summaryrefslogtreecommitdiffstats
path: root/src/gallium/drivers/r600/r600_shader.c
diff options
context:
space:
mode:
Diffstat (limited to 'src/gallium/drivers/r600/r600_shader.c')
-rw-r--r--src/gallium/drivers/r600/r600_shader.c564
1 files changed, 194 insertions, 370 deletions
diff --git a/src/gallium/drivers/r600/r600_shader.c b/src/gallium/drivers/r600/r600_shader.c
index 77b1809..e85e829 100644
--- a/src/gallium/drivers/r600/r600_shader.c
+++ b/src/gallium/drivers/r600/r600_shader.c
@@ -44,6 +44,9 @@ static void r600_pipe_shader_vs(struct pipe_context *ctx, struct r600_pipe_shade
rstate->nregs = 0;
/* so far never got proper semantic id from tgsi */
+ /* FIXME better to move this in config things so they get emited
+ * only one time per cs
+ */
for (i = 0; i < 10; i++) {
spi_vs_out_id[i] = 0;
}
@@ -67,20 +70,11 @@ static void r600_pipe_shader_vs(struct pipe_context *ctx, struct r600_pipe_shade
S_028868_STACK_SIZE(rshader->bc.nstack),
0xFFFFFFFF, NULL);
r600_pipe_state_add_reg(rstate,
- R_0288A4_SQ_PGM_RESOURCES_FS,
- 0x00000000, 0xFFFFFFFF, NULL);
- r600_pipe_state_add_reg(rstate,
R_0288D0_SQ_PGM_CF_OFFSET_VS,
0x00000000, 0xFFFFFFFF, NULL);
r600_pipe_state_add_reg(rstate,
- R_0288DC_SQ_PGM_CF_OFFSET_FS,
- 0x00000000, 0xFFFFFFFF, NULL);
- r600_pipe_state_add_reg(rstate,
R_028858_SQ_PGM_START_VS,
r600_bo_offset(shader->bo) >> 8, 0xFFFFFFFF, shader->bo);
- r600_pipe_state_add_reg(rstate,
- R_028894_SQ_PGM_START_FS,
- r600_bo_offset(shader->bo_fetch) >> 8, 0xFFFFFFFF, shader->bo_fetch);
r600_pipe_state_add_reg(rstate,
R_03E200_SQ_LOOP_CONST_0 + (32 * 4), 0x01000FFF,
@@ -104,37 +98,20 @@ int r600_find_vs_semantic_index(struct r600_shader *vs,
static void r600_pipe_shader_ps(struct pipe_context *ctx, struct r600_pipe_shader *shader)
{
- struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
struct r600_pipe_state *rstate = &shader->rstate;
struct r600_shader *rshader = &shader->shader;
- unsigned i, tmp, exports_ps, num_cout, spi_ps_in_control_0, spi_input_z, spi_ps_in_control_1;
+ unsigned i, exports_ps, num_cout, spi_ps_in_control_0, spi_input_z, spi_ps_in_control_1;
int pos_index = -1, face_index = -1;
- /* clear previous register */
rstate->nregs = 0;
for (i = 0; i < rshader->ninput; i++) {
- tmp = S_028644_SEMANTIC(r600_find_vs_semantic_index(&rctx->vs_shader->shader, rshader, i));
- if (rshader->input[i].centroid)
- tmp |= S_028644_SEL_CENTROID(1);
- if (rshader->input[i].interpolate == TGSI_INTERPOLATE_LINEAR)
- tmp |= S_028644_SEL_LINEAR(1);
-
if (rshader->input[i].name == TGSI_SEMANTIC_POSITION)
pos_index = i;
- if (rshader->input[i].name == TGSI_SEMANTIC_COLOR ||
- rshader->input[i].name == TGSI_SEMANTIC_BCOLOR ||
- rshader->input[i].name == TGSI_SEMANTIC_POSITION) {
- tmp |= S_028644_FLAT_SHADE(rshader->flat_shade);
- }
if (rshader->input[i].name == TGSI_SEMANTIC_FACE)
face_index = i;
- if (rshader->input[i].name == TGSI_SEMANTIC_GENERIC &&
- rctx->sprite_coord_enable & (1 << rshader->input[i].sid)) {
- tmp |= S_028644_PT_SPRITE_TEX(1);
- }
- r600_pipe_state_add_reg(rstate, R_028644_SPI_PS_INPUT_CNTL_0 + i * 4, tmp, 0xFFFFFFFF, NULL);
}
+
for (i = 0; i < rshader->noutput; i++) {
if (rshader->output[i].name == TGSI_SEMANTIC_POSITION)
r600_pipe_state_add_reg(rstate,
@@ -210,22 +187,13 @@ static void r600_pipe_shader_ps(struct pipe_context *ctx, struct r600_pipe_shade
0xFFFFFFFF, NULL);
}
-static int r600_pipe_shader(struct pipe_context *ctx, struct r600_pipe_shader *shader)
+int r600_pipe_shader(struct pipe_context *ctx, struct r600_pipe_shader *shader)
{
struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
struct r600_shader *rshader = &shader->shader;
void *ptr;
/* copy new shader */
- if (rshader->processor_type == TGSI_PROCESSOR_VERTEX && shader->bo_fetch == NULL) {
- shader->bo_fetch = r600_bo(rctx->radeon, rshader->bc_fetch.ndw * 4, 4096, 0, 0);
- if (shader->bo_fetch == NULL) {
- return -ENOMEM;
- }
- ptr = r600_bo_map(rctx->radeon, shader->bo_fetch, 0, NULL);
- memcpy(ptr, rshader->bc_fetch.bytecode, rshader->bc_fetch.ndw * 4);
- r600_bo_unmap(rctx->radeon, shader->bo_fetch);
- }
if (shader->bo == NULL) {
shader->bo = r600_bo(rctx->radeon, rshader->bc.ndw * 4, 4096, 0, 0);
if (shader->bo == NULL) {
@@ -236,7 +204,6 @@ static int r600_pipe_shader(struct pipe_context *ctx, struct r600_pipe_shader *s
r600_bo_unmap(rctx->radeon, shader->bo);
}
/* build state */
- rshader->flat_shade = rctx->flatshade;
switch (rshader->processor_type) {
case TGSI_PROCESSOR_VERTEX:
if (rshader->family >= CHIP_CEDAR) {
@@ -255,116 +222,51 @@ static int r600_pipe_shader(struct pipe_context *ctx, struct r600_pipe_shader *s
default:
return -EINVAL;
}
- r600_context_pipe_state_set(&rctx->ctx, &shader->rstate);
return 0;
}
-static int r600_shader_update(struct pipe_context *ctx, struct r600_pipe_shader *rshader)
-{
- struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
- struct r600_shader *shader = &rshader->shader;
- const struct util_format_description *desc;
- enum pipe_format resource_format[160];
- unsigned i, nresources = 0;
- struct r600_bc *bc = &shader->bc_fetch;
- struct r600_bc_cf *cf;
- struct r600_bc_vtx *vtx;
-
- if (shader->processor_type != TGSI_PROCESSOR_VERTEX)
- return 0;
- /* doing a full memcmp fell over the refcount */
- if ((rshader->vertex_elements.count == rctx->vertex_elements->count) &&
- (!memcmp(&rshader->vertex_elements.elements, &rctx->vertex_elements->elements,
- rctx->vertex_elements->count * sizeof(struct pipe_vertex_element)))) {
- return 0;
- }
- rshader->vertex_elements = *rctx->vertex_elements;
- for (i = 0; i < rctx->vertex_elements->count; i++) {
- resource_format[nresources++] = rctx->vertex_elements->hw_format[i];
- }
- r600_bo_reference(rctx->radeon, &rshader->bo_fetch, NULL);
- LIST_FOR_EACH_ENTRY(cf, &bc->cf, list) {
- switch (cf->inst) {
- case V_SQ_CF_WORD1_SQ_CF_INST_VTX:
- case V_SQ_CF_WORD1_SQ_CF_INST_VTX_TC:
- LIST_FOR_EACH_ENTRY(vtx, &cf->vtx, list) {
- desc = util_format_description(resource_format[vtx->buffer_id]);
- if (desc == NULL) {
- R600_ERR("unknown format %d\n", resource_format[vtx->buffer_id]);
- return -EINVAL;
- }
- vtx->dst_sel_x = desc->swizzle[0];
- vtx->dst_sel_y = desc->swizzle[1];
- vtx->dst_sel_z = desc->swizzle[2];
- vtx->dst_sel_w = desc->swizzle[3];
- }
- break;
- default:
- break;
- }
- }
- return r600_bc_build(&shader->bc_fetch);
-}
-
-int r600_pipe_shader_update(struct pipe_context *ctx, struct r600_pipe_shader *shader)
-{
- struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
- int r;
-
- if (shader == NULL)
- return -EINVAL;
- /* there should be enough input */
- if (rctx->vertex_elements->count < shader->shader.bc.nresource) {
- R600_ERR("%d resources provided, expecting %d\n",
- rctx->vertex_elements->count, shader->shader.bc.nresource);
- return -EINVAL;
- }
- r = r600_shader_update(ctx, shader);
- if (r)
- return r;
- return r600_pipe_shader(ctx, shader);
-}
-
-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;
-//fprintf(stderr, "--------------------------------------------------------------\n");
-//tgsi_dump(tokens, 0);
+ /* Would like some magic "get_bool_option_once" routine.
+ */
+ if (dump_shaders == -1)
+ dump_shaders = debug_get_bool_option("R600_DUMP_SHADERS", FALSE);
+
+ if (dump_shaders) {
+ fprintf(stderr, "--------------------------------------------------------------\n");
+ 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;
}
- if (shader->shader.processor_type == TGSI_PROCESSOR_VERTEX) {
- r = r600_bc_build(&shader->shader.bc_fetch);
- if (r) {
- R600_ERR("building bytecode failed !\n");
- return r;
- }
+ if (dump_shaders) {
+ r600_bc_dump(&shader->shader.bc);
+ fprintf(stderr, "______________________________________________________________\n");
}
-//fprintf(stderr, "______________________________________________________________\n");
- return 0;
+ return r600_pipe_shader(ctx, shader);
}
-void
-r600_pipe_shader_destroy(struct pipe_context *ctx, struct r600_pipe_shader *shader)
+void r600_pipe_shader_destroy(struct pipe_context *ctx, struct r600_pipe_shader *shader)
{
struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
r600_bo_reference(rctx->radeon, &shader->bo, NULL);
-
r600_bc_clear(&shader->shader.bc);
-
- /* FIXME: is there more stuff to free? */
}
/*
@@ -381,9 +283,7 @@ struct r600_shader_ctx {
unsigned temp_reg;
struct r600_shader_tgsi_instruction *inst_info;
struct r600_bc *bc;
- struct r600_bc *bc_fetch;
struct r600_shader *shader;
- u32 value[4];
u32 *literals;
u32 nliterals;
u32 max_driver_temp_used;
@@ -501,9 +401,7 @@ static int evergreen_interp_alu(struct r600_shader_ctx *ctx, int input)
static int tgsi_declaration(struct r600_shader_ctx *ctx)
{
struct tgsi_full_declaration *d = &ctx->parse.FullToken.FullDeclaration;
- struct r600_bc_vtx vtx;
unsigned i;
- int r;
switch (d->Declaration.File) {
case TGSI_FILE_INPUT:
@@ -513,26 +411,6 @@ static int tgsi_declaration(struct r600_shader_ctx *ctx)
ctx->shader->input[i].interpolate = d->Declaration.Interpolate;
ctx->shader->input[i].centroid = d->Declaration.Centroid;
ctx->shader->input[i].gpr = ctx->file_offset[TGSI_FILE_INPUT] + i;
- if (ctx->type == TGSI_PROCESSOR_VERTEX) {
- /* turn input into fetch */
- memset(&vtx, 0, sizeof(struct r600_bc_vtx));
- vtx.inst = 0;
- vtx.fetch_type = 0;
- vtx.buffer_id = i;
- /* register containing the index into the buffer */
- vtx.src_gpr = 0;
- vtx.src_sel_x = 0;
- vtx.mega_fetch_count = 0x1F;
- vtx.dst_gpr = ctx->shader->input[i].gpr;
- vtx.dst_sel_x = 0;
- vtx.dst_sel_y = 1;
- vtx.dst_sel_z = 2;
- vtx.dst_sel_w = 3;
- vtx.use_const_fields = 1;
- r = r600_bc_add_vtx(ctx->bc_fetch, &vtx);
- if (r)
- return r;
- }
if (ctx->type == TGSI_PROCESSOR_FRAGMENT && ctx->bc->chiprev == CHIPREV_EVERGREEN) {
/* turn input into interpolate on EG */
if (ctx->shader->input[i].name != TGSI_SEMANTIC_POSITION) {
@@ -614,7 +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;
@@ -624,7 +502,6 @@ int r600_shader_from_tgsi(const struct tgsi_token *tokens, struct r600_shader *s
int i, r = 0, pos0;
ctx.bc = &shader->bc;
- ctx.bc_fetch = &shader->bc_fetch;
ctx.shader = shader;
r = r600_bc_init(ctx.bc, shader->family);
if (r)
@@ -634,19 +511,15 @@ int r600_shader_from_tgsi(const struct tgsi_token *tokens, struct r600_shader *s
tgsi_parse_init(&ctx.parse, tokens);
ctx.type = ctx.parse.FullHeader.Processor.Processor;
shader->processor_type = ctx.type;
- if (shader->processor_type == TGSI_PROCESSOR_VERTEX) {
- r = r600_bc_init(ctx.bc_fetch, shader->family);
- if (r)
- return r;
- ctx.bc_fetch->type = -1;
- }
ctx.bc->type = shader->processor_type;
/* register allocations */
/* Values [0,127] correspond to GPR[0..127].
* Values [128,159] correspond to constant buffer bank 0
* Values [160,191] correspond to constant buffer bank 1
- * Values [256,511] correspond to cfile constants c[0..255].
+ * Values [256,511] correspond to cfile constants c[0..255]. (Gone on EG)
+ * Values [256,287] correspond to constant buffer bank 2 (EG)
+ * Values [288,319] correspond to constant buffer bank 3 (EG)
* Other special values are shown in the list below.
* 244 ALU_SRC_1_DBL_L: special constant 1.0 double-float, LSW. (RV670+)
* 245 ALU_SRC_1_DBL_M: special constant 1.0 double-float, MSW. (RV670+)
@@ -680,9 +553,11 @@ int r600_shader_from_tgsi(const struct tgsi_token *tokens, struct r600_shader *s
ctx.file_offset[TGSI_FILE_TEMPORARY] = ctx.file_offset[TGSI_FILE_OUTPUT] +
ctx.info.file_count[TGSI_FILE_OUTPUT];
- ctx.file_offset[TGSI_FILE_CONSTANT] = 128;
+ /* Outside the GPR range. This will be translated to one of the
+ * kcache banks later. */
+ ctx.file_offset[TGSI_FILE_CONSTANT] = 512;
- ctx.file_offset[TGSI_FILE_IMMEDIATE] = 253;
+ ctx.file_offset[TGSI_FILE_IMMEDIATE] = V_SQ_ALU_SRC_LITERAL;
ctx.temp_reg = ctx.file_offset[TGSI_FILE_TEMPORARY] +
ctx.info.file_count[TGSI_FILE_TEMPORARY];
@@ -725,9 +600,8 @@ 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;
default:
R600_ERR("unsupported token type %d\n", ctx.parse.FullToken.Token.Type);
@@ -840,21 +714,13 @@ int r600_shader_from_tgsi(const struct tgsi_token *tokens, struct r600_shader *s
output[i].inst = BC_INST(ctx.bc, V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_EXPORT_DONE);
}
}
- /* add return to fetch shader */
- if (ctx.type == TGSI_PROCESSOR_VERTEX) {
- if (ctx.bc->chiprev == CHIPREV_EVERGREEN) {
- r600_bc_add_cfinst(ctx.bc_fetch, EG_V_SQ_CF_WORD1_SQ_CF_INST_RETURN);
- } else {
- r600_bc_add_cfinst(ctx.bc_fetch, V_SQ_CF_WORD1_SQ_CF_INST_RETURN);
- }
- }
/* add output to bytecode */
for (i = 0; i < noutput; i++) {
r = r600_bc_add_output(ctx.bc, &output[i]);
if (r)
goto out_err;
}
- free(ctx.literals);
+ *literals = ctx.literals;
tgsi_parse_free(&ctx.parse);
return 0;
out_err:
@@ -878,22 +744,29 @@ static int tgsi_src(struct r600_shader_ctx *ctx,
const struct tgsi_full_src_register *tgsi_src,
struct r600_bc_alu_src *r600_src)
{
- int index;
memset(r600_src, 0, sizeof(struct r600_bc_alu_src));
- r600_src->sel = tgsi_src->Register.Index;
- if (tgsi_src->Register.File == TGSI_FILE_IMMEDIATE) {
- r600_src->sel = 0;
- index = tgsi_src->Register.Index;
- 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];
- }
- if (tgsi_src->Register.Indirect)
- r600_src->rel = V_SQ_REL_RELATIVE;
r600_src->neg = tgsi_src->Register.Negate;
r600_src->abs = tgsi_src->Register.Absolute;
- r600_src->sel += ctx->file_offset[tgsi_src->Register.File];
+ if (tgsi_src->Register.File == TGSI_FILE_IMMEDIATE) {
+ int index;
+ if((tgsi_src->Register.SwizzleX == tgsi_src->Register.SwizzleY) &&
+ (tgsi_src->Register.SwizzleX == tgsi_src->Register.SwizzleZ) &&
+ (tgsi_src->Register.SwizzleX == tgsi_src->Register.SwizzleW)) {
+
+ index = tgsi_src->Register.Index * 4 + tgsi_src->Register.SwizzleX;
+ r600_bc_special_constants(ctx->literals[index], &r600_src->sel, &r600_src->neg);
+ if (r600_src->sel != V_SQ_ALU_SRC_LITERAL)
+ return 0;
+ }
+ index = tgsi_src->Register.Index;
+ r600_src->sel = V_SQ_ALU_SRC_LITERAL;
+ r600_src->value = ctx->literals + index * 4;
+ } else {
+ if (tgsi_src->Register.Indirect)
+ r600_src->rel = V_SQ_REL_RELATIVE;
+ r600_src->sel = tgsi_src->Register.Index;
+ r600_src->sel += ctx->file_offset[tgsi_src->Register.File];
+ }
return 0;
}
@@ -981,18 +854,19 @@ static int tgsi_split_literal_constant(struct r600_shader_ctx *ctx, struct r600_
int i, j, k, nliteral, r;
for (i = 0, nliteral = 0; i < inst->Instruction.NumSrcRegs; i++) {
- if (inst->Src[i].Register.File == TGSI_FILE_IMMEDIATE) {
+ if (r600_src[i].sel == V_SQ_ALU_SRC_LITERAL) {
nliteral++;
}
}
for (i = 0, j = nliteral - 1; i < inst->Instruction.NumSrcRegs; i++) {
- if (j > 0 && inst->Src[i].Register.File == TGSI_FILE_IMMEDIATE) {
+ if (j > 0 && r600_src[i].sel == V_SQ_ALU_SRC_LITERAL) {
int treg = r600_get_temp(ctx);
for (k = 0; k < 4; k++) {
memset(&alu, 0, sizeof(struct r600_bc_alu));
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;
@@ -1002,9 +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--;
}
@@ -1012,19 +883,25 @@ static int tgsi_split_literal_constant(struct r600_shader_ctx *ctx, struct r600_
return 0;
}
-static int tgsi_op2_s(struct r600_shader_ctx *ctx, int swap)
+static int tgsi_last_instruction(unsigned writemask)
{
- struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction;
- struct r600_bc_alu_src r600_src[3];
- struct r600_bc_alu alu;
- int i, j, r;
- int lasti = 0;
+ int i, lasti = 0;
for (i = 0; i < 4; i++) {
- if (inst->Dst[0].Register.WriteMask & (1 << i)) {
+ if (writemask & (1 << i)) {
lasti = i;
}
}
+ return lasti;
+}
+
+static int tgsi_op2_s(struct r600_shader_ctx *ctx, int swap)
+{
+ struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction;
+ struct r600_bc_alu_src r600_src[3];
+ struct r600_bc_alu alu;
+ int i, j, r;
+ int lasti = tgsi_last_instruction(inst->Dst[0].Register.WriteMask);
r = tgsi_split_constant(ctx, r600_src);
if (r)
@@ -1093,12 +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;
- uint32_t lit_vals[4];
struct r600_bc_alu alu;
- memset(lit_vals, 0, 4*4);
r = tgsi_split_constant(ctx, r600_src);
if (r)
return r;
@@ -1106,9 +985,6 @@ static int tgsi_setup_trig(struct r600_shader_ctx *ctx,
if (r)
return r;
- 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;
@@ -1122,15 +998,13 @@ static int tgsi_setup_trig(struct r600_shader_ctx *ctx,
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);
@@ -1146,14 +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;
@@ -1169,13 +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;
}
@@ -1185,7 +1058,7 @@ static int tgsi_trig(struct r600_shader_ctx *ctx)
struct r600_bc_alu_src r600_src[3];
struct r600_bc_alu alu;
int i, r;
- int lasti = 0;
+ int lasti = tgsi_last_instruction(inst->Dst[0].Register.WriteMask);
r = tgsi_setup_trig(ctx, r600_src);
if (r)
@@ -1205,10 +1078,6 @@ static int tgsi_trig(struct r600_shader_ctx *ctx)
return r;
/* replicate result */
- for (i = 0; i < 4; i++) {
- if (inst->Dst[0].Register.WriteMask & (1 << i))
- lasti = i;
- }
for (i = 0; i < lasti + 1; i++) {
if (!(inst->Dst[0].Register.WriteMask & (1 << i)))
continue;
@@ -1295,10 +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; */
@@ -1319,10 +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;
@@ -1358,9 +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;
@@ -1423,10 +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;
@@ -1445,10 +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;
@@ -1471,9 +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);
@@ -1517,9 +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);
}
@@ -1568,9 +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);
}
@@ -1594,12 +1435,9 @@ 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_IEEE);
+ alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MUL);
r = tgsi_src(ctx, &inst->Src[1], &alu.src[0]);
if (r)
return r;
@@ -1611,9 +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);
@@ -1624,9 +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);
}
@@ -1666,9 +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++) {
@@ -1703,9 +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))) {
@@ -1735,6 +1561,7 @@ static int tgsi_op3(struct r600_shader_ctx *ctx)
struct r600_bc_alu_src r600_src[3];
struct r600_bc_alu alu;
int i, j, r;
+ int lasti = tgsi_last_instruction(inst->Dst[0].Register.WriteMask);
r = tgsi_split_constant(ctx, r600_src);
if (r)
@@ -1742,26 +1569,32 @@ static int tgsi_op3(struct r600_shader_ctx *ctx)
r = tgsi_split_literal_constant(ctx, r600_src);
if (r)
return r;
- /* do it in 2 step as op3 doesn't support writemask */
- for (i = 0; i < 4; i++) {
+ for (i = 0; i < lasti + 1; i++) {
+ if (!(inst->Dst[0].Register.WriteMask & (1 << i)))
+ continue;
+
memset(&alu, 0, sizeof(struct r600_bc_alu));
alu.inst = ctx->inst_info->r600_opcode;
for (j = 0; j < inst->Instruction.NumSrcRegs; j++) {
alu.src[j] = r600_src[j];
alu.src[j].chan = tgsi_chan(&inst->Src[j], i);
}
- alu.dst.sel = ctx->temp_reg;
+
+ r = tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst);
+ if (r)
+ return r;
+
alu.dst.chan = i;
alu.dst.write = 1;
alu.is_op3 = 1;
- if (i == 3) {
+ if (i == lasti) {
alu.last = 1;
}
r = r600_bc_add_alu(ctx->bc, &alu);
if (r)
return r;
}
- return tgsi_helper_copy(ctx, inst);
+ return 0;
}
static int tgsi_dp(struct r600_shader_ctx *ctx)
@@ -1784,9 +1617,13 @@ static int tgsi_dp(struct r600_shader_ctx *ctx)
alu.src[j] = r600_src[j];
alu.src[j].chan = tgsi_chan(&inst->Src[j], i);
}
- alu.dst.sel = ctx->temp_reg;
+
+ r = tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst);
+ if (r)
+ return r;
+
alu.dst.chan = i;
- alu.dst.write = 1;
+ alu.dst.write = (inst->Dst[0].Register.WriteMask >> i) & 1;
/* handle some special cases */
switch (ctx->inst_info->tgsi_opcode) {
case TGSI_OPCODE_DP2:
@@ -1818,19 +1655,21 @@ static int tgsi_dp(struct r600_shader_ctx *ctx)
if (r)
return r;
}
- return tgsi_helper_copy(ctx, inst);
+ return 0;
}
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;
unsigned src_gpr;
int r, i;
int opcode;
- boolean src_not_temp = inst->Src[0].Register.File != TGSI_FILE_TEMPORARY;
- uint32_t lit_vals[4];
+ boolean src_not_temp =
+ inst->Src[0].Register.File != TGSI_FILE_TEMPORARY &&
+ inst->Src[0].Register.File != TGSI_FILE_INPUT;
src_gpr = ctx->file_offset[inst->Src[0].Register.File] + inst->Src[0].Register.Index;
@@ -1959,6 +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;
@@ -1979,6 +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;
@@ -1989,11 +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;
}
@@ -2066,6 +1902,7 @@ static int tgsi_lrp(struct r600_shader_ctx *ctx)
struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction;
struct r600_bc_alu_src r600_src[3];
struct r600_bc_alu alu;
+ int lasti = tgsi_last_instruction(inst->Dst[0].Register.WriteMask);
unsigned i;
int r;
@@ -2075,8 +1912,40 @@ static int tgsi_lrp(struct r600_shader_ctx *ctx)
r = tgsi_split_literal_constant(ctx, r600_src);
if (r)
return r;
+
+ /* optimize if it's just an equal balance */
+ if(r600_src[0].sel == V_SQ_ALU_SRC_0_5) {
+ for (i = 0; i < lasti + 1; i++) {
+ if (!(inst->Dst[0].Register.WriteMask & (1 << i)))
+ continue;
+
+ memset(&alu, 0, sizeof(struct r600_bc_alu));
+ alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_ADD);
+ alu.src[0] = r600_src[1];
+ alu.src[0].chan = tgsi_chan(&inst->Src[1], i);
+ alu.src[1] = r600_src[2];
+ alu.src[1].chan = tgsi_chan(&inst->Src[2], i);
+ alu.omod = 3;
+ r = tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst);
+ if (r)
+ return r;
+
+ alu.dst.chan = i;
+ if (i == lasti) {
+ alu.last = 1;
+ }
+ r = r600_bc_add_alu(ctx->bc, &alu);
+ if (r)
+ return r;
+ }
+ return 0;
+ }
+
/* 1 - src0 */
- for (i = 0; i < 4; i++) {
+ for (i = 0; i < lasti + 1; i++) {
+ if (!(inst->Dst[0].Register.WriteMask & (1 << i)))
+ continue;
+
memset(&alu, 0, sizeof(struct r600_bc_alu));
alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_ADD);
alu.src[0].sel = V_SQ_ALU_SRC_1;
@@ -2086,7 +1955,7 @@ static int tgsi_lrp(struct r600_shader_ctx *ctx)
alu.src[1].neg = 1;
alu.dst.sel = ctx->temp_reg;
alu.dst.chan = i;
- if (i == 3) {
+ if (i == lasti) {
alu.last = 1;
}
alu.dst.write = 1;
@@ -2094,12 +1963,12 @@ 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 < 4; i++) {
+ for (i = 0; i < lasti + 1; i++) {
+ if (!(inst->Dst[0].Register.WriteMask & (1 << i)))
+ continue;
+
memset(&alu, 0, sizeof(struct r600_bc_alu));
alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MUL);
alu.src[0].sel = ctx->temp_reg;
@@ -2108,7 +1977,7 @@ static int tgsi_lrp(struct r600_shader_ctx *ctx)
alu.src[1].chan = tgsi_chan(&inst->Src[2], i);
alu.dst.sel = ctx->temp_reg;
alu.dst.chan = i;
- if (i == 3) {
+ if (i == lasti) {
alu.last = 1;
}
alu.dst.write = 1;
@@ -2116,12 +1985,12 @@ 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 < 4; i++) {
+ for (i = 0; i < lasti + 1; i++) {
+ if (!(inst->Dst[0].Register.WriteMask & (1 << i)))
+ continue;
+
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;
@@ -2131,16 +2000,20 @@ static int tgsi_lrp(struct r600_shader_ctx *ctx)
alu.src[1].chan = tgsi_chan(&inst->Src[1], i);
alu.src[2].sel = ctx->temp_reg;
alu.src[2].chan = i;
- alu.dst.sel = ctx->temp_reg;
+
+ r = tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst);
+ if (r)
+ return r;
+
alu.dst.chan = i;
- if (i == 3) {
+ if (i == lasti) {
alu.last = 1;
}
r = r600_bc_add_alu(ctx->bc, &alu);
if (r)
return r;
}
- return tgsi_helper_copy(ctx, inst);
+ return 0;
}
static int tgsi_cmp(struct r600_shader_ctx *ctx)
@@ -2148,8 +2021,8 @@ static int tgsi_cmp(struct r600_shader_ctx *ctx)
struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction;
struct r600_bc_alu_src r600_src[3];
struct r600_bc_alu alu;
- int use_temp = 0;
int i, r;
+ int lasti = tgsi_last_instruction(inst->Dst[0].Register.WriteMask);
r = tgsi_split_constant(ctx, r600_src);
if (r)
@@ -2158,10 +2031,10 @@ static int tgsi_cmp(struct r600_shader_ctx *ctx)
if (r)
return r;
- if (inst->Dst[0].Register.WriteMask != 0xf)
- use_temp = 1;
+ for (i = 0; i < lasti + 1; i++) {
+ if (!(inst->Dst[0].Register.WriteMask & (1 << i)))
+ continue;
- for (i = 0; i < 4; i++) {
memset(&alu, 0, sizeof(struct r600_bc_alu));
alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP3_SQ_OP3_INST_CNDGE);
alu.src[0] = r600_src[0];
@@ -2173,24 +2046,19 @@ static int tgsi_cmp(struct r600_shader_ctx *ctx)
alu.src[2] = r600_src[1];
alu.src[2].chan = tgsi_chan(&inst->Src[1], i);
- if (use_temp)
- alu.dst.sel = ctx->temp_reg;
- else {
- r = tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst);
- if (r)
- return r;
- }
+ r = tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst);
+ if (r)
+ return r;
+
alu.dst.chan = i;
alu.dst.write = 1;
alu.is_op3 = 1;
- if (i == 3)
+ if (i == lasti)
alu.last = 1;
r = r600_bc_add_alu(ctx->bc, &alu);
if (r)
return r;
}
- if (use_temp)
- return tgsi_helper_copy(ctx, inst);
return 0;
}
@@ -2257,10 +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++) {
@@ -2318,10 +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);
@@ -2354,10 +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;
@@ -2369,10 +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); */
@@ -2398,9 +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);*/
@@ -2421,9 +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;*/
@@ -2441,9 +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);
}
@@ -2473,10 +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;
@@ -2489,10 +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))); */
@@ -2515,10 +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);
@@ -2534,10 +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);
@@ -2553,10 +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);
@@ -2572,10 +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);
@@ -2597,10 +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);*/
@@ -2622,10 +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; */
@@ -2644,10 +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);
@@ -2802,8 +2609,25 @@ static int emit_logic_pred(struct r600_shader_ctx *ctx, int opcode)
static int pops(struct r600_shader_ctx *ctx, int pops)
{
- r600_bc_add_cfinst(ctx->bc, CTX_INST(V_SQ_CF_WORD1_SQ_CF_INST_POP));
- ctx->bc->cf_last->pop_count = pops;
+ int alu_pop = 3;
+ if (ctx->bc->cf_last) {
+ if (ctx->bc->cf_last->inst == CTX_INST(V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU) << 3)
+ alu_pop = 0;
+ else if (ctx->bc->cf_last->inst == CTX_INST(V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_POP_AFTER) << 3)
+ alu_pop = 1;
+ }
+ alu_pop += pops;
+ if (alu_pop == 1) {
+ ctx->bc->cf_last->inst = CTX_INST(V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_POP_AFTER) << 3;
+ ctx->bc->force_add_cf = 1;
+ } else if (alu_pop == 2) {
+ ctx->bc->cf_last->inst = CTX_INST(V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_POP2_AFTER) << 3;
+ ctx->bc->force_add_cf = 1;
+ } else {
+ r600_bc_add_cfinst(ctx->bc, CTX_INST(V_SQ_CF_WORD1_SQ_CF_INST_POP));
+ ctx->bc->cf_last->pop_count = pops;
+ ctx->bc->cf_last->cf_addr = ctx->bc->cf_last->id + 2;
+ }
return 0;
}
generated by cgit v1.1 at 2025-08-17 00:44:51 +0000