r600g: fix up vs export handling

[mesa.git] / src / gallium / drivers / r600 / r600_shader.c

diff --git a/src/gallium/drivers/r600/r600_shader.c b/src/gallium/drivers/r600/r600_shader.c
index 6bb5ceb5450c056614c2b2fd9c97f004e2826880..fc56656f55d034340315824f3eacac0e3810e584 100644 (file)
--- a/src/gallium/drivers/r600/r600_shader.c
+++ b/src/gallium/drivers/r600/r600_shader.c
@@ -99,14 +99,14 @@ static int r600_pipe_shader(struct pipe_context *ctx, struct r600_pipe_shader *s
        /* build state */
        switch (rshader->processor_type) {
        case TGSI_PROCESSOR_VERTEX:
-               if (rshader->family >= CHIP_CEDAR) {
+               if (rctx->chip_class >= EVERGREEN) {
                        evergreen_pipe_shader_vs(ctx, shader);
                } else {
                        r600_pipe_shader_vs(ctx, shader);
                }
                break;
        case TGSI_PROCESSOR_FRAGMENT:
-               if (rshader->family >= CHIP_CEDAR) {
+               if (rctx->chip_class >= EVERGREEN) {
                        evergreen_pipe_shader_ps(ctx, shader);
                } else {
                        r600_pipe_shader_ps(ctx, shader);
@@ -135,7 +135,6 @@ int r600_pipe_shader_create(struct pipe_context *ctx, struct r600_pipe_shader *s
                fprintf(stderr, "--------------------------------------------------------------\n");
                tgsi_dump(shader->tokens, 0);
        }
-       shader->shader.family = r600_get_family(rctx->radeon);
        r = r600_shader_from_tgsi(rctx, shader);
        if (r) {
                R600_ERR("translation from TGSI failed !\n");
@@ -317,7 +316,7 @@ 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_FRAGMENT && ctx->bc->chiprev >= CHIPREV_EVERGREEN) {
+               if (ctx->type == TGSI_PROCESSOR_FRAGMENT && ctx->bc->chip_class >= EVERGREEN) {
                        /* turn input into interpolate on EG */
                        if (ctx->shader->input[i].name != TGSI_SEMANTIC_POSITION) {
                                if (ctx->shader->input[i].interpolate > 0) {
@@ -333,6 +332,12 @@ static int tgsi_declaration(struct r600_shader_ctx *ctx)
                ctx->shader->output[i].sid = d->Semantic.Index;
                ctx->shader->output[i].gpr = ctx->file_offset[TGSI_FILE_OUTPUT] + i;
                ctx->shader->output[i].interpolate = d->Declaration.Interpolate;
+               if (ctx->type == TGSI_PROCESSOR_VERTEX) {
+                       /* these don't count as vertex param exports */
+                       if ((ctx->shader->output[i].name == TGSI_SEMANTIC_POSITION) ||
+                           (ctx->shader->output[i].name == TGSI_SEMANTIC_PSIZE))
+                               ctx->shader->npos++;
+               }
                break;
        case TGSI_FILE_CONSTANT:
        case TGSI_FILE_TEMPORARY:
@@ -610,9 +615,7 @@ static int r600_shader_from_tgsi(struct r600_pipe_context * rctx, struct r600_pi
 
        ctx.bc = &shader->bc;
        ctx.shader = shader;
-       r = r600_bc_init(ctx.bc, shader->family);
-       if (r)
-               return r;
+       r600_bc_init(ctx.bc, rctx->chip_class);
        ctx.tokens = tokens;
        tgsi_scan_shader(tokens, &ctx.info);
        tgsi_parse_init(&ctx.parse, tokens);
@@ -651,19 +654,19 @@ static int r600_shader_from_tgsi(struct r600_pipe_context * rctx, struct r600_pi
        }
        if (ctx.type == TGSI_PROCESSOR_VERTEX) {
                ctx.file_offset[TGSI_FILE_INPUT] = 1;
-               if (ctx.bc->chiprev >= CHIPREV_EVERGREEN) {
+               if (ctx.bc->chip_class >= EVERGREEN) {
                        r600_bc_add_cfinst(ctx.bc, EG_V_SQ_CF_WORD1_SQ_CF_INST_CALL_FS);
                } else {
                        r600_bc_add_cfinst(ctx.bc, V_SQ_CF_WORD1_SQ_CF_INST_CALL_FS);
                }
        }
-       if (ctx.type == TGSI_PROCESSOR_FRAGMENT && ctx.bc->chiprev >= CHIPREV_EVERGREEN) {
+       if (ctx.type == TGSI_PROCESSOR_FRAGMENT && ctx.bc->chip_class >= EVERGREEN) {
                ctx.file_offset[TGSI_FILE_INPUT] = evergreen_gpr_count(&ctx);
        }
        ctx.file_offset[TGSI_FILE_OUTPUT] = ctx.file_offset[TGSI_FILE_INPUT] +
-                                               ctx.info.file_count[TGSI_FILE_INPUT];
+                                               ctx.info.file_max[TGSI_FILE_INPUT] + 1;
        ctx.file_offset[TGSI_FILE_TEMPORARY] = ctx.file_offset[TGSI_FILE_OUTPUT] +
-                                               ctx.info.file_count[TGSI_FILE_OUTPUT];
+                                               ctx.info.file_max[TGSI_FILE_OUTPUT] + 1;
 
        /* Outside the GPR range. This will be translated to one of the
         * kcache banks later. */
@@ -671,7 +674,7 @@ static int r600_shader_from_tgsi(struct r600_pipe_context * rctx, struct r600_pi
 
        ctx.file_offset[TGSI_FILE_IMMEDIATE] = V_SQ_ALU_SRC_LITERAL;
        ctx.ar_reg = ctx.file_offset[TGSI_FILE_TEMPORARY] +
-                       ctx.info.file_count[TGSI_FILE_TEMPORARY];
+                       ctx.info.file_max[TGSI_FILE_TEMPORARY] + 1;
        ctx.temp_reg = ctx.ar_reg + 1;
 
        ctx.nliterals = 0;
@@ -711,9 +714,9 @@ static int r600_shader_from_tgsi(struct r600_pipe_context * rctx, struct r600_pi
                                goto out_err;
                        if ((r = tgsi_split_literal_constant(&ctx)))
                                goto out_err;
-                       if (ctx.bc->chiprev == CHIPREV_CAYMAN)
+                       if (ctx.bc->chip_class == CAYMAN)
                                ctx.inst_info = &cm_shader_tgsi_instruction[opcode];
-                       else if (ctx.bc->chiprev >= CHIPREV_EVERGREEN)
+                       else if (ctx.bc->chip_class >= EVERGREEN)
                                ctx.inst_info = &eg_shader_tgsi_instruction[opcode];
                        else
                                ctx.inst_info = &r600_shader_tgsi_instruction[opcode];
@@ -802,7 +805,7 @@ static int r600_shader_from_tgsi(struct r600_pipe_context * rctx, struct r600_pi
                        if (shader->output[i].name == TGSI_SEMANTIC_COLOR) {
                                output[i + j].array_base = shader->output[i].sid;
                                output[i + j].type = V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PIXEL;
-                               if (shader->fs_write_all && (shader->family >= CHIP_CEDAR)) {
+                               if (shader->fs_write_all && (rctx->chip_class >= EVERGREEN)) {
                                        for (j = 1; j < shader->nr_cbufs; j++) {
                                                memset(&output[i + j], 0, sizeof(struct r600_bc_output));
                                                output[i + j].gpr = shader->output[i].gpr;
@@ -886,7 +889,7 @@ static int r600_shader_from_tgsi(struct r600_pipe_context * rctx, struct r600_pi
        }
        /* set export done on last export of each type */
        for (i = noutput - 1, output_done = 0; i >= 0; i--) {
-               if (ctx.bc->chiprev < CHIPREV_CAYMAN) {
+               if (ctx.bc->chip_class < CAYMAN) {
                        if (i == (noutput - 1)) {
                                output[i].end_of_program = 1;
                        }
@@ -903,7 +906,7 @@ static int r600_shader_from_tgsi(struct r600_pipe_context * rctx, struct r600_pi
                        goto out_err;
        }
        /* add program end */
-       if (ctx.bc->chiprev == CHIPREV_CAYMAN)
+       if (ctx.bc->chip_class == CAYMAN)
                cm_bc_add_cf_end(ctx.bc);
 
        free(ctx.literals);
@@ -1123,7 +1126,7 @@ static int tgsi_setup_trig(struct r600_shader_ctx *ctx)
        alu.src[2].sel = V_SQ_ALU_SRC_LITERAL;
        alu.src[2].chan = 0;
 
-       if (ctx->bc->chiprev == CHIPREV_R600) {
+       if (ctx->bc->chip_class == R600) {
                alu.src[1].value = *(uint32_t *)&double_pi;
                alu.src[2].value = *(uint32_t *)&neg_pi;
        } else {
@@ -1230,7 +1233,7 @@ static int tgsi_scs(struct r600_shader_ctx *ctx)
 
        /* dst.x = COS */
        if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) {
-               if (ctx->bc->chiprev == CHIPREV_CAYMAN) {
+               if (ctx->bc->chip_class == CAYMAN) {
                        for (i = 0 ; i < 3; i++) {
                                memset(&alu, 0, sizeof(struct r600_bc_alu));
                                alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_COS);
@@ -1264,7 +1267,7 @@ static int tgsi_scs(struct r600_shader_ctx *ctx)
 
        /* dst.y = SIN */
        if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) {
-               if (ctx->bc->chiprev == CHIPREV_CAYMAN) {
+               if (ctx->bc->chip_class == CAYMAN) {
                        for (i = 0 ; i < 3; i++) {
                                memset(&alu, 0, sizeof(struct r600_bc_alu));
                                alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_SIN);
@@ -1373,19 +1376,37 @@ static int tgsi_lit(struct r600_shader_ctx *ctx)
        struct r600_bc_alu alu;
        int r;
 
+       /* tmp.x = max(src.y, 0.0) */
+       memset(&alu, 0, sizeof(struct r600_bc_alu));
+       alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MAX);
+       r600_bc_src(&alu.src[0], &ctx->src[0], 1);
+       alu.src[1].sel  = V_SQ_ALU_SRC_0; /*0.0*/
+       alu.src[1].chan = 1;
+
+       alu.dst.sel = ctx->temp_reg;
+       alu.dst.chan = 0;
+       alu.dst.write = 1;
+
+       alu.last = 1;
+       r = r600_bc_add_alu(ctx->bc, &alu);
+       if (r)
+               return r;
+
        if (inst->Dst[0].Register.WriteMask & (1 << 2))
        {
                int chan;
                int sel;
                int i;
 
-               if (ctx->bc->chiprev == CHIPREV_CAYMAN) {
+               if (ctx->bc->chip_class == CAYMAN) {
                        for (i = 0; i < 3; i++) {
-                               /* dst.z = log(src.y) */
+                               /* tmp.z = log(tmp.x) */
                                memset(&alu, 0, sizeof(struct r600_bc_alu));
                                alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_LOG_CLAMPED);
-                               r600_bc_src(&alu.src[0], &ctx->src[0], 1);
-                               tgsi_dst(ctx, &inst->Dst[0], i, &alu.dst);
+                               alu.src[0].sel = ctx->temp_reg;
+                               alu.src[0].chan = 0;
+                               alu.dst.sel = ctx->temp_reg;
+                               alu.dst.chan = i;
                                if (i == 2) {
                                        alu.dst.write = 1;
                                        alu.last = 1;
@@ -1397,10 +1418,11 @@ static int tgsi_lit(struct r600_shader_ctx *ctx)
                                        return r;
                        }
                } else {
-                       /* tmp.z = log(src.y) */
+                       /* tmp.z = log(tmp.x) */
                        memset(&alu, 0, sizeof(struct r600_bc_alu));
                        alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_LOG_CLAMPED);
-                       r600_bc_src(&alu.src[0], &ctx->src[0], 1);
+                       alu.src[0].sel = ctx->temp_reg;
+                       alu.src[0].chan = 0;
                        alu.dst.sel = ctx->temp_reg;
                        alu.dst.chan = 2;
                        alu.dst.write = 1;
@@ -1429,7 +1451,7 @@ static int tgsi_lit(struct r600_shader_ctx *ctx)
                if (r)
                        return r;
 
-               if (ctx->bc->chiprev == CHIPREV_CAYMAN) {
+               if (ctx->bc->chip_class == CAYMAN) {
                        for (i = 0; i < 3; i++) {
                                /* dst.z = exp(tmp.x) */
                                memset(&alu, 0, sizeof(struct r600_bc_alu));
@@ -1906,7 +1928,7 @@ static int tgsi_tex(struct r600_shader_ctx *ctx)
        } else if (inst->Instruction.Opcode == TGSI_OPCODE_TXP) {
                int out_chan;
                /* Add perspective divide */
-               if (ctx->bc->chiprev == CHIPREV_CAYMAN) {
+               if (ctx->bc->chip_class == CAYMAN) {
                        out_chan = 2;
                        for (i = 0; i < 3; i++) {
                                memset(&alu, 0, sizeof(struct r600_bc_alu));
@@ -1988,7 +2010,7 @@ static int tgsi_tex(struct r600_shader_ctx *ctx)
                }
 
                /* tmp1.z = RCP_e(|tmp1.z|) */
-               if (ctx->bc->chiprev == CHIPREV_CAYMAN) {
+               if (ctx->bc->chip_class == CAYMAN) {
                        for (i = 0; i < 3; i++) {
                                memset(&alu, 0, sizeof(struct r600_bc_alu));
                                alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_RECIP_IEEE);
@@ -2381,7 +2403,7 @@ static int tgsi_exp(struct r600_shader_ctx *ctx)
                if (r)
                        return r;
 
-               if (ctx->bc->chiprev == CHIPREV_CAYMAN) {
+               if (ctx->bc->chip_class == CAYMAN) {
                        for (i = 0; i < 3; i++) {
                                alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_EXP_IEEE);
                                alu.src[0].sel = ctx->temp_reg;
@@ -2437,7 +2459,7 @@ static int tgsi_exp(struct r600_shader_ctx *ctx)
 
        /* result.z = RoughApprox2ToX(tmp);*/
        if ((inst->Dst[0].Register.WriteMask >> 2) & 0x1) {
-               if (ctx->bc->chiprev == CHIPREV_CAYMAN) {
+               if (ctx->bc->chip_class == CAYMAN) {
                        for (i = 0; i < 3; i++) {
                                memset(&alu, 0, sizeof(struct r600_bc_alu));
                                alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_EXP_IEEE);
@@ -2499,7 +2521,7 @@ static int tgsi_log(struct r600_shader_ctx *ctx)
 
        /* result.x = floor(log2(|src|)); */
        if (inst->Dst[0].Register.WriteMask & 1) {
-               if (ctx->bc->chiprev == CHIPREV_CAYMAN) {
+               if (ctx->bc->chip_class == CAYMAN) {
                        for (i = 0; i < 3; i++) {
                                memset(&alu, 0, sizeof(struct r600_bc_alu));
 
@@ -2551,7 +2573,7 @@ static int tgsi_log(struct r600_shader_ctx *ctx)
        /* result.y = |src.x| / (2 ^ floor(log2(|src.x|))); */
        if ((inst->Dst[0].Register.WriteMask >> 1) & 1) {
 
-               if (ctx->bc->chiprev == CHIPREV_CAYMAN) {
+               if (ctx->bc->chip_class == CAYMAN) {
                        for (i = 0; i < 3; i++) {
                                memset(&alu, 0, sizeof(struct r600_bc_alu));
 
@@ -2602,7 +2624,7 @@ static int tgsi_log(struct r600_shader_ctx *ctx)
                if (r)
                        return r;
 
-               if (ctx->bc->chiprev == CHIPREV_CAYMAN) {
+               if (ctx->bc->chip_class == CAYMAN) {
                        for (i = 0; i < 3; i++) {
                                memset(&alu, 0, sizeof(struct r600_bc_alu));
                                alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_EXP_IEEE);
@@ -2636,7 +2658,7 @@ static int tgsi_log(struct r600_shader_ctx *ctx)
                                return r;
                }
 
-               if (ctx->bc->chiprev == CHIPREV_CAYMAN) {
+               if (ctx->bc->chip_class == CAYMAN) {
                        for (i = 0; i < 3; i++) {
                                memset(&alu, 0, sizeof(struct r600_bc_alu));
                                alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_RECIP_IEEE);
@@ -2692,7 +2714,7 @@ static int tgsi_log(struct r600_shader_ctx *ctx)
 
        /* result.z = log2(|src|);*/
        if ((inst->Dst[0].Register.WriteMask >> 2) & 1) {
-               if (ctx->bc->chiprev == CHIPREV_CAYMAN) {
+               if (ctx->bc->chip_class == CAYMAN) {
                        for (i = 0; i < 3; i++) {
                                memset(&alu, 0, sizeof(struct r600_bc_alu));