15 files changed, 135 insertions, 211 deletions

diff --git a/mesalib/src/mesa/state_tracker/st_atom.c b/mesalib/src/mesa/state_tracker/st_atom.c
index f0fe11ffa..428f2d9d7 100644
--- a/mesalib/src/mesa/state_tracker/st_atom.c
+++ b/mesalib/src/mesa/state_tracker/st_atom.c
@@ -183,7 +183,7 @@ void st_validate_state( struct st_context *st )
    if (state->st == 0)
       return;
 
-   /*printf("%s %x/%x\n", __FUNCTION__, state->mesa, state->st);*/
+   /*printf("%s %x/%x\n", __func__, state->mesa, state->st);*/
 
 #ifdef DEBUG
    if (1) {
diff --git a/mesalib/src/mesa/state_tracker/st_atom_array.c b/mesalib/src/mesa/state_tracker/st_atom_array.c
index 9b52f9779..d4fb8b862 100644
--- a/mesalib/src/mesa/state_tracker/st_atom_array.c
+++ b/mesalib/src/mesa/state_tracker/st_atom_array.c
@@ -598,7 +598,7 @@ static void update_array(struct st_context *st)
 const struct st_tracked_state st_update_array = {
    "st_update_array",					/* name */
    {							/* dirty */
-      0,                                                /* mesa */
+      _NEW_CURRENT_ATTRIB,                              /* mesa */
       ST_NEW_VERTEX_ARRAYS | ST_NEW_VERTEX_PROGRAM,     /* st */
    },
    update_array						/* update */
diff --git a/mesalib/src/mesa/state_tracker/st_atom_constbuf.c b/mesalib/src/mesa/state_tracker/st_atom_constbuf.c
index 7984bf742..a54e0d9db 100644
--- a/mesalib/src/mesa/state_tracker/st_atom_constbuf.c
+++ b/mesalib/src/mesa/state_tracker/st_atom_constbuf.c
@@ -92,7 +92,7 @@ void st_upload_constants( struct st_context *st,
 
       if (ST_DEBUG & DEBUG_CONSTANTS) {
          debug_printf("%s(shader=%d, numParams=%d, stateFlags=0x%x)\n",
-                      __FUNCTION__, shader_type, params->NumParameters,
+                      __func__, shader_type, params->NumParameters,
                       params->StateFlags);
          _mesa_print_parameter_list(params);
       }
diff --git a/mesalib/src/mesa/state_tracker/st_atom_shader.c b/mesalib/src/mesa/state_tracker/st_atom_shader.c
index 73768ed12..629f54f25 100644
--- a/mesalib/src/mesa/state_tracker/st_atom_shader.c
+++ b/mesalib/src/mesa/state_tracker/st_atom_shader.c
@@ -40,9 +40,8 @@
 #include "program/program.h"
 
 #include "pipe/p_context.h"
-
+#include "pipe/p_shader_tokens.h"
 #include "util/u_simple_shaders.h"
-
 #include "cso_cache/cso_context.h"
 
 #include "st_context.h"
diff --git a/mesalib/src/mesa/state_tracker/st_cb_clear.c b/mesalib/src/mesa/state_tracker/st_cb_clear.c
index dd81a6273..f10e9063a 100644
--- a/mesalib/src/mesa/state_tracker/st_cb_clear.c
+++ b/mesalib/src/mesa/state_tracker/st_cb_clear.c
@@ -247,7 +247,7 @@ clear_with_quad(struct gl_context *ctx, unsigned clear_buffers)
       util_framebuffer_get_num_layers(&st->state.framebuffer);
 
    /*
-   printf("%s %s%s%s %f,%f %f,%f\n", __FUNCTION__, 
+   printf("%s %s%s%s %f,%f %f,%f\n", __func__,
 	  color ? "color, " : "",
 	  depth ? "depth, " : "",
 	  stencil ? "stencil" : "",
diff --git a/mesalib/src/mesa/state_tracker/st_cb_program.c b/mesalib/src/mesa/state_tracker/st_cb_program.c
index aa301d830..c382d7d2c 100644
--- a/mesalib/src/mesa/state_tracker/st_cb_program.c
+++ b/mesalib/src/mesa/state_tracker/st_cb_program.c
@@ -41,6 +41,7 @@
 #include "draw/draw_context.h"
 
 #include "st_context.h"
+#include "st_debug.h"
 #include "st_program.h"
 #include "st_mesa_to_tgsi.h"
 #include "st_cb_program.h"
@@ -214,6 +215,9 @@ st_program_string_notify( struct gl_context *ctx,
 	 st->dirty.st |= ST_NEW_VERTEX_PROGRAM;
    }
 
+   if (ST_DEBUG & DEBUG_PRECOMPILE)
+      st_precompile_shader_variant(st, prog);
+
    /* XXX check if program is legal, within limits */
    return GL_TRUE;
 }
diff --git a/mesalib/src/mesa/state_tracker/st_cb_texture.c b/mesalib/src/mesa/state_tracker/st_cb_texture.c
index 5c520b44f..bdf236e82 100644
--- a/mesalib/src/mesa/state_tracker/st_cb_texture.c
+++ b/mesalib/src/mesa/state_tracker/st_cb_texture.c
@@ -123,7 +123,7 @@ gl_target_to_pipe(GLenum target)
 static struct gl_texture_image *
 st_NewTextureImage(struct gl_context * ctx)
 {
-   DBG("%s\n", __FUNCTION__);
+   DBG("%s\n", __func__);
    (void) ctx;
    return (struct gl_texture_image *) ST_CALLOC_STRUCT(st_texture_image);
 }
@@ -144,7 +144,7 @@ st_NewTextureObject(struct gl_context * ctx, GLuint name, GLenum target)
 {
    struct st_texture_object *obj = ST_CALLOC_STRUCT(st_texture_object);
 
-   DBG("%s\n", __FUNCTION__);
+   DBG("%s\n", __func__);
    _mesa_initialize_texture_object(ctx, &obj->base, name, target);
 
    return &obj->base;
@@ -172,7 +172,7 @@ st_FreeTextureImageBuffer(struct gl_context *ctx,
 {
    struct st_texture_image *stImage = st_texture_image(texImage);
 
-   DBG("%s\n", __FUNCTION__);
+   DBG("%s\n", __func__);
 
    if (stImage->pt) {
       pipe_resource_reference(&stImage->pt, NULL);
@@ -405,7 +405,7 @@ guess_and_alloc_texture(struct st_context *st,
    GLuint ptWidth, ptHeight, ptDepth, ptLayers;
    enum pipe_format fmt;
 
-   DBG("%s\n", __FUNCTION__);
+   DBG("%s\n", __func__);
 
    assert(!stObj->pt);
 
@@ -473,7 +473,7 @@ guess_and_alloc_texture(struct st_context *st,
 
    stObj->lastLevel = lastLevel;
 
-   DBG("%s returning %d\n", __FUNCTION__, (stObj->pt != NULL));
+   DBG("%s returning %d\n", __func__, (stObj->pt != NULL));
 
    return stObj->pt != NULL;
 }
@@ -496,7 +496,7 @@ st_AllocTextureImageBuffer(struct gl_context *ctx,
    GLuint height = texImage->Height;
    GLuint depth = texImage->Depth;
 
-   DBG("%s\n", __FUNCTION__);
+   DBG("%s\n", __func__);
 
    assert(!stImage->pt); /* xxx this might be wrong */
 
@@ -738,6 +738,11 @@ st_TexSubImage(struct gl_context *ctx, GLuint dims,
    if (gl_target == GL_TEXTURE_CUBE_MAP) {
       gl_target = GL_TEXTURE_2D;
    }
+   /* TexSubImage can specify subsets of cube map array faces
+    * so we need to upload via 2D array instead */
+   if (gl_target == GL_TEXTURE_CUBE_MAP_ARRAY) {
+      gl_target = GL_TEXTURE_2D_ARRAY;
+   }
 
    /* Initialize the source texture description. */
    memset(&src_templ, 0, sizeof(src_templ));
@@ -1148,7 +1153,7 @@ st_GetTexImage(struct gl_context * ctx,
       }
 
       if (ST_DEBUG & DEBUG_FALLBACK)
-         debug_printf("%s: fallback format translation\n", __FUNCTION__);
+         debug_printf("%s: fallback format translation\n", __func__);
 
       dstMesaFormat = _mesa_format_from_format_and_type(format, type);
       dstStride = _mesa_image_row_stride(&ctx->Pack, width, format, type);
@@ -1234,7 +1239,7 @@ fallback_copy_texsubimage(struct gl_context *ctx,
    struct pipe_transfer *transfer;
 
    if (ST_DEBUG & DEBUG_FALLBACK)
-      debug_printf("%s: fallback processing\n", __FUNCTION__);
+      debug_printf("%s: fallback processing\n", __func__);
 
    if (st_fb_orientation(ctx->ReadBuffer) == Y_0_TOP) {
       srcY = strb->Base.Height - srcY - height;
@@ -1389,7 +1394,7 @@ st_CopyTexSubImage(struct gl_context *ctx, GLuint dims,
           texImage->TexFormat != MESA_FORMAT_ETC1_RGB8);
 
    if (!strb || !strb->surface || !stImage->pt) {
-      debug_printf("%s: null strb or stImage\n", __FUNCTION__);
+      debug_printf("%s: null strb or stImage\n", __func__);
       return;
    }
 
diff --git a/mesalib/src/mesa/state_tracker/st_debug.c b/mesalib/src/mesa/state_tracker/st_debug.c
index de3e3a9cd..50891c112 100644
--- a/mesalib/src/mesa/state_tracker/st_debug.c
+++ b/mesalib/src/mesa/state_tracker/st_debug.c
@@ -56,6 +56,7 @@ static const struct debug_named_value st_debug_flags[] = {
    { "draw",     DEBUG_DRAW, NULL },
    { "buffer",   DEBUG_BUFFER, NULL },
    { "wf",       DEBUG_WIREFRAME, NULL },
+   { "precompile",  DEBUG_PRECOMPILE, NULL },
    DEBUG_NAMED_VALUE_END
 };
 
diff --git a/mesalib/src/mesa/state_tracker/st_debug.h b/mesalib/src/mesa/state_tracker/st_debug.h
index cc8197836..288eccf9f 100644
--- a/mesalib/src/mesa/state_tracker/st_debug.h
+++ b/mesalib/src/mesa/state_tracker/st_debug.h
@@ -47,6 +47,7 @@ st_print_current(void);
 #define DEBUG_DRAW      0x100
 #define DEBUG_BUFFER    0x200
 #define DEBUG_WIREFRAME 0x400
+#define DEBUG_PRECOMPILE   0x800
 
 #ifdef DEBUG
 extern int ST_DEBUG;
diff --git a/mesalib/src/mesa/state_tracker/st_glsl_to_tgsi.cpp b/mesalib/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
index efee4b258..93671ba9c 100644
--- a/mesalib/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
+++ b/mesalib/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
@@ -30,40 +30,25 @@
  * Translate GLSL IR to TGSI.
  */
 
-#include <stdio.h>
-#include "main/compiler.h"
-#include "ir.h"
-#include "ir_visitor.h"
-#include "ir_expression_flattening.h"
-#include "glsl_types.h"
+#include "st_glsl_to_tgsi.h"
+
 #include "glsl_parser_extras.h"
-#include "../glsl/program.h"
 #include "ir_optimization.h"
-#include "ast.h"
 
-#include "main/mtypes.h"
+#include "main/errors.h"
 #include "main/shaderobj.h"
 #include "main/uniforms.h"
 #include "main/shaderapi.h"
-#include "program/hash_table.h"
 #include "program/prog_instruction.h"
-#include "program/prog_optimize.h"
-#include "program/prog_print.h"
-#include "program/program.h"
-#include "program/prog_parameter.h"
 #include "program/sampler.h"
 
-#include "pipe/p_compiler.h"
 #include "pipe/p_context.h"
 #include "pipe/p_screen.h"
-#include "pipe/p_shader_tokens.h"
-#include "pipe/p_state.h"
-#include "util/u_math.h"
 #include "tgsi/tgsi_ureg.h"
 #include "tgsi/tgsi_info.h"
-#include "st_context.h"
+#include "util/u_math.h"
+#include "util/u_memory.h"
 #include "st_program.h"
-#include "st_glsl_to_tgsi.h"
 #include "st_mesa_to_tgsi.h"
 
 
@@ -328,6 +313,7 @@ public:
    int num_address_regs;
    int samplers_used;
    bool indirect_addr_consts;
+   int wpos_transform_const;
 
    int glsl_version;
    bool native_integers;
@@ -441,9 +427,6 @@ public:
 
    void emit_arl(ir_instruction *ir, st_dst_reg dst, st_src_reg src0);
 
-   void emit_scs(ir_instruction *ir, unsigned op,
-                 st_dst_reg dst, const st_src_reg &src);
-
    bool try_emit_mad(ir_expression *ir,
               int mul_operand);
    bool try_emit_mad_for_and_not(ir_expression *ir,
@@ -966,101 +949,6 @@ glsl_to_tgsi_visitor::emit_arl(ir_instruction *ir,
    emit(NULL, op, dst, src0);
 }
 
-/**
- * Emit an TGSI_OPCODE_SCS instruction
- *
- * The \c SCS opcode functions a bit differently than the other TGSI opcodes.
- * Instead of splatting its result across all four components of the
- * destination, it writes one value to the \c x component and another value to
- * the \c y component.
- *
- * \param ir        IR instruction being processed
- * \param op        Either \c TGSI_OPCODE_SIN or \c TGSI_OPCODE_COS depending
- *                  on which value is desired.
- * \param dst       Destination register
- * \param src       Source register
- */
-void
-glsl_to_tgsi_visitor::emit_scs(ir_instruction *ir, unsigned op,
-                               st_dst_reg dst,
-                               const st_src_reg &src)
-{
-   /* Vertex programs cannot use the SCS opcode.
-    */
-   if (this->prog->Target == GL_VERTEX_PROGRAM_ARB) {
-      emit_scalar(ir, op, dst, src);
-      return;
-   }
-
-   const unsigned component = (op == TGSI_OPCODE_SIN) ? 0 : 1;
-   const unsigned scs_mask = (1U << component);
-   int done_mask = ~dst.writemask;
-   st_src_reg tmp;
-
-   assert(op == TGSI_OPCODE_SIN || op == TGSI_OPCODE_COS);
-
-   /* If there are compnents in the destination that differ from the component
-    * that will be written by the SCS instrution, we'll need a temporary.
-    */
-   if (scs_mask != unsigned(dst.writemask)) {
-      tmp = get_temp(glsl_type::vec4_type);
-   }
-
-   for (unsigned i = 0; i < 4; i++) {
-      unsigned this_mask = (1U << i);
-      st_src_reg src0 = src;
-
-      if ((done_mask & this_mask) != 0)
-         continue;
-
-      /* The source swizzle specified which component of the source generates
-       * sine / cosine for the current component in the destination.  The SCS
-       * instruction requires that this value be swizzle to the X component.
-       * Replace the current swizzle with a swizzle that puts the source in
-       * the X component.
-       */
-      unsigned src0_swiz = GET_SWZ(src.swizzle, i);
-
-      src0.swizzle = MAKE_SWIZZLE4(src0_swiz, src0_swiz,
-                                   src0_swiz, src0_swiz);
-      for (unsigned j = i + 1; j < 4; j++) {
-         /* If there is another enabled component in the destination that is
-          * derived from the same inputs, generate its value on this pass as
-          * well.
-          */
-         if (!(done_mask & (1 << j)) &&
-             GET_SWZ(src0.swizzle, j) == src0_swiz) {
-            this_mask |= (1 << j);
-         }
-      }
-
-      if (this_mask != scs_mask) {
-         glsl_to_tgsi_instruction *inst;
-         st_dst_reg tmp_dst = st_dst_reg(tmp);
-
-         /* Emit the SCS instruction.
-          */
-         inst = emit(ir, TGSI_OPCODE_SCS, tmp_dst, src0);
-         inst->dst[0].writemask = scs_mask;
-
-         /* Move the result of the SCS instruction to the desired location in
-          * the destination.
-          */
-         tmp.swizzle = MAKE_SWIZZLE4(component, component,
-                                     component, component);
-         inst = emit(ir, TGSI_OPCODE_SCS, dst, tmp);
-         inst->dst[0].writemask = this_mask;
-      } else {
-         /* Emit the SCS instruction to write directly to the destination.
-          */
-         glsl_to_tgsi_instruction *inst = emit(ir, TGSI_OPCODE_SCS, dst, src0);
-         inst->dst[0].writemask = scs_mask;
-      }
-
-      done_mask |= this_mask;
-   }
-}
-
 int
 glsl_to_tgsi_visitor::add_constant(gl_register_file file,
                                    gl_constant_value values[8], int size, int datatype,
@@ -1611,12 +1499,6 @@ glsl_to_tgsi_visitor::visit(ir_expression *ir)
    case ir_unop_cos:
       emit_scalar(ir, TGSI_OPCODE_COS, result_dst, op[0]);
       break;
-   case ir_unop_sin_reduced:
-      emit_scs(ir, TGSI_OPCODE_SIN, result_dst, op[0]);
-      break;
-   case ir_unop_cos_reduced:
-      emit_scs(ir, TGSI_OPCODE_COS, result_dst, op[0]);
-      break;
    case ir_unop_saturate: {
       glsl_to_tgsi_instruction *inst;
       inst = emit(ir, TGSI_OPCODE_MOV, result_dst, op[0]);
@@ -3134,7 +3016,7 @@ glsl_to_tgsi_visitor::visit(ir_texture *ir)
       break;
    case ir_query_levels:
       opcode = TGSI_OPCODE_TXQ;
-      lod_info = st_src_reg(PROGRAM_IMMEDIATE, 0, GLSL_TYPE_INT);
+      lod_info = undef_src;
       levels_src = get_temp(ir->type);
       break;
    case ir_txf:
@@ -3456,6 +3338,7 @@ glsl_to_tgsi_visitor::glsl_to_tgsi_visitor()
    num_address_regs = 0;
    samplers_used = 0;
    indirect_addr_consts = false;
+   wpos_transform_const = -1;
    glsl_version = 0;
    native_integers = false;
    mem_ctx = ralloc_context(NULL);
@@ -3465,6 +3348,7 @@ glsl_to_tgsi_visitor::glsl_to_tgsi_visitor()
    shader = NULL;
    options = NULL;
    have_sqrt = false;
+   have_fma = false;
 }
 
 glsl_to_tgsi_visitor::~glsl_to_tgsi_visitor()
@@ -3937,6 +3821,7 @@ glsl_to_tgsi_visitor::copy_propagate(void)
              inst->dst[0].index == inst->src[0].index) &&
           !inst->dst[0].reladdr &&
           !inst->saturate &&
+          inst->src[0].file != PROGRAM_ARRAY &&
           !inst->src[0].reladdr &&
           !inst->src[0].reladdr2 &&
           !inst->src[0].negate) {
@@ -4464,7 +4349,9 @@ struct st_translate {
 
    struct ureg_dst arrays[MAX_ARRAYS];
    struct ureg_src *constants;
+   int num_constants;
    struct ureg_src *immediates;
+   int num_immediates;
    struct ureg_dst outputs[PIPE_MAX_SHADER_OUTPUTS];
    struct ureg_src inputs[PIPE_MAX_SHADER_INPUTS];
    struct ureg_dst address[3];
@@ -4666,7 +4553,7 @@ src_register(struct st_translate *t, const st_src_reg *reg)
 {
    switch(reg->file) {
    case PROGRAM_UNDEFINED:
-      return ureg_src_undef();
+      return ureg_imm4f(t->ureg, 0, 0, 0, 0);
 
    case PROGRAM_TEMPORARY:
    case PROGRAM_ARRAY:
@@ -4674,17 +4561,18 @@ src_register(struct st_translate *t, const st_src_reg *reg)
 
    case PROGRAM_UNIFORM:
       assert(reg->index >= 0);
-      return t->constants[reg->index];
+      return reg->index < t->num_constants ?
+               t->constants[reg->index] : ureg_imm4f(t->ureg, 0, 0, 0, 0);
    case PROGRAM_STATE_VAR:
    case PROGRAM_CONSTANT:       /* ie, immediate */
       if (reg->has_index2)
          return ureg_src_register(TGSI_FILE_CONSTANT, reg->index);
-      else if (reg->index < 0)
-         return ureg_DECL_constant(t->ureg, 0);
       else
-         return t->constants[reg->index];
+         return reg->index >= 0 && reg->index < t->num_constants ?
+                  t->constants[reg->index] : ureg_imm4f(t->ureg, 0, 0, 0, 0);
 
    case PROGRAM_IMMEDIATE:
+      assert(reg->index >= 0 && reg->index < t->num_immediates);
       return t->immediates[reg->index];
 
    case PROGRAM_INPUT:
@@ -4805,6 +4693,7 @@ translate_tex_offset(struct st_translate *t,
 
    switch (in_offset->file) {
    case PROGRAM_IMMEDIATE:
+      assert(in_offset->index >= 0 && in_offset->index < t->num_immediates);
       imm_src = t->immediates[in_offset->index];
 
       offset.File = imm_src.File;
@@ -4867,10 +4756,8 @@ compile_tgsi_instruction(struct st_translate *t,
                              inst->saturate,
                              clamp_dst_color_output);
 
-   for (i = 0; i < num_src; i++) {
-      assert(inst->src[i].file != PROGRAM_UNDEFINED);
+   for (i = 0; i < num_src; i++)
       src[i] = translate_src(t, &inst->src[i]);
-   }
 
    switch(inst->op) {
    case TGSI_OPCODE_BGNLOOP:
@@ -4939,28 +4826,19 @@ compile_tgsi_instruction(struct st_translate *t,
  */
 static void
 emit_wpos_adjustment( struct st_translate *t,
-                      const struct gl_program *program,
+                      int wpos_transform_const,
                       boolean invert,
                       GLfloat adjX, GLfloat adjY[2])
 {
    struct ureg_program *ureg = t->ureg;
 
+   assert(wpos_transform_const >= 0);
+
    /* Fragment program uses fragment position input.
     * Need to replace instances of INPUT[WPOS] with temp T
-    * where T = INPUT[WPOS] by y is inverted.
+    * where T = INPUT[WPOS] is inverted by Y.
     */
-   static const gl_state_index wposTransformState[STATE_LENGTH]
-      = { STATE_INTERNAL, STATE_FB_WPOS_Y_TRANSFORM,
-          (gl_state_index)0, (gl_state_index)0, (gl_state_index)0 };
-
-   /* XXX: note we are modifying the incoming shader here!  Need to
-    * do this before emitting the constant decls below, or this
-    * will be missed:
-    */
-   unsigned wposTransConst = _mesa_add_state_reference(program->Parameters,
-                                                       wposTransformState);
-
-   struct ureg_src wpostrans = ureg_DECL_constant( ureg, wposTransConst );
+   struct ureg_src wpostrans = ureg_DECL_constant(ureg, wpos_transform_const);
    struct ureg_dst wpos_temp = ureg_DECL_temporary( ureg );
    struct ureg_src wpos_input = t->inputs[t->inputMapping[VARYING_SLOT_POS]];
 
@@ -5024,7 +4902,8 @@ static void
 emit_wpos(struct st_context *st,
           struct st_translate *t,
           const struct gl_program *program,
-          struct ureg_program *ureg)
+          struct ureg_program *ureg,
+          int wpos_transform_const)
 {
    const struct gl_fragment_program *fp =
       (const struct gl_fragment_program *) program;
@@ -5121,7 +5000,7 @@ emit_wpos(struct st_context *st,
 
    /* we invert after adjustment so that we avoid the MOV to temporary,
     * and reuse the adjustment ADD instead */
-   emit_wpos_adjustment(t, program, invert, adjX, adjY);
+   emit_wpos_adjustment(t, wpos_transform_const, invert, adjX, adjY);
 }
 
 /**
@@ -5238,15 +5117,6 @@ st_translate_program(
    t->outputMapping = outputMapping;
    t->ureg = ureg;
 
-   if (program->shader_program) {
-      for (i = 0; i < program->shader_program->NumUserUniformStorage; i++) {
-         struct gl_uniform_storage *const storage =
-               &program->shader_program->UniformStorage[i];
-
-         _mesa_uniform_detach_all_driver_storage(storage);
-      }
-   }
-
    /*
     * Declare input attributes.
     */
@@ -5260,10 +5130,9 @@ st_translate_program(
       }
 
       if (proginfo->InputsRead & VARYING_BIT_POS) {
-         /* Must do this after setting up t->inputs, and before
-          * emitting constant references, below:
-          */
-          emit_wpos(st_context(ctx), t, proginfo, ureg);
+          /* Must do this after setting up t->inputs. */
+          emit_wpos(st_context(ctx), t, proginfo, ureg,
+                    program->wpos_transform_const);
       }
 
       if (proginfo->InputsRead & VARYING_BIT_FACE)
@@ -5400,6 +5269,7 @@ st_translate_program(
          ret = PIPE_ERROR_OUT_OF_MEMORY;
          goto out;
       }
+      t->num_constants = proginfo->Parameters->NumParameters;
 
       for (i = 0; i < proginfo->Parameters->NumParameters; i++) {
          switch (proginfo->Parameters->Parameters[i].Type) {
@@ -5451,6 +5321,8 @@ st_translate_program(
       ret = PIPE_ERROR_OUT_OF_MEMORY;
       goto out;
    }
+   t->num_immediates = program->num_immediates;
+
    i = 0;
    foreach_in_list(immediate_storage, imm, &program->immediates) {
       assert(i < program->num_immediates);
@@ -5479,33 +5351,21 @@ st_translate_program(
                        t->insn[t->labels[i].branch_target]);
    }
 
-   if (program->shader_program) {
-      /* This has to be done last.  Any operation the can cause
-       * prog->ParameterValues to get reallocated (e.g., anything that adds a
-       * program constant) has to happen before creating this linkage.
-       */
-      for (unsigned i = 0; i < MESA_SHADER_STAGES; i++) {
-         if (program->shader_program->_LinkedShaders[i] == NULL)
-            continue;
-
-         _mesa_associate_uniform_storage(ctx, program->shader_program,
-               program->shader_program->_LinkedShaders[i]->Program->Parameters);
-      }
-   }
-
 out:
    if (t) {
       free(t->temps);
       free(t->insn);
       free(t->labels);
       free(t->constants);
+      t->num_constants = 0;
       free(t->immediates);
+      t->num_immediates = 0;
 
       if (t->error) {
-         debug_printf("%s: translate error flag set\n", __FUNCTION__);
+         debug_printf("%s: translate error flag set\n", __func__);
       }
 
-      free(t);
+      FREE(t);
    }
 
    return ret;
@@ -5634,14 +5494,12 @@ get_mesa_program(struct gl_context *ctx,
    v->emit(NULL, TGSI_OPCODE_END);
 
    if (ctx->_Shader->Flags & GLSL_DUMP) {
-      printf("\n");
-      printf("GLSL IR for linked %s program %d:\n",
+      _mesa_log("\n");
+      _mesa_log("GLSL IR for linked %s program %d:\n",
              _mesa_shader_stage_to_string(shader->Stage),
              shader_program->Name);
-      _mesa_print_ir(stdout, shader->ir, NULL);
-      printf("\n");
-      printf("\n");
-      fflush(stdout);
+      _mesa_print_ir(_mesa_get_log_file(), shader->ir, NULL);
+      _mesa_log("\n\n");
    }
 
    prog->Instructions = NULL;
@@ -5650,6 +5508,17 @@ get_mesa_program(struct gl_context *ctx,
    do_set_program_inouts(shader->ir, prog, shader->Stage);
    count_resources(v, prog);
 
+   /* This must be done before the uniform storage is associated. */
+   if (shader->Type == GL_FRAGMENT_SHADER &&
+       prog->InputsRead & VARYING_BIT_POS){
+      static const gl_state_index wposTransformState[STATE_LENGTH] = {
+         STATE_INTERNAL, STATE_FB_WPOS_Y_TRANSFORM
+      };
+
+      v->wpos_transform_const = _mesa_add_state_reference(prog->Parameters,
+                                                          wposTransformState);
+   }
+
    _mesa_reference_program(ctx, &shader->Program, prog);
 
    /* This has to be done last.  Any operation the can cause
diff --git a/mesalib/src/mesa/state_tracker/st_glsl_to_tgsi.h b/mesalib/src/mesa/state_tracker/st_glsl_to_tgsi.h
index 5ed640747..2cb80bcf9 100644
--- a/mesalib/src/mesa/state_tracker/st_glsl_to_tgsi.h
+++ b/mesalib/src/mesa/state_tracker/st_glsl_to_tgsi.h
@@ -22,17 +22,18 @@
  * DEALINGS IN THE SOFTWARE.
  */
 
+#include "pipe/p_defines.h"
+#include "main/mtypes.h"
+
 #ifdef __cplusplus
 extern "C" {
 #endif
 
-#include "main/glheader.h"
-#include "tgsi/tgsi_ureg.h"
-
 struct gl_context;
 struct gl_shader;
 struct gl_shader_program;
 struct glsl_to_tgsi_visitor;
+struct ureg_program;
 
 enum pipe_error st_translate_program(
    struct gl_context *ctx,
diff --git a/mesalib/src/mesa/state_tracker/st_mesa_to_tgsi.c b/mesalib/src/mesa/state_tracker/st_mesa_to_tgsi.c
index 2f1016110..98d525c86 100644
--- a/mesalib/src/mesa/state_tracker/st_mesa_to_tgsi.c
+++ b/mesalib/src/mesa/state_tracker/st_mesa_to_tgsi.c
@@ -1249,7 +1249,7 @@ out:
    free(t->constants);
 
    if (t->error) {
-      debug_printf("%s: translate error flag set\n", __FUNCTION__);
+      debug_printf("%s: translate error flag set\n", __func__);
    }
 
    return ret;
diff --git a/mesalib/src/mesa/state_tracker/st_program.c b/mesalib/src/mesa/state_tracker/st_program.c
index 4cfd817ce..d93b3c7bc 100644
--- a/mesalib/src/mesa/state_tracker/st_program.c
+++ b/mesalib/src/mesa/state_tracker/st_program.c
@@ -185,9 +185,6 @@ st_prepare_vertex_program(struct gl_context *ctx,
    if (stvp->Base.IsPositionInvariant)
       _mesa_insert_mvp_code(ctx, &stvp->Base);
 
-   if (!stvp->glsl_to_tgsi)
-      assert(stvp->Base.Base.NumInstructions > 1);
-
    /*
     * Determine number of inputs, the mappings between VERT_ATTRIB_x
     * and TGSI generic input indexes, plus input attrib semantic info.
@@ -403,7 +400,7 @@ st_translate_vertex_program(struct st_context *st,
    return vpv;
 
 fail:
-   debug_printf("%s: failed to translate Mesa program:\n", __FUNCTION__);
+   debug_printf("%s: failed to translate Mesa program:\n", __func__);
    _mesa_print_program(&stvp->Base.Base);
    debug_assert(0);
 
@@ -1318,3 +1315,47 @@ st_print_current_vertex_program(void)
       }
    }
 }
+
+
+/**
+ * Compile one shader variant.
+ */
+void
+st_precompile_shader_variant(struct st_context *st,
+                             struct gl_program *prog)
+{
+   switch (prog->Target) {
+   case GL_VERTEX_PROGRAM_ARB: {
+      struct st_vertex_program *p = (struct st_vertex_program *)prog;
+      struct st_vp_variant_key key;
+
+      memset(&key, 0, sizeof(key));
+      key.st = st;
+      st_get_vp_variant(st, p, &key);
+      break;
+   }
+
+   case GL_GEOMETRY_PROGRAM_NV: {
+      struct st_geometry_program *p = (struct st_geometry_program *)prog;
+      struct st_gp_variant_key key;
+
+      memset(&key, 0, sizeof(key));
+      key.st = st;
+      st_get_gp_variant(st, p, &key);
+      break;
+   }
+
+   case GL_FRAGMENT_PROGRAM_ARB: {
+      struct st_fragment_program *p = (struct st_fragment_program *)prog;
+      struct st_fp_variant_key key;
+
+      memset(&key, 0, sizeof(key));
+      key.st = st;
+      st_get_fp_variant(st, p, &key);
+      break;
+   }
+
+   default:
+      assert(0);
+   }
+}
diff --git a/mesalib/src/mesa/state_tracker/st_program.h b/mesalib/src/mesa/state_tracker/st_program.h
index 451d7bb6a..b2c86faec 100644
--- a/mesalib/src/mesa/state_tracker/st_program.h
+++ b/mesalib/src/mesa/state_tracker/st_program.h
@@ -329,6 +329,9 @@ st_destroy_program_variants(struct st_context *st);
 extern void
 st_print_current_vertex_program(void);
 
+extern void
+st_precompile_shader_variant(struct st_context *st,
+                             struct gl_program *prog);
 
 #ifdef __cplusplus
 }
diff --git a/mesalib/src/mesa/state_tracker/st_texture.c b/mesalib/src/mesa/state_tracker/st_texture.c
index ca7c83c21..6beb21e33 100644
--- a/mesalib/src/mesa/state_tracker/st_texture.c
+++ b/mesalib/src/mesa/state_tracker/st_texture.c
@@ -74,7 +74,7 @@ st_texture_create(struct st_context *st,
    if (target == PIPE_TEXTURE_CUBE)
       assert(layers == 6);
 
-   DBG("%s target %d format %s last_level %d\n", __FUNCTION__,
+   DBG("%s target %d format %s last_level %d\n", __func__,
        (int) target, util_format_name(format), last_level);
 
    assert(format);
@@ -177,7 +177,7 @@ st_gl_texture_dims_to_pipe_dims(GLenum texture,
       *widthOut = widthIn;
       *heightOut = heightIn;
       *depthOut = 1;
-      *layersOut = depthIn;
+      *layersOut = util_align_npot(depthIn, 6);
       break;
    default:
       assert(0 && "Unexpected texture in st_gl_texture_dims_to_pipe_dims()");
@@ -250,7 +250,7 @@ st_texture_image_map(struct st_context *st, struct st_texture_image *stImage,
    GLuint level;
    void *map;
 
-   DBG("%s \n", __FUNCTION__);
+   DBG("%s \n", __func__);
 
    if (!stImage->pt)
       return NULL;
@@ -304,7 +304,7 @@ st_texture_image_unmap(struct st_context *st,
       slice += stObj->base.MinLayer;
    transfer = &stImage->transfer[slice + stImage->base.Face].transfer;
 
-   DBG("%s\n", __FUNCTION__);
+   DBG("%s\n", __func__);
 
    pipe_transfer_unmap(pipe, *transfer);
    *transfer = NULL;