summaryrefslogtreecommitdiffstats
path: root/src/gallium
diff options
context:
space:
mode:
Diffstat (limited to 'src/gallium')
-rw-r--r--src/gallium/drivers/r600/r600_asm.c189
-rw-r--r--src/gallium/drivers/r600/r600_asm.h6
-rw-r--r--src/gallium/drivers/r600/r600_shader.c221
-rw-r--r--src/gallium/drivers/r600/r600_shader.h2
-rw-r--r--src/gallium/drivers/r600/r700_asm.c10
5 files changed, 151 insertions, 277 deletions
diff --git a/src/gallium/drivers/r600/r600_asm.c b/src/gallium/drivers/r600/r600_asm.c
index ca2bf93b0b8..e96236e06ee 100644
--- a/src/gallium/drivers/r600/r600_asm.c
+++ b/src/gallium/drivers/r600/r600_asm.c
@@ -608,10 +608,90 @@ static int replace_gpr_with_pv_ps(struct r600_bc_alu *slots[5], struct r600_bc_a
return 0;
}
+void r600_bc_special_constants(u32 value, unsigned *sel, unsigned *neg)
+{
+ switch(value) {
+ case 0:
+ *sel = V_SQ_ALU_SRC_0;
+ break;
+ case 1:
+ *sel = V_SQ_ALU_SRC_1_INT;
+ break;
+ case -1:
+ *sel = V_SQ_ALU_SRC_M_1_INT;
+ break;
+ case 0x3F800000: // 1.0f
+ *sel = V_SQ_ALU_SRC_1;
+ break;
+ case 0x3F000000: // 0.5f
+ *sel = V_SQ_ALU_SRC_0_5;
+ break;
+ case 0xBF800000: // -1.0f
+ *sel = V_SQ_ALU_SRC_1;
+ *neg ^= 1;
+ break;
+ case 0xBF000000: // -0.5f
+ *sel = V_SQ_ALU_SRC_0_5;
+ *neg ^= 1;
+ break;
+ default:
+ *sel = V_SQ_ALU_SRC_LITERAL;
+ break;
+ }
+}
+
+/* compute how many literal are needed */
+static int r600_bc_alu_nliterals(struct r600_bc_alu *alu, uint32_t literal[4], unsigned *nliteral)
+{
+ unsigned num_src = r600_bc_get_num_operands(alu);
+ unsigned i, j;
+
+ for (i = 0; i < num_src; ++i) {
+ if (alu->src[i].sel == V_SQ_ALU_SRC_LITERAL) {
+ uint32_t value = alu->src[i].value[alu->src[i].chan];
+ unsigned found = 0;
+ for (j = 0; j < *nliteral; ++j) {
+ if (literal[j] == value) {
+ found = 1;
+ break;
+ }
+ }
+ if (!found) {
+ if (*nliteral >= 4)
+ return -EINVAL;
+ literal[(*nliteral)++] = value;
+ }
+ }
+ }
+ return 0;
+}
+
+static void r600_bc_alu_adjust_literals(struct r600_bc_alu *alu, uint32_t literal[4], unsigned nliteral)
+{
+ unsigned num_src = r600_bc_get_num_operands(alu);
+ unsigned i, j;
+
+ for (i = 0; i < num_src; ++i) {
+ if (alu->src[i].sel == V_SQ_ALU_SRC_LITERAL) {
+ uint32_t value = alu->src[i].value[alu->src[i].chan];
+ for (j = 0; j < nliteral; ++j) {
+ if (literal[j] == value) {
+ alu->src[i].chan = j;
+ break;
+ }
+ }
+ }
+ }
+}
+
static int merge_inst_groups(struct r600_bc *bc, struct r600_bc_alu *slots[5], struct r600_bc_alu *alu_prev)
{
struct r600_bc_alu *prev[5];
struct r600_bc_alu *result[5] = { NULL };
+
+ uint32_t literal[4];
+ unsigned nliteral = 0;
+
int i, j, r, src, num_src;
int num_once_inst = 0;
@@ -620,13 +700,12 @@ static int merge_inst_groups(struct r600_bc *bc, struct r600_bc_alu *slots[5], s
return r;
for (i = 0; i < 5; ++i) {
- // TODO: we have literals? forget it!
- if (prev[i] && prev[i]->nliteral)
+ /* check number of literals */
+ if (prev[i] && r600_bc_alu_nliterals(prev[i], literal, &nliteral))
return 0;
- if (slots[i] && slots[i]->nliteral)
+ if (slots[i] && r600_bc_alu_nliterals(slots[i], literal, &nliteral))
return 0;
-
// let's check used slots
if (prev[i] && !slots[i]) {
result[i] = prev[i];
@@ -834,7 +913,6 @@ int r600_bc_add_alu_type(struct r600_bc *bc, const struct r600_bc_alu *alu, int
if (nalu == NULL)
return -ENOMEM;
memcpy(nalu, alu, sizeof(struct r600_bc_alu));
- nalu->nliteral = 0;
if (bc->cf_last != NULL && bc->cf_last->inst != (type << 3)) {
/* check if we could add it anyway */
@@ -880,20 +958,10 @@ int r600_bc_add_alu_type(struct r600_bc *bc, const struct r600_bc_alu *alu, int
if (nalu->src[i].sel >= bc->ngpr && nalu->src[i].sel < 128) {
bc->ngpr = nalu->src[i].sel + 1;
}
- /* compute how many literal are needed
- * either 2 or 4 literals
- */
- if (nalu->src[i].sel == 253) {
- if (((nalu->src[i].chan + 2) & 0x6) > nalu->nliteral) {
- nalu->nliteral = (nalu->src[i].chan + 2) & 0x6;
- }
- }
- }
- if (!LIST_IS_EMPTY(&bc->cf_last->alu)) {
- lalu = LIST_ENTRY(struct r600_bc_alu, bc->cf_last->alu.prev, list);
- if (!lalu->last && lalu->nliteral > nalu->nliteral) {
- nalu->nliteral = lalu->nliteral;
- }
+ if (nalu->src[i].sel == V_SQ_ALU_SRC_LITERAL)
+ r600_bc_special_constants(
+ nalu->src[i].value[nalu->src[i].chan],
+ &nalu->src[i].sel, &nalu->src[i].neg);
}
if (nalu->dst.sel >= bc->ngpr) {
bc->ngpr = nalu->dst.sel + 1;
@@ -938,46 +1006,6 @@ int r600_bc_add_alu(struct r600_bc *bc, const struct r600_bc_alu *alu)
return r600_bc_add_alu_type(bc, alu, BC_INST(bc, V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU));
}
-int r600_bc_add_literal(struct r600_bc *bc, const u32 *value)
-{
- struct r600_bc_alu *alu;
-
- if (bc->cf_last == NULL) {
- return 0;
- }
- if (bc->cf_last->inst == V_SQ_CF_WORD1_SQ_CF_INST_TEX) {
- return 0;
- }
- /* all same on EG */
- if (bc->cf_last->inst == V_SQ_CF_WORD1_SQ_CF_INST_JUMP ||
- bc->cf_last->inst == V_SQ_CF_WORD1_SQ_CF_INST_ELSE ||
- bc->cf_last->inst == V_SQ_CF_WORD1_SQ_CF_INST_LOOP_START_NO_AL ||
- bc->cf_last->inst == V_SQ_CF_WORD1_SQ_CF_INST_LOOP_BREAK ||
- bc->cf_last->inst == V_SQ_CF_WORD1_SQ_CF_INST_LOOP_CONTINUE ||
- bc->cf_last->inst == V_SQ_CF_WORD1_SQ_CF_INST_LOOP_END ||
- bc->cf_last->inst == V_SQ_CF_WORD1_SQ_CF_INST_POP) {
- return 0;
- }
- /* same on EG */
- if (((bc->cf_last->inst != (V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU << 3)) &&
- (bc->cf_last->inst != (V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_POP_AFTER << 3)) &&
- (bc->cf_last->inst != (V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_POP2_AFTER << 3)) &&
- (bc->cf_last->inst != (V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_PUSH_BEFORE << 3))) ||
- LIST_IS_EMPTY(&bc->cf_last->alu)) {
- R600_ERR("last CF is not ALU (%p)\n", bc->cf_last);
- return -EINVAL;
- }
- alu = LIST_ENTRY(struct r600_bc_alu, bc->cf_last->alu.prev, list);
- if (!alu->last || !alu->nliteral || alu->literal_added) {
- return 0;
- }
- memcpy(alu->value, value, 4 * 4);
- bc->cf_last->ndw += alu->nliteral;
- bc->ndw += alu->nliteral;
- alu->literal_added = 1;
- return 0;
-}
-
int r600_bc_add_vtx(struct r600_bc *bc, const struct r600_bc_vtx *vtx)
{
struct r600_bc_vtx *nvtx = r600_bc_vtx();
@@ -1134,8 +1162,6 @@ static int r600_bc_tex_build(struct r600_bc *bc, struct r600_bc_tex *tex, unsign
/* r600 only, r700/eg bits in r700_asm.c */
static int r600_bc_alu_build(struct r600_bc *bc, struct r600_bc_alu *alu, unsigned id)
{
- unsigned i;
-
/* don't replace gpr by pv or ps for destination register */
bc->bytecode[id++] = S_SQ_ALU_WORD0_SRC0_SEL(alu->src[0].sel) |
S_SQ_ALU_WORD0_SRC0_REL(alu->src[0].rel) |
@@ -1172,14 +1198,6 @@ static int r600_bc_alu_build(struct r600_bc *bc, struct r600_bc_alu *alu, unsign
S_SQ_ALU_WORD1_OP2_UPDATE_EXECUTE_MASK(alu->predicate) |
S_SQ_ALU_WORD1_OP2_UPDATE_PRED(alu->predicate);
}
- if (alu->last) {
- if (alu->nliteral && !alu->literal_added) {
- R600_ERR("Bug in ALU processing for instruction 0x%08x, literal not added correctly\n", alu->inst);
- }
- for (i = 0; i < alu->nliteral; i++) {
- bc->bytecode[id++] = alu->value[i];
- }
- }
return 0;
}
@@ -1257,8 +1275,10 @@ int r600_bc_build(struct r600_bc *bc)
struct r600_bc_alu *alu;
struct r600_bc_vtx *vtx;
struct r600_bc_tex *tex;
+ uint32_t literal[4];
+ unsigned nliteral;
unsigned addr;
- int r;
+ int i, r;
if (bc->callstack[0].max > 0)
bc->nstack = ((bc->callstack[0].max + 3) >> 2) + 2;
@@ -1275,6 +1295,16 @@ int r600_bc_build(struct r600_bc *bc)
case (V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_POP_AFTER << 3):
case (V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_POP2_AFTER << 3):
case (V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_PUSH_BEFORE << 3):
+ nliteral = 0;
+ LIST_FOR_EACH_ENTRY(alu, &cf->alu, list) {
+ r = r600_bc_alu_nliterals(alu, literal, &nliteral);
+ if (r)
+ return r;
+ if (alu->last) {
+ cf->ndw += align(nliteral, 2);
+ nliteral = 0;
+ }
+ }
break;
case V_SQ_CF_WORD1_SQ_CF_INST_TEX:
case V_SQ_CF_WORD1_SQ_CF_INST_VTX:
@@ -1323,7 +1353,12 @@ int r600_bc_build(struct r600_bc *bc)
case (V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_POP_AFTER << 3):
case (V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_POP2_AFTER << 3):
case (V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_PUSH_BEFORE << 3):
+ nliteral = 0;
LIST_FOR_EACH_ENTRY(alu, &cf->alu, list) {
+ r = r600_bc_alu_nliterals(alu, literal, &nliteral);
+ if (r)
+ return r;
+ r600_bc_alu_adjust_literals(alu, literal, nliteral);
switch(bc->chiprev) {
case CHIPREV_R600:
r = r600_bc_alu_build(bc, alu, addr);
@@ -1340,7 +1375,10 @@ int r600_bc_build(struct r600_bc *bc)
return r;
addr += 2;
if (alu->last) {
- addr += alu->nliteral;
+ for (i = 0; i < align(nliteral, 2); ++i) {
+ bc->bytecode[addr++] = literal[i];
+ }
+ nliteral = 0;
}
}
break;
@@ -1427,6 +1465,8 @@ void r600_bc_dump(struct r600_bc *bc)
struct r600_bc_tex *tex = NULL;
unsigned i, id;
+ uint32_t literal[4];
+ unsigned nliteral;
char chip = '6';
switch (bc->chiprev) {
@@ -1513,7 +1553,10 @@ void r600_bc_dump(struct r600_bc *bc)
}
id = cf->addr;
+ nliteral = 0;
LIST_FOR_EACH_ENTRY(alu, &cf->alu, list) {
+ r600_bc_alu_nliterals(alu, literal, &nliteral);
+
fprintf(stderr, "%04d %08X ", id, bc->bytecode[id]);
fprintf(stderr, "SRC0(SEL:%d ", alu->src[0].sel);
fprintf(stderr, "REL:%d ", alu->src[0].rel);
@@ -1548,10 +1591,12 @@ void r600_bc_dump(struct r600_bc *bc)
id++;
if (alu->last) {
- for (i = 0; i < alu->nliteral; i++, id++) {
+ for (i = 0; i < nliteral; i++, id++) {
float *f = (float*)(bc->bytecode + id);
fprintf(stderr, "%04d %08X\t%f\n", id, bc->bytecode[id], *f);
}
+ id += nliteral & 1;
+ nliteral = 0;
}
}
diff --git a/src/gallium/drivers/r600/r600_asm.h b/src/gallium/drivers/r600/r600_asm.h
index 570292e9fdc..278b4466cb0 100644
--- a/src/gallium/drivers/r600/r600_asm.h
+++ b/src/gallium/drivers/r600/r600_asm.h
@@ -34,6 +34,7 @@ struct r600_bc_alu_src {
unsigned neg;
unsigned abs;
unsigned rel;
+ u32 *value;
};
struct r600_bc_alu_dst {
@@ -52,11 +53,8 @@ struct r600_bc_alu {
unsigned last;
unsigned is_op3;
unsigned predicate;
- unsigned nliteral;
- unsigned literal_added;
unsigned bank_swizzle;
unsigned bank_swizzle_force;
- u32 value[4];
unsigned omod;
};
@@ -196,13 +194,13 @@ void eg_cf_vtx(struct r600_vertex_element *ve, u32 *bytecode, unsigned count);
int r600_bc_init(struct r600_bc *bc, enum radeon_family family);
void r600_bc_clear(struct r600_bc *bc);
int r600_bc_add_alu(struct r600_bc *bc, const struct r600_bc_alu *alu);
-int r600_bc_add_literal(struct r600_bc *bc, const u32 *value);
int r600_bc_add_vtx(struct r600_bc *bc, const struct r600_bc_vtx *vtx);
int r600_bc_add_tex(struct r600_bc *bc, const struct r600_bc_tex *tex);
int r600_bc_add_output(struct r600_bc *bc, const struct r600_bc_output *output);
int r600_bc_build(struct r600_bc *bc);
int r600_bc_add_cfinst(struct r600_bc *bc, int inst);
int r600_bc_add_alu_type(struct r600_bc *bc, const struct r600_bc_alu *alu, int type);
+void r600_bc_special_constants(u32 value, unsigned *sel, unsigned *neg);
void r600_bc_dump(struct r600_bc *bc);
void r600_cf_vtx(struct r600_vertex_element *ve, u32 *bytecode, unsigned count);
void r600_cf_vtx_tc(struct r600_vertex_element *ve, u32 *bytecode, unsigned count);
diff --git a/src/gallium/drivers/r600/r600_shader.c b/src/gallium/drivers/r600/r600_shader.c
index 78739bf89d8..e85e829badd 100644
--- a/src/gallium/drivers/r600/r600_shader.c
+++ b/src/gallium/drivers/r600/r600_shader.c
@@ -225,11 +225,12 @@ int r600_pipe_shader(struct pipe_context *ctx, struct r600_pipe_shader *shader)
return 0;
}
-int r600_shader_from_tgsi(const struct tgsi_token *tokens, struct r600_shader *shader);
+int r600_shader_from_tgsi(const struct tgsi_token *tokens, struct r600_shader *shader, u32 **literals);
int r600_pipe_shader_create(struct pipe_context *ctx, struct r600_pipe_shader *shader, const struct tgsi_token *tokens)
{
static int dump_shaders = -1;
struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
+ u32 *literals;
int r;
/* Would like some magic "get_bool_option_once" routine.
@@ -242,12 +243,13 @@ int r600_pipe_shader_create(struct pipe_context *ctx, struct r600_pipe_shader *s
tgsi_dump(tokens, 0);
}
shader->shader.family = r600_get_family(rctx->radeon);
- r = r600_shader_from_tgsi(tokens, &shader->shader);
+ r = r600_shader_from_tgsi(tokens, &shader->shader, &literals);
if (r) {
R600_ERR("translation from TGSI failed !\n");
return r;
}
r = r600_bc_build(&shader->shader.bc);
+ free(literals);
if (r) {
R600_ERR("building bytecode failed !\n");
return r;
@@ -282,7 +284,6 @@ struct r600_shader_ctx {
struct r600_shader_tgsi_instruction *inst_info;
struct r600_bc *bc;
struct r600_shader *shader;
- u32 value[4];
u32 *literals;
u32 nliterals;
u32 max_driver_temp_used;
@@ -491,7 +492,7 @@ static int evergreen_gpr_count(struct r600_shader_ctx *ctx)
return ctx->num_interp_gpr;
}
-int r600_shader_from_tgsi(const struct tgsi_token *tokens, struct r600_shader *shader)
+int r600_shader_from_tgsi(const struct tgsi_token *tokens, struct r600_shader *shader, u32 **literals)
{
struct tgsi_full_immediate *immediate;
struct r600_shader_ctx ctx;
@@ -599,9 +600,6 @@ int r600_shader_from_tgsi(const struct tgsi_token *tokens, struct r600_shader *s
r = ctx.inst_info->process(&ctx);
if (r)
goto out_err;
- r = r600_bc_add_literal(ctx.bc, ctx.value);
- if (r)
- goto out_err;
break;
case TGSI_TOKEN_TYPE_PROPERTY:
break;
@@ -722,7 +720,7 @@ int r600_shader_from_tgsi(const struct tgsi_token *tokens, struct r600_shader *s
if (r)
goto out_err;
}
- free(ctx.literals);
+ *literals = ctx.literals;
tgsi_parse_free(&ctx.parse);
return 0;
out_err:
@@ -756,38 +754,13 @@ static int tgsi_src(struct r600_shader_ctx *ctx,
(tgsi_src->Register.SwizzleX == tgsi_src->Register.SwizzleW)) {
index = tgsi_src->Register.Index * 4 + tgsi_src->Register.SwizzleX;
- switch(ctx->literals[index]) {
- case 0:
- r600_src->sel = V_SQ_ALU_SRC_0;
- return 0;
- case 1:
- r600_src->sel = V_SQ_ALU_SRC_1_INT;
- return 0;
- case -1:
- r600_src->sel = V_SQ_ALU_SRC_M_1_INT;
- return 0;
- case 0x3F800000: // 1.0f
- r600_src->sel = V_SQ_ALU_SRC_1;
+ r600_bc_special_constants(ctx->literals[index], &r600_src->sel, &r600_src->neg);
+ if (r600_src->sel != V_SQ_ALU_SRC_LITERAL)
return 0;
- case 0x3F000000: // 0.5f
- r600_src->sel = V_SQ_ALU_SRC_0_5;
- return 0;
- case 0xBF800000: // -1.0f
- r600_src->sel = V_SQ_ALU_SRC_1;
- r600_src->neg ^= 1;
- return 0;
- case 0xBF000000: // -0.5f
- r600_src->sel = V_SQ_ALU_SRC_0_5;
- r600_src->neg ^= 1;
- return 0;
- }
}
index = tgsi_src->Register.Index;
r600_src->sel = V_SQ_ALU_SRC_LITERAL;
- ctx->value[0] = ctx->literals[index * 4 + 0];
- ctx->value[1] = ctx->literals[index * 4 + 1];
- ctx->value[2] = ctx->literals[index * 4 + 2];
- ctx->value[3] = ctx->literals[index * 4 + 3];
+ r600_src->value = ctx->literals + index * 4;
} else {
if (tgsi_src->Register.Indirect)
r600_src->rel = V_SQ_REL_RELATIVE;
@@ -893,6 +866,7 @@ static int tgsi_split_literal_constant(struct r600_shader_ctx *ctx, struct r600_
alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV);
alu.src[0].sel = r600_src[i].sel;
alu.src[0].chan = k;
+ alu.src[0].value = r600_src[i].value;
alu.dst.sel = treg;
alu.dst.chan = k;
alu.dst.write = 1;
@@ -902,9 +876,6 @@ static int tgsi_split_literal_constant(struct r600_shader_ctx *ctx, struct r600_
if (r)
return r;
}
- r = r600_bc_add_literal(ctx->bc, &ctx->literals[inst->Src[i].Register.Index * 4]);
- if (r)
- return r;
r600_src[i].sel = treg;
j--;
}
@@ -999,12 +970,14 @@ static int tgsi_op2_swap(struct r600_shader_ctx *ctx)
static int tgsi_setup_trig(struct r600_shader_ctx *ctx,
struct r600_bc_alu_src r600_src[3])
{
+ static float half_inv_pi = 1.0 /(3.1415926535 * 2);
+ static float double_pi = 3.1415926535 * 2;
+ static float neg_pi = -3.1415926535;
+
struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction;
- int r, src0_chan;
- uint32_t lit_vals[4];
+ int r;
struct r600_bc_alu alu;
- memset(lit_vals, 0, 4*4);
r = tgsi_split_constant(ctx, r600_src);
if (r)
return r;
@@ -1012,22 +985,6 @@ static int tgsi_setup_trig(struct r600_shader_ctx *ctx,
if (r)
return r;
- src0_chan = tgsi_chan(&inst->Src[0], 0);
-
- /* We are going to feed two literals to the MAD below,
- * which means that if the first operand is a literal as well,
- * we need to copy its value manually.
- */
- if (r600_src[0].sel == V_SQ_ALU_SRC_LITERAL) {
- unsigned index = inst->Src[0].Register.Index;
-
- lit_vals[2] = ctx->literals[index * 4 + src0_chan];
- src0_chan = 2;
- }
-
- lit_vals[0] = fui(1.0 /(3.1415926535 * 2));
- lit_vals[1] = fui(0.5f);
-
memset(&alu, 0, sizeof(struct r600_bc_alu));
alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP3_SQ_OP3_INST_MULADD);
alu.is_op3 = 1;
@@ -1037,19 +994,17 @@ static int tgsi_setup_trig(struct r600_shader_ctx *ctx,
alu.dst.write = 1;
alu.src[0] = r600_src[0];
- alu.src[0].chan = src0_chan;
+ alu.src[0].chan = tgsi_chan(&inst->Src[0], 0);
alu.src[1].sel = V_SQ_ALU_SRC_LITERAL;
alu.src[1].chan = 0;
- alu.src[2].sel = V_SQ_ALU_SRC_LITERAL;
+ alu.src[1].value = (uint32_t *)&half_inv_pi;
+ alu.src[2].sel = V_SQ_ALU_SRC_0_5;
alu.src[2].chan = 1;
alu.last = 1;
r = r600_bc_add_alu(ctx->bc, &alu);
if (r)
return r;
- r = r600_bc_add_literal(ctx->bc, lit_vals);
- if (r)
- return r;
memset(&alu, 0, sizeof(struct r600_bc_alu));
alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FRACT);
@@ -1065,14 +1020,6 @@ static int tgsi_setup_trig(struct r600_shader_ctx *ctx,
if (r)
return r;
- if (ctx->bc->chiprev == CHIPREV_R600) {
- lit_vals[0] = fui(3.1415926535897f * 2.0f);
- lit_vals[1] = fui(-3.1415926535897f);
- } else {
- lit_vals[0] = fui(1.0f);
- lit_vals[1] = fui(-0.5f);
- }
-
memset(&alu, 0, sizeof(struct r600_bc_alu));
alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP3_SQ_OP3_INST_MULADD);
alu.is_op3 = 1;
@@ -1088,13 +1035,20 @@ static int tgsi_setup_trig(struct r600_shader_ctx *ctx,
alu.src[1].chan = 0;
alu.src[2].sel = V_SQ_ALU_SRC_LITERAL;
alu.src[2].chan = 1;
+
+ if (ctx->bc->chiprev == CHIPREV_R600) {
+ alu.src[1].value = (uint32_t *)&double_pi;
+ alu.src[2].value = (uint32_t *)&neg_pi;
+ } else {
+ alu.src[1].sel = V_SQ_ALU_SRC_1;
+ alu.src[2].sel = V_SQ_ALU_SRC_0_5;
+ alu.src[2].neg = 1;
+ }
+
alu.last = 1;
r = r600_bc_add_alu(ctx->bc, &alu);
if (r)
return r;
- r = r600_bc_add_literal(ctx->bc, lit_vals);
- if (r)
- return r;
return 0;
}
@@ -1210,10 +1164,6 @@ static int tgsi_scs(struct r600_shader_ctx *ctx)
r = r600_bc_add_alu(ctx->bc, &alu);
if (r)
return r;
-
- r = r600_bc_add_literal(ctx->bc, ctx->value);
- if (r)
- return r;
}
/* dst.w = 1.0; */
@@ -1234,10 +1184,6 @@ static int tgsi_scs(struct r600_shader_ctx *ctx)
r = r600_bc_add_alu(ctx->bc, &alu);
if (r)
return r;
-
- r = r600_bc_add_literal(ctx->bc, ctx->value);
- if (r)
- return r;
}
return 0;
@@ -1273,9 +1219,6 @@ static int tgsi_kill(struct r600_shader_ctx *ctx)
if (r)
return r;
}
- r = r600_bc_add_literal(ctx->bc, ctx->value);
- if (r)
- return r;
/* kill must be last in ALU */
ctx->bc->force_add_cf = 1;
@@ -1338,10 +1281,6 @@ static int tgsi_lit(struct r600_shader_ctx *ctx)
if (r)
return r;
- r = r600_bc_add_literal(ctx->bc, ctx->value);
- if (r)
- return r;
-
if (inst->Dst[0].Register.WriteMask & (1 << 2))
{
int chan;
@@ -1360,10 +1299,6 @@ static int tgsi_lit(struct r600_shader_ctx *ctx)
if (r)
return r;
- r = r600_bc_add_literal(ctx->bc, ctx->value);
- if (r)
- return r;
-
chan = alu.dst.chan;
sel = alu.dst.sel;
@@ -1386,9 +1321,6 @@ static int tgsi_lit(struct r600_shader_ctx *ctx)
if (r)
return r;
- r = r600_bc_add_literal(ctx->bc, ctx->value);
- if (r)
- return r;
/* dst.z = exp(tmp.x) */
memset(&alu, 0, sizeof(struct r600_bc_alu));
alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_EXP_IEEE);
@@ -1432,9 +1364,6 @@ static int tgsi_rsq(struct r600_shader_ctx *ctx)
r = r600_bc_add_alu(ctx->bc, &alu);
if (r)
return r;
- r = r600_bc_add_literal(ctx->bc, ctx->value);
- if (r)
- return r;
/* replicate result */
return tgsi_helper_tempx_replicate(ctx);
}
@@ -1483,9 +1412,6 @@ static int tgsi_trans_srcx_replicate(struct r600_shader_ctx *ctx)
r = r600_bc_add_alu(ctx->bc, &alu);
if (r)
return r;
- r = r600_bc_add_literal(ctx->bc, ctx->value);
- if (r)
- return r;
/* replicate result */
return tgsi_helper_tempx_replicate(ctx);
}
@@ -1509,9 +1435,6 @@ static int tgsi_pow(struct r600_shader_ctx *ctx)
r = r600_bc_add_alu(ctx->bc, &alu);
if (r)
return r;
- r = r600_bc_add_literal(ctx->bc,ctx->value);
- if (r)
- return r;
/* b * LOG2(a) */
memset(&alu, 0, sizeof(struct r600_bc_alu));
alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MUL);
@@ -1526,9 +1449,6 @@ static int tgsi_pow(struct r600_shader_ctx *ctx)
r = r600_bc_add_alu(ctx->bc, &alu);
if (r)
return r;
- r = r600_bc_add_literal(ctx->bc,ctx->value);
- if (r)
- return r;
/* POW(a,b) = EXP2(b * LOG2(a))*/
memset(&alu, 0, sizeof(struct r600_bc_alu));
alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_EXP_IEEE);
@@ -1539,9 +1459,6 @@ static int tgsi_pow(struct r600_shader_ctx *ctx)
r = r600_bc_add_alu(ctx->bc, &alu);
if (r)
return r;
- r = r600_bc_add_literal(ctx->bc,ctx->value);
- if (r)
- return r;
return tgsi_helper_tempx_replicate(ctx);
}
@@ -1581,9 +1498,6 @@ static int tgsi_ssg(struct r600_shader_ctx *ctx)
if (r)
return r;
}
- r = r600_bc_add_literal(ctx->bc, ctx->value);
- if (r)
- return r;
/* dst = (-tmp > 0 ? -1 : tmp) */
for (i = 0; i < 4; i++) {
@@ -1618,9 +1532,6 @@ static int tgsi_helper_copy(struct r600_shader_ctx *ctx, struct tgsi_full_instru
struct r600_bc_alu alu;
int i, r;
- r = r600_bc_add_literal(ctx->bc, ctx->value);
- if (r)
- return r;
for (i = 0; i < 4; i++) {
memset(&alu, 0, sizeof(struct r600_bc_alu));
if (!(inst->Dst[0].Register.WriteMask & (1 << i))) {
@@ -1749,6 +1660,7 @@ static int tgsi_dp(struct r600_shader_ctx *ctx)
static int tgsi_tex(struct r600_shader_ctx *ctx)
{
+ static float one_point_five = 1.5f;
struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction;
struct r600_bc_tex tex;
struct r600_bc_alu alu;
@@ -1758,7 +1670,6 @@ static int tgsi_tex(struct r600_shader_ctx *ctx)
boolean src_not_temp =
inst->Src[0].Register.File != TGSI_FILE_TEMPORARY &&
inst->Src[0].Register.File != TGSI_FILE_INPUT;
- uint32_t lit_vals[4];
src_gpr = ctx->file_offset[inst->Src[0].Register.File] + inst->Src[0].Register.Index;
@@ -1887,6 +1798,7 @@ static int tgsi_tex(struct r600_shader_ctx *ctx)
alu.src[2].sel = V_SQ_ALU_SRC_LITERAL;
alu.src[2].chan = 0;
+ alu.src[2].value = (u32*)&one_point_five;
alu.dst.sel = ctx->temp_reg;
alu.dst.chan = 0;
@@ -1907,6 +1819,7 @@ static int tgsi_tex(struct r600_shader_ctx *ctx)
alu.src[2].sel = V_SQ_ALU_SRC_LITERAL;
alu.src[2].chan = 0;
+ alu.src[2].value = (u32*)&one_point_five;
alu.dst.sel = ctx->temp_reg;
alu.dst.chan = 1;
@@ -1917,11 +1830,6 @@ static int tgsi_tex(struct r600_shader_ctx *ctx)
if (r)
return r;
- lit_vals[0] = fui(1.5f);
-
- r = r600_bc_add_literal(ctx->bc, lit_vals);
- if (r)
- return r;
src_not_temp = FALSE;
src_gpr = ctx->temp_reg;
}
@@ -2055,9 +1963,6 @@ static int tgsi_lrp(struct r600_shader_ctx *ctx)
if (r)
return r;
}
- r = r600_bc_add_literal(ctx->bc, ctx->value);
- if (r)
- return r;
/* (1 - src0) * src2 */
for (i = 0; i < lasti + 1; i++) {
@@ -2080,9 +1985,6 @@ static int tgsi_lrp(struct r600_shader_ctx *ctx)
if (r)
return r;
}
- r = r600_bc_add_literal(ctx->bc, ctx->value);
- if (r)
- return r;
/* src0 * src1 + (1 - src0) * src2 */
for (i = 0; i < lasti + 1; i++) {
@@ -2223,10 +2125,6 @@ static int tgsi_xpd(struct r600_shader_ctx *ctx)
r = r600_bc_add_alu(ctx->bc, &alu);
if (r)
return r;
-
- r = r600_bc_add_literal(ctx->bc, ctx->value);
- if (r)
- return r;
}
for (i = 0; i < 4; i++) {
@@ -2284,10 +2182,6 @@ static int tgsi_xpd(struct r600_shader_ctx *ctx)
r = r600_bc_add_alu(ctx->bc, &alu);
if (r)
return r;
-
- r = r600_bc_add_literal(ctx->bc, ctx->value);
- if (r)
- return r;
}
if (use_temp)
return tgsi_helper_copy(ctx, inst);
@@ -2320,10 +2214,6 @@ static int tgsi_exp(struct r600_shader_ctx *ctx)
if (r)
return r;
- r = r600_bc_add_literal(ctx->bc, ctx->value);
- if (r)
- return r;
-
alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_EXP_IEEE);
alu.src[0].sel = ctx->temp_reg;
alu.src[0].chan = 0;
@@ -2335,10 +2225,6 @@ static int tgsi_exp(struct r600_shader_ctx *ctx)
r = r600_bc_add_alu(ctx->bc, &alu);
if (r)
return r;
-
- r = r600_bc_add_literal(ctx->bc, ctx->value);
- if (r)
- return r;
}
/* result.y = tmp - floor(tmp); */
@@ -2364,9 +2250,6 @@ static int tgsi_exp(struct r600_shader_ctx *ctx)
r = r600_bc_add_alu(ctx->bc, &alu);
if (r)
return r;
- r = r600_bc_add_literal(ctx->bc, ctx->value);
- if (r)
- return r;
}
/* result.z = RoughApprox2ToX(tmp);*/
@@ -2387,9 +2270,6 @@ static int tgsi_exp(struct r600_shader_ctx *ctx)
r = r600_bc_add_alu(ctx->bc, &alu);
if (r)
return r;
- r = r600_bc_add_literal(ctx->bc, ctx->value);
- if (r)
- return r;
}
/* result.w = 1.0;*/
@@ -2407,9 +2287,6 @@ static int tgsi_exp(struct r600_shader_ctx *ctx)
r = r600_bc_add_alu(ctx->bc, &alu);
if (r)
return r;
- r = r600_bc_add_literal(ctx->bc, ctx->value);
- if (r)
- return r;
}
return tgsi_helper_copy(ctx, inst);
}
@@ -2439,10 +2316,6 @@ static int tgsi_log(struct r600_shader_ctx *ctx)
if (r)
return r;
- r = r600_bc_add_literal(ctx->bc, ctx->value);
- if (r)
- return r;
-
alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FLOOR);
alu.src[0].sel = ctx->temp_reg;
alu.src[0].chan = 0;
@@ -2455,10 +2328,6 @@ static int tgsi_log(struct r600_shader_ctx *ctx)
r = r600_bc_add_alu(ctx->bc, &alu);
if (r)
return r;
-
- r = r600_bc_add_literal(ctx->bc, ctx->value);
- if (r)
- return r;
}
/* result.y = src.x / (2 ^ floor(log2(src.x))); */
@@ -2481,10 +2350,6 @@ static int tgsi_log(struct r600_shader_ctx *ctx)
if (r)
return r;
- r = r600_bc_add_literal(ctx->bc, ctx->value);
- if (r)
- return r;
-
memset(&alu, 0, sizeof(struct r600_bc_alu));
alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FLOOR);
@@ -2500,10 +2365,6 @@ static int tgsi_log(struct r600_shader_ctx *ctx)
if (r)
return r;
- r = r600_bc_add_literal(ctx->bc, ctx->value);
- if (r)
- return r;
-
memset(&alu, 0, sizeof(struct r600_bc_alu));
alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_EXP_IEEE);
@@ -2519,10 +2380,6 @@ static int tgsi_log(struct r600_shader_ctx *ctx)
if (r)
return r;
- r = r600_bc_add_literal(ctx->bc, ctx->value);
- if (r)
- return r;
-
memset(&alu, 0, sizeof(struct r600_bc_alu));
alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_RECIP_IEEE);
@@ -2538,10 +2395,6 @@ static int tgsi_log(struct r600_shader_ctx *ctx)
if (r)
return r;
- r = r600_bc_add_literal(ctx->bc, ctx->value);
- if (r)
- return r;
-
memset(&alu, 0, sizeof(struct r600_bc_alu));
alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MUL);
@@ -2563,10 +2416,6 @@ static int tgsi_log(struct r600_shader_ctx *ctx)
r = r600_bc_add_alu(ctx->bc, &alu);
if (r)
return r;
-
- r = r600_bc_add_literal(ctx->bc, ctx->value);
- if (r)
- return r;
}
/* result.z = log2(src);*/
@@ -2588,10 +2437,6 @@ static int tgsi_log(struct r600_shader_ctx *ctx)
r = r600_bc_add_alu(ctx->bc, &alu);
if (r)
return r;
-
- r = r600_bc_add_literal(ctx->bc, ctx->value);
- if (r)
- return r;
}
/* result.w = 1.0; */
@@ -2610,10 +2455,6 @@ static int tgsi_log(struct r600_shader_ctx *ctx)
r = r600_bc_add_alu(ctx->bc, &alu);
if (r)
return r;
-
- r = r600_bc_add_literal(ctx->bc, ctx->value);
- if (r)
- return r;
}
return tgsi_helper_copy(ctx, inst);
diff --git a/src/gallium/drivers/r600/r600_shader.h b/src/gallium/drivers/r600/r600_shader.h
index 35b0331525a..935dd6fe3ab 100644
--- a/src/gallium/drivers/r600/r600_shader.h
+++ b/src/gallium/drivers/r600/r600_shader.h
@@ -47,6 +47,6 @@ struct r600_shader {
boolean uses_kill;
};
-int r600_shader_from_tgsi(const struct tgsi_token *tokens, struct r600_shader *shader);
+int r600_shader_from_tgsi(const struct tgsi_token *tokens, struct r600_shader *shader, u32 **literals);
#endif
diff --git a/src/gallium/drivers/r600/r700_asm.c b/src/gallium/drivers/r600/r700_asm.c
index 3eb6fb50ca7..a7f2f54736e 100644
--- a/src/gallium/drivers/r600/r700_asm.c
+++ b/src/gallium/drivers/r600/r700_asm.c
@@ -29,8 +29,6 @@
int r700_bc_alu_build(struct r600_bc *bc, struct r600_bc_alu *alu, unsigned id)
{
- unsigned i;
-
bc->bytecode[id++] = S_SQ_ALU_WORD0_SRC0_SEL(alu->src[0].sel) |
S_SQ_ALU_WORD0_SRC0_REL(alu->src[0].rel) |
S_SQ_ALU_WORD0_SRC0_CHAN(alu->src[0].chan) |
@@ -67,13 +65,5 @@ int r700_bc_alu_build(struct r600_bc *bc, struct r600_bc_alu *alu, unsigned id)
S_SQ_ALU_WORD1_OP2_UPDATE_EXECUTE_MASK(alu->predicate) |
S_SQ_ALU_WORD1_OP2_UPDATE_PRED(alu->predicate);
}
- if (alu->last) {
- if (alu->nliteral && !alu->literal_added) {
- R600_ERR("Bug in ALU processing for instruction 0x%08x, literal not added correctly\n", alu->inst);
- }
- for (i = 0; i < alu->nliteral; i++) {
- bc->bytecode[id++] = alu->value[i];
- }
- }
return 0;
}
generated by cgit v1.2.3 (git 2.39.1) at 2024-11-29 17:41:53 +0000