u32 value[4];
u32 *literals;
u32 nliterals;
+ u32 max_driver_temp_used;
};
struct r600_shader_tgsi_instruction {
struct radeon_state *state;
unsigned i, tmp;
- rpshader->rstate = radeon_state_decref(rpshader->rstate);
- state = radeon_state_shader(rscreen->rw, R600_STATE_SHADER, 0, R600_SHADER_VS);
- if (state == NULL)
- return -ENOMEM;
+ state = &rpshader->rstate[0];
+ radeon_state_fini(&rpshader->rstate[0]);
+ radeon_state_init(state, rscreen->rw, R600_STATE_SHADER, 0, R600_SHADER_VS);
for (i = 0; i < 10; i++) {
state->states[R600_VS_SHADER__SPI_VS_OUT_ID_0 + i] = 0;
}
state->states[R600_VS_SHADER__SPI_VS_OUT_CONFIG] = S_0286C4_VS_EXPORT_COUNT(rshader->noutput - 2);
state->states[R600_VS_SHADER__SQ_PGM_RESOURCES_VS] = S_028868_NUM_GPRS(rshader->bc.ngpr) |
S_028868_STACK_SIZE(rshader->bc.nstack);
- rpshader->rstate = state;
- rpshader->rstate->bo[0] = radeon_bo_incref(rscreen->rw, rpshader->bo);
- rpshader->rstate->bo[1] = radeon_bo_incref(rscreen->rw, rpshader->bo);
- rpshader->rstate->nbo = 2;
- rpshader->rstate->placement[0] = RADEON_GEM_DOMAIN_GTT;
- rpshader->rstate->placement[2] = RADEON_GEM_DOMAIN_GTT;
+ state->bo[0] = radeon_bo_incref(rscreen->rw, rpshader->bo);
+ state->bo[1] = radeon_bo_incref(rscreen->rw, rpshader->bo);
+ state->nbo = 2;
+ state->placement[0] = RADEON_GEM_DOMAIN_GTT;
+ state->placement[2] = RADEON_GEM_DOMAIN_GTT;
return radeon_state_pm4(state);
}
struct r600_context *rctx = r600_context(ctx);
struct radeon_state *state;
unsigned i, tmp, exports_ps, num_cout;
+ boolean have_pos = FALSE;
+ state = &rpshader->rstate[0];
rasterizer = &rctx->rasterizer->state.rasterizer;
- rpshader->rstate = radeon_state_decref(rpshader->rstate);
- state = radeon_state_shader(rscreen->rw, R600_STATE_SHADER, 0, R600_SHADER_PS);
- if (state == NULL)
- return -ENOMEM;
+ radeon_state_fini(state);
+ radeon_state_init(state, rscreen->rw, R600_STATE_SHADER, 0, R600_SHADER_PS);
for (i = 0; i < rshader->ninput; i++) {
tmp = S_028644_SEMANTIC(i);
tmp |= S_028644_SEL_CENTROID(1);
+ if (rshader->input[i].name == TGSI_SEMANTIC_POSITION)
+ have_pos = TRUE;
if (rshader->input[i].name == TGSI_SEMANTIC_COLOR ||
- rshader->input[i].name == TGSI_SEMANTIC_BCOLOR) {
+ rshader->input[i].name == TGSI_SEMANTIC_BCOLOR ||
+ rshader->input[i].name == TGSI_SEMANTIC_POSITION) {
tmp |= S_028644_FLAT_SHADE(rshader->flat_shade);
}
if (rasterizer->sprite_coord_enable & (1 << i)) {
}
state->states[R600_PS_SHADER__SPI_PS_IN_CONTROL_0] = S_0286CC_NUM_INTERP(rshader->ninput) |
S_0286CC_PERSP_GRADIENT_ENA(1);
+ if (have_pos) {
+ state->states[R600_PS_SHADER__SPI_PS_IN_CONTROL_0] |= S_0286CC_POSITION_ENA(1) |
+ S_0286CC_BARYC_SAMPLE_CNTL(1);
+ state->states[R600_PS_SHADER__SPI_INPUT_Z] |= 1;
+ }
state->states[R600_PS_SHADER__SPI_PS_IN_CONTROL_1] = 0x00000000;
state->states[R600_PS_SHADER__SQ_PGM_RESOURCES_PS] = S_028868_NUM_GPRS(rshader->bc.ngpr) |
S_028868_STACK_SIZE(rshader->bc.nstack);
state->states[R600_PS_SHADER__SQ_PGM_EXPORTS_PS] = exports_ps;
- rpshader->rstate = state;
- rpshader->rstate->bo[0] = radeon_bo_incref(rscreen->rw, rpshader->bo);
- rpshader->rstate->nbo = 1;
- rpshader->rstate->placement[0] = RADEON_GEM_DOMAIN_GTT;
+ state->bo[0] = radeon_bo_incref(rscreen->rw, rpshader->bo);
+ state->nbo = 1;
+ state->placement[0] = RADEON_GEM_DOMAIN_GTT;
return radeon_state_pm4(state);
}
return 0;
}
+static int r600_get_temp(struct r600_shader_ctx *ctx)
+{
+ return ctx->temp_reg + ctx->max_driver_temp_used++;
+}
+
int r600_shader_from_tgsi(const struct tgsi_token *tokens, struct r600_shader *shader)
{
struct tgsi_full_immediate *immediate;
r = tgsi_is_supported(&ctx);
if (r)
goto out_err;
+ ctx.max_driver_temp_used = 0;
+ /* reserve first tmp for everyone */
+ r600_get_temp(&ctx);
opcode = ctx.parse.FullToken.FullInstruction.Instruction.Opcode;
ctx.inst_info = &r600_shader_tgsi_instruction[opcode];
r = ctx.inst_info->process(&ctx);
output[i].type = V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PIXEL;
} else if (shader->output[i].name == TGSI_SEMANTIC_POSITION) {
output[i].array_base = 61;
+ output[i].swizzle_x = 2;
+ output[i].swizzle_y = output[i].swizzle_z = output[i].swizzle_w = 7;
output[i].type = V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PIXEL;
} else {
R600_ERR("unsupported fragment output name %d\n", shader->output[i].name);
}
for (i = 0, j = nconst - 1; i < inst->Instruction.NumSrcRegs; i++) {
if (inst->Src[j].Register.File == TGSI_FILE_CONSTANT && j > 0) {
+ int treg = r600_get_temp(ctx);
for (k = 0; k < 4; k++) {
memset(&alu, 0, sizeof(struct r600_bc_alu));
alu.inst = V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV;
alu.src[0].sel = r600_src[j].sel;
alu.src[0].chan = k;
- alu.dst.sel = ctx->temp_reg + j;
+ alu.dst.sel = treg;
alu.dst.chan = k;
alu.dst.write = 1;
if (k == 3)
if (r)
return r;
}
- r600_src[j].sel = ctx->temp_reg + j;
+ r600_src[j].sel = treg;
j--;
}
}
return 0;
}
+/* need to move any immediate into a temp - for trig functions which use literal for PI stuff */
+static int tgsi_split_literal_constant(struct r600_shader_ctx *ctx, struct r600_bc_alu_src r600_src[3])
+{
+ struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction;
+ struct r600_bc_alu alu;
+ 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) {
+ nliteral++;
+ }
+ }
+ for (i = 0, j = 0; i < inst->Instruction.NumSrcRegs; i++) {
+ if (inst->Src[j].Register.File == TGSI_FILE_IMMEDIATE) {
+ int treg = r600_get_temp(ctx);
+ for (k = 0; k < 4; k++) {
+ memset(&alu, 0, sizeof(struct r600_bc_alu));
+ alu.inst = V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV;
+ alu.src[0].sel = r600_src[j].sel;
+ alu.src[0].chan = k;
+ alu.dst.sel = treg;
+ alu.dst.chan = k;
+ alu.dst.write = 1;
+ if (k == 3)
+ alu.last = 1;
+ 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;
+ r600_src[j].sel = treg;
+ j++;
+ }
+ }
+ return 0;
+}
+
static int tgsi_op2_s(struct r600_shader_ctx *ctx, int swap)
{
struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction;
r = tgsi_split_constant(ctx, r600_src);
if (r)
return r;
+
+ r = tgsi_split_literal_constant(ctx, r600_src);
+ if (r)
+ return r;
+
lit_vals[0] = fui(1.0 /(3.1415926535 * 2));
lit_vals[1] = fui(0.5f);
struct r600_bc_alu_src r600_src[3];
struct r600_bc_alu alu;
int i, r;
-
- r = tgsi_split_constant(ctx, r600_src);
- if (r)
- return r;
+ int lasti = 0;
r = tgsi_setup_trig(ctx, r600_src);
if (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;
+
memset(&alu, 0, sizeof(struct r600_bc_alu));
- alu.src[0].sel = ctx->temp_reg;
alu.inst = V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV;
- alu.dst.chan = i;
+
+ alu.src[0].sel = ctx->temp_reg;
r = tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst);
if (r)
return r;
- alu.dst.write = (inst->Dst[0].Register.WriteMask >> i) & 1;
- if (i == 3)
+ if (i == lasti)
alu.last = 1;
r = r600_bc_add_alu(ctx->bc, &alu);
if (r)
struct r600_bc_alu alu;
int r;
- r = tgsi_split_constant(ctx, r600_src);
- if (r)
- return r;
-
r = tgsi_setup_trig(ctx, r600_src);
if (r)
return r;
{
struct tgsi_full_instruction *inst = &ctx->parse.FullToken.FullInstruction;
struct r600_bc_alu alu;
+ struct r600_bc_alu_src r600_src[3];
int r;
+ r = tgsi_split_constant(ctx, r600_src);
+ if (r)
+ return r;
+ r = tgsi_split_literal_constant(ctx, r600_src);
+ if (r)
+ return r;
+
/* dst.x, <- 1.0 */
memset(&alu, 0, sizeof(struct r600_bc_alu));
alu.inst = V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV;
/* dst.y = max(src.x, 0.0) */
memset(&alu, 0, sizeof(struct r600_bc_alu));
alu.inst = V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MAX;
- r = tgsi_src(ctx, &inst->Src[0], &alu.src[0]);
- if (r)
- return r;
+ alu.src[0] = r600_src[0];
alu.src[1].sel = V_SQ_ALU_SRC_0; /*0.0*/
alu.src[1].chan = 0;
r = tgsi_dst(ctx, &inst->Dst[0], 1, &alu.dst);
if (r)
return r;
- /* dst.z = NOP - fill Z slot */
- memset(&alu, 0, sizeof(struct r600_bc_alu));
- alu.inst = V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP;
- alu.dst.chan = 2;
- r = r600_bc_add_alu(ctx->bc, &alu);
- if (r)
- return r;
-
/* dst.w, <- 1.0 */
memset(&alu, 0, sizeof(struct r600_bc_alu));
alu.inst = V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV;
/* dst.z = log(src.y) */
memset(&alu, 0, sizeof(struct r600_bc_alu));
alu.inst = V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_LOG_CLAMPED;
- r = tgsi_src(ctx, &inst->Src[0], &alu.src[0]);
- if (r)
- return r;
- alu.src[0].chan = tgsi_chan(&inst->Src[0], 3);
+ alu.src[0] = r600_src[0];
+ alu.src[0].chan = tgsi_chan(&inst->Src[0], 1);
r = tgsi_dst(ctx, &inst->Dst[0], 2, &alu.dst);
if (r)
return r;
/* tmp.x = amd MUL_LIT(src.w, dst.z, src.x ) */
memset(&alu, 0, sizeof(struct r600_bc_alu));
alu.inst = V_SQ_ALU_WORD1_OP3_SQ_OP3_INST_MUL_LIT;
- r = tgsi_src(ctx, &inst->Src[0], &alu.src[0]);
- if (r)
- return r;
+ alu.src[0] = r600_src[0];
alu.src[0].chan = tgsi_chan(&inst->Src[0], 3);
alu.src[1].sel = sel;
alu.src[1].chan = chan;
- r = tgsi_src(ctx, &inst->Src[0], &alu.src[2]);
- if (r)
- return r;
+
+ alu.src[2] = r600_src[0];
alu.src[2].chan = tgsi_chan(&inst->Src[0], 0);
alu.dst.sel = ctx->temp_reg;
alu.dst.chan = 0;
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];
src_gpr = ctx->file_offset[inst->Src[0].Register.File] + inst->Src[0].Register.Index;
/* Add perspective divide */
memset(&alu, 0, sizeof(struct r600_bc_alu));
alu.inst = V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_RECIP_IEEE;
- alu.src[0].sel = src_gpr;
+ r = tgsi_src(ctx, &inst->Src[0], &alu.src[0]);
+ if (r)
+ return r;
+
alu.src[0].chan = tgsi_chan(&inst->Src[0], 3);
alu.dst.sel = ctx->temp_reg;
alu.dst.chan = 3;
alu.inst = V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MUL;
alu.src[0].sel = ctx->temp_reg;
alu.src[0].chan = 3;
- alu.src[1].sel = src_gpr;
+ r = tgsi_src(ctx, &inst->Src[0], &alu.src[1]);
+ if (r)
+ return r;
alu.src[1].chan = tgsi_chan(&inst->Src[0], i);
alu.dst.sel = ctx->temp_reg;
alu.dst.chan = i;
r = r600_bc_add_alu(ctx->bc, &alu);
if (r)
return r;
+ src_not_temp = false;
src_gpr = ctx->temp_reg;
- } else if (inst->Src[0].Register.File != TGSI_FILE_TEMPORARY) {
+ }
+
+ if (inst->Texture.Texture == TGSI_TEXTURE_CUBE) {
+ int src_chan, src2_chan;
+
+ /* tmp1.xyzw = CUBE(R0.zzxy, R0.yxzz) */
+ for (i = 0; i < 4; i++) {
+ memset(&alu, 0, sizeof(struct r600_bc_alu));
+ alu.inst = V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_CUBE;
+ switch (i) {
+ case 0:
+ src_chan = 2;
+ src2_chan = 1;
+ break;
+ case 1:
+ src_chan = 2;
+ src2_chan = 0;
+ break;
+ case 2:
+ src_chan = 0;
+ src2_chan = 2;
+ break;
+ case 3:
+ src_chan = 1;
+ src2_chan = 2;
+ break;
+ }
+ r = tgsi_src(ctx, &inst->Src[0], &alu.src[0]);
+ if (r)
+ return r;
+ alu.src[0].chan = tgsi_chan(&inst->Src[0], src_chan);
+ r = tgsi_src(ctx, &inst->Src[0], &alu.src[1]);
+ if (r)
+ return r;
+ alu.src[1].chan = tgsi_chan(&inst->Src[0], src2_chan);
+ alu.dst.sel = ctx->temp_reg;
+ alu.dst.chan = i;
+ if (i == 3)
+ alu.last = 1;
+ alu.dst.write = 1;
+ r = r600_bc_add_alu(ctx->bc, &alu);
+ if (r)
+ return r;
+ }
+
+ /* tmp1.z = RCP_e(|tmp1.z|) */
+ memset(&alu, 0, sizeof(struct r600_bc_alu));
+ alu.inst = V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_RECIP_IEEE;
+ alu.src[0].sel = ctx->temp_reg;
+ alu.src[0].chan = 2;
+ alu.src[0].abs = 1;
+ alu.dst.sel = ctx->temp_reg;
+ alu.dst.chan = 2;
+ alu.dst.write = 1;
+ alu.last = 1;
+ r = r600_bc_add_alu(ctx->bc, &alu);
+ if (r)
+ return r;
+
+ /* MULADD R0.x, R0.x, PS1, (0x3FC00000, 1.5f).x
+ * MULADD R0.y, R0.y, PS1, (0x3FC00000, 1.5f).x
+ * muladd has no writemask, have to use another temp
+ */
+ memset(&alu, 0, sizeof(struct r600_bc_alu));
+ alu.inst = V_SQ_ALU_WORD1_OP3_SQ_OP3_INST_MULADD;
+ alu.is_op3 = 1;
+
+ alu.src[0].sel = ctx->temp_reg;
+ alu.src[0].chan = 0;
+ alu.src[1].sel = ctx->temp_reg;
+ alu.src[1].chan = 2;
+
+ alu.src[2].sel = V_SQ_ALU_SRC_LITERAL;
+ alu.src[2].chan = 0;
+
+ alu.dst.sel = ctx->temp_reg;
+ alu.dst.chan = 0;
+ alu.dst.write = 1;
+
+ r = r600_bc_add_alu(ctx->bc, &alu);
+ if (r)
+ return r;
+
+ memset(&alu, 0, sizeof(struct r600_bc_alu));
+ alu.inst = V_SQ_ALU_WORD1_OP3_SQ_OP3_INST_MULADD;
+ alu.is_op3 = 1;
+
+ alu.src[0].sel = ctx->temp_reg;
+ alu.src[0].chan = 1;
+ alu.src[1].sel = ctx->temp_reg;
+ alu.src[1].chan = 2;
+
+ alu.src[2].sel = V_SQ_ALU_SRC_LITERAL;
+ alu.src[2].chan = 0;
+
+ alu.dst.sel = ctx->temp_reg;
+ alu.dst.chan = 1;
+ alu.dst.write = 1;
+
+ alu.last = 1;
+ r = r600_bc_add_alu(ctx->bc, &alu);
+ 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;
+ }
+
+ if (src_not_temp) {
for (i = 0; i < 4; i++) {
memset(&alu, 0, sizeof(struct r600_bc_alu));
alu.inst = V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV;
}
src_gpr = ctx->temp_reg;
}
+
+ opcode = ctx->inst_info->r600_opcode;
+ if (opcode == SQ_TEX_INST_SAMPLE &&
+ (inst->Texture.Texture == TGSI_TEXTURE_SHADOW1D || inst->Texture.Texture == TGSI_TEXTURE_SHADOW2D))
+ opcode = SQ_TEX_INST_SAMPLE_C;
memset(&tex, 0, sizeof(struct r600_bc_tex));
- tex.inst = ctx->inst_info->r600_opcode;
+ tex.inst = opcode;
tex.resource_id = ctx->file_offset[inst->Src[1].Register.File] + inst->Src[1].Register.Index;
tex.sampler_id = tex.resource_id;
tex.src_gpr = src_gpr;
tex.src_sel_z = 2;
tex.src_sel_w = 3;
+ if (inst->Texture.Texture == TGSI_TEXTURE_CUBE) {
+ tex.src_sel_x = 1;
+ tex.src_sel_y = 0;
+ tex.src_sel_z = 3;
+ tex.src_sel_w = 1;
+ }
+
if (inst->Texture.Texture != TGSI_TEXTURE_RECT) {
tex.coord_type_x = 1;
tex.coord_type_y = 1;
tex.coord_type_z = 1;
tex.coord_type_w = 1;
}
- return r600_bc_add_tex(ctx->bc, &tex);
+
+ if (inst->Texture.Texture == TGSI_TEXTURE_SHADOW1D || inst->Texture.Texture == TGSI_TEXTURE_SHADOW2D)
+ tex.src_sel_w = 2;
+
+ r = r600_bc_add_tex(ctx->bc, &tex);
+ if (r)
+ return r;
+
+ /* add shadow ambient support - gallium doesn't do it yet */
+ return 0;
+
}
static int tgsi_lrp(struct r600_shader_ctx *ctx)