nv30/shader: SSG, LIT only requires one source register
[mesa.git] / src / gallium / drivers / r600 / r600_shader.c
index 6d05df6ae285dbc1a38da19bf8428ffa8f62e47b..cd78104a010a97df32908cb2d70f792a731dd038 100644 (file)
  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  * USE OR OTHER DEALINGS IN THE SOFTWARE.
  */
+#include "r600_sq.h"
+#include "r600_llvm.h"
+#include "r600_formats.h"
+#include "r600_opcodes.h"
+#include "r600d.h"
+
 #include "pipe/p_shader_tokens.h"
 #include "tgsi/tgsi_info.h"
 #include "tgsi/tgsi_parse.h"
 #include "tgsi/tgsi_scan.h"
 #include "tgsi/tgsi_dump.h"
-#include "util/u_format.h"
-#include "r600_pipe.h"
-#include "r600_asm.h"
-#include "r600_sq.h"
-#include "r600_formats.h"
-#include "r600_opcodes.h"
-#include "r600d.h"
+#include "util/u_memory.h"
 #include <stdio.h>
 #include <errno.h>
 #include <byteswap.h>
@@ -71,7 +71,7 @@ static int r600_pipe_shader(struct pipe_context *ctx, struct r600_pipe_shader *s
                if (shader->bo == NULL) {
                        return -ENOMEM;
                }
-               ptr = (uint32_t*)rctx->ws->buffer_map(shader->bo->buf, rctx->cs, PIPE_TRANSFER_WRITE);
+               ptr = (uint32_t*)rctx->ws->buffer_map(shader->bo->cs_buf, rctx->cs, PIPE_TRANSFER_WRITE);
                if (R600_BIG_ENDIAN) {
                        for (i = 0; i < rshader->bc.ndw; ++i) {
                                ptr[i] = bswap_32(rshader->bc.bytecode[i]);
@@ -79,7 +79,7 @@ static int r600_pipe_shader(struct pipe_context *ctx, struct r600_pipe_shader *s
                } else {
                        memcpy(ptr, rshader->bc.bytecode, rshader->bc.ndw * sizeof(*ptr));
                }
-               rctx->ws->buffer_unmap(shader->bo->buf);
+               rctx->ws->buffer_unmap(shader->bo->cs_buf);
        }
        /* build state */
        switch (rshader->processor_type) {
@@ -198,6 +198,7 @@ struct r600_shader_ctx {
        boolean                 clip_vertex_write;
        unsigned                cv_output;
        int                                     fragcoord_input;
+       int                                     native_integers;
 };
 
 struct r600_shader_tgsi_instruction {
@@ -209,6 +210,276 @@ struct r600_shader_tgsi_instruction {
 
 static struct r600_shader_tgsi_instruction r600_shader_tgsi_instruction[], eg_shader_tgsi_instruction[], cm_shader_tgsi_instruction[];
 static int tgsi_helper_tempx_replicate(struct r600_shader_ctx *ctx);
+static inline void callstack_check_depth(struct r600_shader_ctx *ctx, unsigned reason, unsigned check_max_only);
+static void fc_pushlevel(struct r600_shader_ctx *ctx, int type);
+static int tgsi_else(struct r600_shader_ctx *ctx);
+static int tgsi_endif(struct r600_shader_ctx *ctx);
+static int tgsi_bgnloop(struct r600_shader_ctx *ctx);
+static int tgsi_endloop(struct r600_shader_ctx *ctx);
+static int tgsi_loop_brk_cont(struct r600_shader_ctx *ctx);
+
+/*
+ * bytestream -> r600 shader
+ *
+ * These functions are used to transform the output of the LLVM backend into
+ * struct r600_bytecode.
+ */
+
+static unsigned r600_src_from_byte_stream(unsigned char * bytes,
+               unsigned bytes_read, struct r600_bytecode_alu * alu, unsigned src_idx)
+{
+       unsigned i;
+       unsigned sel0, sel1;
+       sel0 = bytes[bytes_read++];
+       sel1 = bytes[bytes_read++];
+       alu->src[src_idx].sel = sel0 | (sel1 << 8);
+       alu->src[src_idx].chan = bytes[bytes_read++];
+       alu->src[src_idx].neg = bytes[bytes_read++];
+       alu->src[src_idx].abs = bytes[bytes_read++];
+       alu->src[src_idx].rel = bytes[bytes_read++];
+       alu->src[src_idx].kc_bank = bytes[bytes_read++];
+       for (i = 0; i < 4; i++) {
+               alu->src[src_idx].value |= bytes[bytes_read++] << (i * 8);
+       }
+       return bytes_read;
+}
+
+static unsigned r600_alu_from_byte_stream(struct r600_shader_ctx *ctx,
+                               unsigned char * bytes, unsigned bytes_read)
+{
+       unsigned src_idx;
+       unsigned inst0, inst1;
+       struct r600_bytecode_alu alu;
+       memset(&alu, 0, sizeof(alu));
+       for(src_idx = 0; src_idx < 3; src_idx++) {
+               bytes_read = r600_src_from_byte_stream(bytes, bytes_read,
+                                                               &alu, src_idx);
+       }
+
+       alu.dst.sel = bytes[bytes_read++];
+       alu.dst.chan = bytes[bytes_read++];
+       alu.dst.clamp = bytes[bytes_read++];
+       alu.dst.write = bytes[bytes_read++];
+       alu.dst.rel = bytes[bytes_read++];
+       inst0 = bytes[bytes_read++];
+       inst1 = bytes[bytes_read++];
+       alu.inst = inst0 | (inst1 << 8);
+       alu.last = bytes[bytes_read++];
+       alu.is_op3 = bytes[bytes_read++];
+       alu.predicate = bytes[bytes_read++];
+       alu.bank_swizzle = bytes[bytes_read++];
+       alu.bank_swizzle_force = bytes[bytes_read++];
+       alu.omod = bytes[bytes_read++];
+       alu.index_mode = bytes[bytes_read++];
+       r600_bytecode_add_alu(ctx->bc, &alu);
+
+       /* XXX: Handle other KILL instructions */
+       if (alu.inst == CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_KILLGT)) {
+               ctx->shader->uses_kill = 1;
+               /* XXX: This should be enforced in the LLVM backend. */
+               ctx->bc->force_add_cf = 1;
+       }
+       return bytes_read;
+}
+
+static void llvm_if(struct r600_shader_ctx *ctx, struct r600_bytecode_alu * alu,
+       unsigned pred_inst)
+{
+       alu->inst = pred_inst; 
+       alu->predicate = 1;
+       alu->dst.write = 0;
+       alu->src[1].sel = V_SQ_ALU_SRC_0;
+       alu->src[1].chan = 0;
+       alu->last = 1;
+       r600_bytecode_add_alu_type(ctx->bc, alu,
+               CTX_INST(V_SQ_CF_ALU_WORD1_SQ_CF_INST_ALU_PUSH_BEFORE));
+
+       r600_bytecode_add_cfinst(ctx->bc, CTX_INST(V_SQ_CF_WORD1_SQ_CF_INST_JUMP));
+       fc_pushlevel(ctx, FC_IF);
+       callstack_check_depth(ctx, FC_PUSH_VPM, 0);
+}
+
+static void r600_break_from_byte_stream(struct r600_shader_ctx *ctx,
+                       struct r600_bytecode_alu *alu, unsigned compare_opcode)
+{
+       unsigned opcode = TGSI_OPCODE_BRK;
+       if (ctx->bc->chip_class == CAYMAN)
+               ctx->inst_info = &cm_shader_tgsi_instruction[opcode];
+       else if (ctx->bc->chip_class >= EVERGREEN)
+               ctx->inst_info = &eg_shader_tgsi_instruction[opcode];
+       else
+               ctx->inst_info = &r600_shader_tgsi_instruction[opcode];
+       llvm_if(ctx, alu, compare_opcode);
+       tgsi_loop_brk_cont(ctx);
+       tgsi_endif(ctx);
+}
+
+static unsigned r600_fc_from_byte_stream(struct r600_shader_ctx *ctx,
+                               unsigned char * bytes, unsigned bytes_read)
+{
+       struct r600_bytecode_alu alu;
+       unsigned inst;
+       memset(&alu, 0, sizeof(alu));
+       bytes_read = r600_src_from_byte_stream(bytes, bytes_read, &alu, 0);
+       inst = bytes[bytes_read++];
+       switch (inst) {
+       case 0:
+               llvm_if(ctx, &alu,
+                       CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_PRED_SETNE));
+               break;
+       case 1:
+               tgsi_else(ctx);
+               break;
+       case 2:
+               tgsi_endif(ctx);
+               break;
+       case 3:
+               tgsi_bgnloop(ctx);
+               break;
+       case 4:
+               tgsi_endloop(ctx);
+               break;
+       case 5:
+               r600_break_from_byte_stream(ctx, &alu,
+                       CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_PRED_SETE));
+               break;
+       case 6:
+               r600_break_from_byte_stream(ctx, &alu,
+                       CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_PRED_SETNE_INT));
+               break;
+       case 7:
+               {
+                       unsigned opcode = TGSI_OPCODE_CONT;
+                       if (ctx->bc->chip_class == CAYMAN) {
+                               ctx->inst_info =
+                                       &cm_shader_tgsi_instruction[opcode];
+                       } else if (ctx->bc->chip_class >= EVERGREEN) {
+                               ctx->inst_info =
+                                       &eg_shader_tgsi_instruction[opcode];
+                       } else {
+                               ctx->inst_info =
+                                       &r600_shader_tgsi_instruction[opcode];
+                       }
+                       tgsi_loop_brk_cont(ctx);
+               }
+               break;
+       case 8:
+               r600_break_from_byte_stream(ctx, &alu,
+                       CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_PRED_SETE_INT));
+               break;
+       }
+
+       return bytes_read;
+}
+
+static unsigned r600_tex_from_byte_stream(struct r600_shader_ctx *ctx,
+                               unsigned char * bytes, unsigned bytes_read)
+{
+       struct r600_bytecode_tex tex;
+
+       tex.inst = bytes[bytes_read++];
+       tex.resource_id = bytes[bytes_read++];
+       tex.src_gpr = bytes[bytes_read++];
+       tex.src_rel = bytes[bytes_read++];
+       tex.dst_gpr = bytes[bytes_read++];
+       tex.dst_rel = bytes[bytes_read++];
+       tex.dst_sel_x = bytes[bytes_read++];
+       tex.dst_sel_y = bytes[bytes_read++];
+       tex.dst_sel_z = bytes[bytes_read++];
+       tex.dst_sel_w = bytes[bytes_read++];
+       tex.lod_bias = bytes[bytes_read++];
+       tex.coord_type_x = bytes[bytes_read++];
+       tex.coord_type_y = bytes[bytes_read++];
+       tex.coord_type_z = bytes[bytes_read++];
+       tex.coord_type_w = bytes[bytes_read++];
+       tex.offset_x = bytes[bytes_read++];
+       tex.offset_y = bytes[bytes_read++];
+       tex.offset_z = bytes[bytes_read++];
+       tex.sampler_id = bytes[bytes_read++];
+       tex.src_sel_x = bytes[bytes_read++];
+       tex.src_sel_y = bytes[bytes_read++];
+       tex.src_sel_z = bytes[bytes_read++];
+       tex.src_sel_w = bytes[bytes_read++];
+
+       r600_bytecode_add_tex(ctx->bc, &tex);
+
+       return bytes_read;
+}
+
+static int r600_vtx_from_byte_stream(struct r600_shader_ctx *ctx,
+       unsigned char * bytes, unsigned bytes_read)
+{
+       struct r600_bytecode_vtx vtx;
+       memset(&vtx, 0, sizeof(vtx));
+       vtx.inst = bytes[bytes_read++];
+       vtx.fetch_type = bytes[bytes_read++];
+       vtx.buffer_id = bytes[bytes_read++];
+       vtx.src_gpr = bytes[bytes_read++];
+       vtx.src_sel_x = bytes[bytes_read++];
+       vtx.mega_fetch_count = bytes[bytes_read++];
+       vtx.dst_gpr = bytes[bytes_read++];
+       vtx.dst_sel_x = bytes[bytes_read++];
+       vtx.dst_sel_y = bytes[bytes_read++];
+       vtx.dst_sel_z = bytes[bytes_read++];
+       vtx.dst_sel_w = bytes[bytes_read++];
+       vtx.use_const_fields = bytes[bytes_read++];
+       vtx.data_format = bytes[bytes_read++];
+       vtx.num_format_all = bytes[bytes_read++];
+       vtx.format_comp_all = bytes[bytes_read++];
+       vtx.srf_mode_all = bytes[bytes_read++];
+       vtx.offset = bytes[bytes_read++];
+       vtx.endian = bytes[bytes_read++];
+
+       if (r600_bytecode_add_vtx(ctx->bc, &vtx)) {
+               fprintf(stderr, "Error adding vtx\n");
+       }
+       /* Use the Texture Cache */
+       ctx->bc->cf_last->inst = EG_V_SQ_CF_WORD1_SQ_CF_INST_TEX;
+       return bytes_read;
+}
+
+static void r600_bytecode_from_byte_stream(struct r600_shader_ctx *ctx,
+                               unsigned char * bytes,  unsigned num_bytes)
+{
+       unsigned bytes_read = 0;
+       unsigned i, byte;
+       while (bytes_read < num_bytes) {
+               char inst_type = bytes[bytes_read++];
+               switch (inst_type) {
+               case 0:
+                       bytes_read = r600_alu_from_byte_stream(ctx, bytes,
+                                                               bytes_read);
+                       break;
+               case 1:
+                       bytes_read = r600_tex_from_byte_stream(ctx, bytes,
+                                                               bytes_read);
+                       break;
+               case 2:
+                       bytes_read = r600_fc_from_byte_stream(ctx, bytes,
+                                                               bytes_read);
+                       break;
+               case 3:
+                       r600_bytecode_add_cfinst(ctx->bc, CF_NATIVE);
+                       for (i = 0; i < 2; i++) {
+                               for (byte = 0 ; byte < 4; byte++) {
+                                       ctx->bc->cf_last->isa[i] |=
+                                       (bytes[bytes_read++] << (byte * 8));
+                               }
+                       }
+                       break;
+
+               case 4:
+                       bytes_read = r600_vtx_from_byte_stream(ctx, bytes,
+                                                               bytes_read);
+                       break;
+               default:
+                       /* XXX: Error here */
+                       break;
+               }
+       }
+}
+
+/* End bytestream -> r600 shader functions*/
 
 static int tgsi_is_supported(struct r600_shader_ctx *ctx)
 {
@@ -451,8 +722,8 @@ static int tgsi_declaration(struct r600_shader_ctx *ctx)
                ctx->shader->input[i].name = d->Semantic.Name;
                ctx->shader->input[i].sid = d->Semantic.Index;
                ctx->shader->input[i].spi_sid = r600_spi_sid(&ctx->shader->input[i]);
-               ctx->shader->input[i].interpolate = d->Declaration.Interpolate;
-               ctx->shader->input[i].centroid = d->Declaration.Centroid;
+               ctx->shader->input[i].interpolate = d->Interp.Interpolate;
+               ctx->shader->input[i].centroid = d->Interp.Centroid;
                ctx->shader->input[i].gpr = ctx->file_offset[TGSI_FILE_INPUT] + d->Range.First;
                if (ctx->type == TGSI_PROCESSOR_FRAGMENT) {
                        switch (ctx->shader->input[i].name) {
@@ -478,7 +749,7 @@ static int tgsi_declaration(struct r600_shader_ctx *ctx)
                ctx->shader->output[i].sid = d->Semantic.Index;
                ctx->shader->output[i].spi_sid = r600_spi_sid(&ctx->shader->output[i]);
                ctx->shader->output[i].gpr = ctx->file_offset[TGSI_FILE_OUTPUT] + d->Range.First;
-               ctx->shader->output[i].interpolate = d->Declaration.Interpolate;
+               ctx->shader->output[i].interpolate = d->Interp.Interpolate;
                ctx->shader->output[i].write_mask = d->Declaration.UsageMask;
                if (ctx->type == TGSI_PROCESSOR_VERTEX) {
                        switch (d->Semantic.Name) {
@@ -504,20 +775,22 @@ static int tgsi_declaration(struct r600_shader_ctx *ctx)
 
        case TGSI_FILE_SYSTEM_VALUE:
                if (d->Semantic.Name == TGSI_SEMANTIC_INSTANCEID) {
-                       struct r600_bytecode_alu alu;
-                       memset(&alu, 0, sizeof(struct r600_bytecode_alu));
+                       if (!ctx->native_integers) {
+                               struct r600_bytecode_alu alu;
+                               memset(&alu, 0, sizeof(struct r600_bytecode_alu));
 
-                       alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_INT_TO_FLT);
-                       alu.src[0].sel = 0;
-                       alu.src[0].chan = 3;
+                               alu.inst = CTX_INST(V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_INT_TO_FLT);
+                               alu.src[0].sel = 0;
+                               alu.src[0].chan = 3;
 
-                       alu.dst.sel = 0;
-                       alu.dst.chan = 3;
-                       alu.dst.write = 1;
-                       alu.last = 1;
+                               alu.dst.sel = 0;
+                               alu.dst.chan = 3;
+                               alu.dst.write = 1;
+                               alu.last = 1;
 
-                       if ((r = r600_bytecode_add_alu(ctx->bc, &alu)))
-                               return r;
+                               if ((r = r600_bytecode_add_alu(ctx->bc, &alu)))
+                                       return r;
+                       }
                        break;
                } else if (d->Semantic.Name == TGSI_SEMANTIC_VERTEXID)
                        break;
@@ -576,7 +849,7 @@ static int evergreen_gpr_count(struct r600_shader_ctx *ctx)
 
        ctx->num_interp_gpr += (num_baryc + 1) >> 1;
 
-       /* TODO PULL MODEL and LINE STIPPLE, FIXED PT POS */
+       /* XXX PULL MODEL and LINE STIPPLE, FIXED PT POS */
        return ctx->num_interp_gpr;
 }
 
@@ -819,9 +1092,18 @@ static int r600_shader_from_tgsi(struct r600_context * rctx, struct r600_pipe_sh
        unsigned opcode;
        int i, j, k, r = 0;
        int next_pixel_base = 0, next_pos_base = 60, next_param_base = 0;
+       /* Declarations used by llvm code */
+       bool use_llvm = false;
+       unsigned char * inst_bytes = NULL;
+       unsigned inst_byte_count = 0;
 
+#ifdef R600_USE_LLVM
+       use_llvm = debug_get_bool_option("R600_LLVM", TRUE);
+#endif
        ctx.bc = &shader->bc;
        ctx.shader = shader;
+       ctx.native_integers = (rctx->screen->glsl_feature_level >= 130);
+
        r600_bytecode_init(ctx.bc, rctx->chip_class, rctx->family);
        ctx.tokens = tokens;
        tgsi_scan_shader(tokens, &ctx.info);
@@ -873,8 +1155,46 @@ static int r600_shader_from_tgsi(struct r600_context * rctx, struct r600_pipe_sh
        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_max[TGSI_FILE_INPUT] + 1;
+
+       /* LLVM backend setup */
+#ifdef R600_USE_LLVM
+       if (use_llvm && ctx.info.indirect_files) {
+               fprintf(stderr, "Warning: R600 LLVM backend does not support "
+                               "indirect adressing.  Falling back to TGSI "
+                               "backend.\n");
+               use_llvm = 0;
+       }
+       if (use_llvm) {
+               struct radeon_llvm_context radeon_llvm_ctx;
+               LLVMModuleRef mod;
+               unsigned dump = 0;
+               memset(&radeon_llvm_ctx, 0, sizeof(radeon_llvm_ctx));
+               radeon_llvm_ctx.reserved_reg_count = ctx.file_offset[TGSI_FILE_INPUT];
+               mod = r600_tgsi_llvm(&radeon_llvm_ctx, tokens);
+               if (debug_get_bool_option("R600_DUMP_SHADERS", FALSE)) {
+                       dump = 1;
+               }
+               if (r600_llvm_compile(mod, &inst_bytes, &inst_byte_count,
+                                                       rctx->family, dump)) {
+                       FREE(inst_bytes);
+                       radeon_llvm_dispose(&radeon_llvm_ctx);
+                       use_llvm = 0;
+                       fprintf(stderr, "R600 LLVM backend failed to compile "
+                               "shader.  Falling back to TGSI\n");
+               } else {
+                       ctx.file_offset[TGSI_FILE_OUTPUT] =
+                                       ctx.file_offset[TGSI_FILE_INPUT];
+               }
+               radeon_llvm_dispose(&radeon_llvm_ctx);
+       }
+#endif
+       /* End of LLVM backend setup */
+
+       if (!use_llvm) {
+               ctx.file_offset[TGSI_FILE_OUTPUT] =
+                       ctx.file_offset[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_max[TGSI_FILE_OUTPUT] + 1;
 
@@ -975,6 +1295,9 @@ static int r600_shader_from_tgsi(struct r600_context * rctx, struct r600_pipe_sh
                tgsi_parse_token(&ctx.parse);
                switch (ctx.parse.FullToken.Token.Type) {
                case TGSI_TOKEN_TYPE_INSTRUCTION:
+                       if (use_llvm) {
+                               continue;
+                       }
                        r = tgsi_is_supported(&ctx);
                        if (r)
                                goto out_err;
@@ -1002,6 +1325,12 @@ static int r600_shader_from_tgsi(struct r600_context * rctx, struct r600_pipe_sh
                }
        }
 
+       /* Get instructions if we are using the LLVM backend. */
+       if (use_llvm) {
+               r600_bytecode_from_byte_stream(&ctx, inst_bytes, inst_byte_count);
+               FREE(inst_bytes);
+       }
+
        noutput = shader->noutput;
 
        if (ctx.clip_vertex_write) {
@@ -1270,6 +1599,14 @@ static int r600_shader_from_tgsi(struct r600_context * rctx, struct r600_pipe_sh
        if (ctx.bc->chip_class == CAYMAN)
                cm_bytecode_add_cf_end(ctx.bc);
 
+       /* check GPR limit - we have 124 = 128 - 4
+        * (4 are reserved as alu clause temporary registers) */
+       if (ctx.bc->ngpr > 124) {
+               R600_ERR("GPR limit exceeded - shader requires %d registers\n", ctx.bc->ngpr);
+               r = -ENOMEM;
+               goto out_err;
+       }
+
        free(ctx.literals);
        tgsi_parse_free(&ctx.parse);
        return 0;
@@ -1953,7 +2290,7 @@ static int tgsi_rsq(struct r600_shader_ctx *ctx)
 
        memset(&alu, 0, sizeof(struct r600_bytecode_alu));
 
-       /* FIXME:
+       /* XXX:
         * For state trackers other than OpenGL, we'll want to use
         * _RECIPSQRT_IEEE instead.
         */
@@ -4528,7 +4865,7 @@ static int emit_jump_to_offset(struct r600_shader_ctx *ctx, int pops, int offset
 
        r600_bytecode_add_cfinst(ctx->bc, CTX_INST(V_SQ_CF_WORD1_SQ_CF_INST_JUMP));
        ctx->bc->cf_last->pop_count = pops;
-       /* TODO work out offset */
+       /* XXX work out offset */
        return 0;
 }
 
@@ -4640,7 +4977,7 @@ static int tgsi_endloop(struct r600_shader_ctx *ctx)
        for (i = 0; i < ctx->bc->fc_stack[ctx->bc->fc_sp].num_mid; i++) {
                ctx->bc->fc_stack[ctx->bc->fc_sp].mid[i]->cf_addr = ctx->bc->cf_last->id;
        }
-       /* TODO add LOOPRET support */
+       /* XXX add LOOPRET support */
        fc_poplevel(ctx);
        callstack_decrease_current(ctx, FC_LOOP);
        return 0;
@@ -4727,7 +5064,7 @@ static struct r600_shader_tgsi_instruction r600_shader_tgsi_instruction[] = {
        {TGSI_OPCODE_MOV,       0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOV, tgsi_op2},
        {TGSI_OPCODE_LIT,       0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_lit},
 
-       /* FIXME:
+       /* XXX:
         * For state trackers other than OpenGL, we'll want to use
         * _RECIP_IEEE instead.
         */
@@ -4817,7 +5154,7 @@ static struct r600_shader_tgsi_instruction r600_shader_tgsi_instruction[] = {
        {80,                    0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported},
        {TGSI_OPCODE_PUSHA,     0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported},
        {TGSI_OPCODE_POPA,      0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported},
-       {TGSI_OPCODE_CEIL,      0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported},
+       {TGSI_OPCODE_CEIL,      0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_CEIL, tgsi_op2},
        {TGSI_OPCODE_I2F,       0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_INT_TO_FLT, tgsi_op2_trans},
        {TGSI_OPCODE_NOT,       0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOT_INT, tgsi_op2},
        {TGSI_OPCODE_TRUNC,     0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_TRUNC, tgsi_op2},
@@ -4883,16 +5220,16 @@ static struct r600_shader_tgsi_instruction r600_shader_tgsi_instruction[] = {
        {TGSI_OPCODE_CASE,      0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported},
        {TGSI_OPCODE_DEFAULT,   0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported},
        {TGSI_OPCODE_ENDSWITCH, 0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported},
-       {TGSI_OPCODE_LOAD,      0, 0, tgsi_unsupported},
-       {TGSI_OPCODE_LOAD_MS,   0, 0, tgsi_unsupported},
        {TGSI_OPCODE_SAMPLE,    0, 0, tgsi_unsupported},
+       {TGSI_OPCODE_SAMPLE_I,  0, 0, tgsi_unsupported},
+       {TGSI_OPCODE_SAMPLE_I_MS, 0, 0, tgsi_unsupported},
        {TGSI_OPCODE_SAMPLE_B,  0, 0, tgsi_unsupported},
        {TGSI_OPCODE_SAMPLE_C,  0, 0, tgsi_unsupported},
        {TGSI_OPCODE_SAMPLE_C_LZ, 0, 0, tgsi_unsupported},
        {TGSI_OPCODE_SAMPLE_D,  0, 0, tgsi_unsupported},
        {TGSI_OPCODE_SAMPLE_L,  0, 0, tgsi_unsupported},
        {TGSI_OPCODE_GATHER4,   0, 0, tgsi_unsupported},
-       {TGSI_OPCODE_RESINFO,   0, 0, tgsi_unsupported},
+       {TGSI_OPCODE_SVIEWINFO, 0, 0, tgsi_unsupported},
        {TGSI_OPCODE_SAMPLE_POS, 0, 0, tgsi_unsupported},
        {TGSI_OPCODE_SAMPLE_INFO, 0, 0, tgsi_unsupported},
        {TGSI_OPCODE_UARL,      0, V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOVA_INT, tgsi_r600_arl},
@@ -4991,7 +5328,7 @@ static struct r600_shader_tgsi_instruction eg_shader_tgsi_instruction[] = {
        {80,                    0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported},
        {TGSI_OPCODE_PUSHA,     0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported},
        {TGSI_OPCODE_POPA,      0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported},
-       {TGSI_OPCODE_CEIL,      0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported},
+       {TGSI_OPCODE_CEIL,      0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_CEIL, tgsi_op2},
        {TGSI_OPCODE_I2F,       0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_INT_TO_FLT, tgsi_op2_trans},
        {TGSI_OPCODE_NOT,       0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOT_INT, tgsi_op2},
        {TGSI_OPCODE_TRUNC,     0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_TRUNC, tgsi_op2},
@@ -5057,16 +5394,16 @@ static struct r600_shader_tgsi_instruction eg_shader_tgsi_instruction[] = {
        {TGSI_OPCODE_CASE,      0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported},
        {TGSI_OPCODE_DEFAULT,   0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported},
        {TGSI_OPCODE_ENDSWITCH, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported},
-       {TGSI_OPCODE_LOAD,      0, 0, tgsi_unsupported},
-       {TGSI_OPCODE_LOAD_MS,   0, 0, tgsi_unsupported},
        {TGSI_OPCODE_SAMPLE,    0, 0, tgsi_unsupported},
+       {TGSI_OPCODE_SAMPLE_I,      0, 0, tgsi_unsupported},
+       {TGSI_OPCODE_SAMPLE_I_MS,   0, 0, tgsi_unsupported},
        {TGSI_OPCODE_SAMPLE_B,  0, 0, tgsi_unsupported},
        {TGSI_OPCODE_SAMPLE_C,  0, 0, tgsi_unsupported},
        {TGSI_OPCODE_SAMPLE_C_LZ, 0, 0, tgsi_unsupported},
        {TGSI_OPCODE_SAMPLE_D,  0, 0, tgsi_unsupported},
        {TGSI_OPCODE_SAMPLE_L,  0, 0, tgsi_unsupported},
        {TGSI_OPCODE_GATHER4,   0, 0, tgsi_unsupported},
-       {TGSI_OPCODE_RESINFO,   0, 0, tgsi_unsupported},
+       {TGSI_OPCODE_SVIEWINFO, 0, 0, tgsi_unsupported},
        {TGSI_OPCODE_SAMPLE_POS, 0, 0, tgsi_unsupported},
        {TGSI_OPCODE_SAMPLE_INFO, 0, 0, tgsi_unsupported},
        {TGSI_OPCODE_UARL,      0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOVA_INT, tgsi_eg_arl},
@@ -5165,7 +5502,7 @@ static struct r600_shader_tgsi_instruction cm_shader_tgsi_instruction[] = {
        {80,                    0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported},
        {TGSI_OPCODE_PUSHA,     0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported},
        {TGSI_OPCODE_POPA,      0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported},
-       {TGSI_OPCODE_CEIL,      0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported},
+       {TGSI_OPCODE_CEIL,      0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_CEIL, tgsi_op2},
        {TGSI_OPCODE_I2F,       0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_INT_TO_FLT, tgsi_op2},
        {TGSI_OPCODE_NOT,       0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOT_INT, tgsi_op2},
        {TGSI_OPCODE_TRUNC,     0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_TRUNC, tgsi_op2},
@@ -5231,16 +5568,16 @@ static struct r600_shader_tgsi_instruction cm_shader_tgsi_instruction[] = {
        {TGSI_OPCODE_CASE,      0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported},
        {TGSI_OPCODE_DEFAULT,   0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported},
        {TGSI_OPCODE_ENDSWITCH, 0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_NOP, tgsi_unsupported},
-       {TGSI_OPCODE_LOAD,      0, 0, tgsi_unsupported},
-       {TGSI_OPCODE_LOAD_MS,   0, 0, tgsi_unsupported},
        {TGSI_OPCODE_SAMPLE,    0, 0, tgsi_unsupported},
+       {TGSI_OPCODE_SAMPLE_I,      0, 0, tgsi_unsupported},
+       {TGSI_OPCODE_SAMPLE_I_MS,   0, 0, tgsi_unsupported},
        {TGSI_OPCODE_SAMPLE_B,  0, 0, tgsi_unsupported},
        {TGSI_OPCODE_SAMPLE_C,  0, 0, tgsi_unsupported},
        {TGSI_OPCODE_SAMPLE_C_LZ, 0, 0, tgsi_unsupported},
        {TGSI_OPCODE_SAMPLE_D,  0, 0, tgsi_unsupported},
        {TGSI_OPCODE_SAMPLE_L,  0, 0, tgsi_unsupported},
        {TGSI_OPCODE_GATHER4,   0, 0, tgsi_unsupported},
-       {TGSI_OPCODE_RESINFO,   0, 0, tgsi_unsupported},
+       {TGSI_OPCODE_SVIEWINFO, 0, 0, tgsi_unsupported},
        {TGSI_OPCODE_SAMPLE_POS, 0, 0, tgsi_unsupported},
        {TGSI_OPCODE_SAMPLE_INFO, 0, 0, tgsi_unsupported},
        {TGSI_OPCODE_UARL,      0, EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_MOVA_INT, tgsi_eg_arl},