WIP

1 files changed, 44 insertions, 1 deletions

diff --git a/src/vrend_shader.c b/src/vrend_shader.c
index 1934840..b778aa1 100644
--- a/src/vrend_shader.c
+++ b/src/vrend_shader.c
@@ -138,6 +138,7 @@ struct dump_ctx {
    bool has_clipvertex_so;
    bool has_viewport_idx;
    bool uses_gpu_shader5;
+   bool write_mul_temp;
 };
 
 static inline const char *tgsi_proc_to_prefix(int shader_type)
@@ -812,6 +813,9 @@ static int prepare_so_movs(struct dump_ctx *ctx)
          continue;
 
       ctx->outputs[ctx->so->output[i].register_index].stream = ctx->so->output[i].stream;
+      if (ctx->prog_type == TGSI_PROCESSOR_GEOMETRY && ctx->so->output[i].stream)
+         ctx->uses_gpu_shader5 = true;
+ 
       ctx->write_so_outputs[i] = false;
    }
    return 0;
@@ -1129,7 +1133,7 @@ static int translate_tex(struct dump_ctx *ctx,
 
       sampler_index = 2;
       ctx->uses_tg4 = true;
-      if (inst->Texture.NumOffsets > 1 || is_shad)
+      if (inst->Texture.NumOffsets > 1 || is_shad || ctx->uses_sampler_rect)
          ctx->uses_gpu_shader5 = true;
       if (inst->Texture.NumOffsets == 1) {
          if (inst->TexOffsets[0].File != TGSI_FILE_IMMEDIATE)
@@ -1985,6 +1989,40 @@ iter_instruction(struct tgsi_iterate_context *iter,
       EMIT_BUF_WITH_RET(ctx, buf);
       ctx->uses_gpu_shader5 = true;
       break;
+   case TGSI_OPCODE_UMUL_HI:
+      snprintf(buf, 255, "umulExtended(%s, %s, %s, mul_temp);\n", srcs[0], srcs[1], dsts[0]);
+      EMIT_BUF_WITH_RET(ctx, buf);
+      ctx->uses_gpu_shader5 = true;
+      ctx->write_mul_temp = true;
+      break;
+   case TGSI_OPCODE_IMUL_HI:
+      snprintf(buf, 255, "imulExtended(%s, %s, %s, mul_temp);\n", srcs[0], srcs[1], dsts[0]);
+      EMIT_BUF_WITH_RET(ctx, buf);
+      ctx->uses_gpu_shader5 = true;
+      ctx->write_mul_temp = true;
+      break;
+
+   case TGSI_OPCODE_IBFE:
+   case TGSI_OPCODE_UBFE:
+   case TGSI_OPCODE_BFI:
+   case TGSI_OPCODE_BREV:
+      break;
+   case TGSI_OPCODE_POPC:
+      snprintf(buf, 255, "%s = bitCount(%s);\n", dsts[0], srcs[0]);
+      EMIT_BUF_WITH_RET(ctx, buf);
+      ctx->uses_gpu_shader5 = true;
+      break;
+   case TGSI_OPCODE_LSB:
+      snprintf(buf, 255, "%s = findLSB(%s);\n", dsts[0], srcs[0]);
+      EMIT_BUF_WITH_RET(ctx, buf);
+      ctx->uses_gpu_shader5 = true;
+      break;
+   case TGSI_OPCODE_IMSB:
+   case TGSI_OPCODE_UMSB:
+      snprintf(buf, 255, "%s = findMSB(%s);\n", dsts[0], srcs[0]);
+      EMIT_BUF_WITH_RET(ctx, buf);
+      ctx->uses_gpu_shader5 = true;
+      break;
    default:
       fprintf(stderr,"failed to convert opcode %d\n", inst->Instruction.Opcode);
       break;
@@ -2262,6 +2300,11 @@ static char *emit_ios(struct dump_ctx *ctx, char *glsl_hdr)
       STRCAT_WITH_RET(glsl_hdr, buf);
    }
 
+   if (ctx->write_mul_temp) {
+      snprintf(buf, 255, "vec4 mul_temp;\n");
+      STRCAT_WITH_RET(glsl_hdr, buf);
+   }
+      
    for (i = 0; i < ctx->num_address; i++) {
       snprintf(buf, 255, "int addr%d;\n", i);
       STRCAT_WITH_RET(glsl_hdr, buf);