+static void
+vc4_setup_shared_key(struct vc4_key *key, struct vc4_texture_stateobj *texstate)
+{
+ 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];
+
+ if (sampler) {
+ struct pipe_resource *prsc = sampler->texture;
+ key->tex[i].format = prsc->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;
+ }
+ }
}
static void
struct vc4_fs_key *key = &local_key;
memset(key, 0, sizeof(*key));
+ vc4_setup_shared_key(&key->base, &vc4->fragtex);
key->base.shader_state = vc4->prog.bind_fs;
key->is_points = (prim_mode == PIPE_PRIM_POINTS);
key->is_lines = (prim_mode >= PIPE_PRIM_LINES &&
prim_mode <= PIPE_PRIM_LINE_STRIP);
+ key->blend = vc4->blend->rt[0];
if (vc4->framebuffer.cbufs[0])
key->color_format = vc4->framebuffer.cbufs[0]->format;
- key->depth_enabled = vc4->zsa->base.depth.enabled;
+ key->stencil_enabled = vc4->zsa->stencil_uniforms[0] != 0;
+ key->stencil_twoside = vc4->zsa->stencil_uniforms[1] != 0;
+ 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);
+ }
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;
memset(key, 0, sizeof(*key));
+ vc4_setup_shared_key(&key->base, &vc4->verttex);
key->base.shader_state = vc4->prog.bind_vs;
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
static void
write_texture_p0(struct vc4_context *vc4,
struct vc4_texture_stateobj *texstate,
- uint32_t tex_and_sampler)
+ uint32_t unit)
{
- uint32_t texi = (tex_and_sampler >> 0) & 0xff;
- struct pipe_sampler_view *texture = texstate->textures[texi];
+ struct pipe_sampler_view *texture = texstate->textures[unit];
struct vc4_resource *rsc = vc4_resource(texture->texture);
cl_reloc(vc4, &vc4->uniforms, rsc->bo,
- texture->u.tex.last_level);
+ rsc->slices[0].offset | texture->u.tex.last_level |
+ ((rsc->vc4_format & 7) << 4));
}
static void
write_texture_p1(struct vc4_context *vc4,
struct vc4_texture_stateobj *texstate,
- uint32_t tex_and_sampler)
+ uint32_t unit)
{
- uint32_t texi = (tex_and_sampler >> 0) & 0xff;
- uint32_t sampi = (tex_and_sampler >> 8) & 0xff;
- struct pipe_sampler_view *texture = texstate->textures[texi];
- struct pipe_sampler_state *sampler = texstate->samplers[sampi];
+ 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,
};
cl_u32(&vc4->uniforms,
- (1 << 31) /* XXX: data type */|
+ ((rsc->vc4_format >> 4) << 31) |
(texture->texture->height0 << 20) |
(texture->texture->width0 << 8) |
(imgfilter_map[sampler->mag_img_filter] << 7) |
gallium_uniforms[uinfo->data[i]]);
break;
case QUNIFORM_VIEWPORT_X_SCALE:
- cl_u32(&vc4->uniforms, fui(vc4->framebuffer.width *
- 16.0f / 2.0f));
+ cl_f(&vc4->uniforms, vc4->viewport.scale[0] * 16.0f);
break;
case QUNIFORM_VIEWPORT_Y_SCALE:
- cl_u32(&vc4->uniforms, fui(vc4->framebuffer.height *
- -16.0f / 2.0f));
+ cl_f(&vc4->uniforms, vc4->viewport.scale[1] * 16.0f);
break;
case QUNIFORM_VIEWPORT_Z_OFFSET:
- cl_u32(&vc4->uniforms, fui(vc4->viewport.translate[2]));
+ cl_f(&vc4->uniforms, vc4->viewport.translate[2]);
break;
case QUNIFORM_VIEWPORT_Z_SCALE:
- cl_u32(&vc4->uniforms, fui(vc4->viewport.scale[2]));
+ cl_f(&vc4->uniforms, vc4->viewport.scale[2]);
break;
case QUNIFORM_TEXTURE_CONFIG_P0:
uinfo->contents[i],
uinfo->data[i]));
break;
+
+ case QUNIFORM_BLEND_CONST_COLOR:
+ cl_f(&vc4->uniforms,
+ vc4->blend_color.color[uinfo->data[i]]);
+ break;
+
+ case QUNIFORM_STENCIL:
+ cl_u32(&vc4->uniforms,
+ vc4->zsa->stencil_uniforms[uinfo->data[i]] |
+ (uinfo->data[i] <= 1 ?
+ (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);