r500: fragprog
authorDave Airlie <airlied@redhat.com>
Mon, 17 Mar 2008 09:05:15 +0000 (19:05 +1000)
committerDave Airlie <airlied@redhat.com>
Mon, 17 Mar 2008 09:05:44 +0000 (19:05 +1000)
src/mesa/drivers/dri/r300/Makefile
src/mesa/drivers/dri/r300/r500_fragprog.c [new file with mode: 0644]
src/mesa/drivers/dri/r300/r500_fragprog.h [new file with mode: 0644]

index 44248964fdb7d889b1f3528c1b7aad8c046b97f1..5b2bd0bc2b005432252c78622eb00420d419222d 100644 (file)
@@ -39,6 +39,7 @@ DRIVER_SOURCES = \
                 r300_texstate.c \
                 r300_vertprog.c \
                 r300_fragprog.c \
+                r500_fragprog.c \
                 r300_shader.c \
                 r300_emit.c \
                 r300_swtcl.c \
diff --git a/src/mesa/drivers/dri/r300/r500_fragprog.c b/src/mesa/drivers/dri/r300/r500_fragprog.c
new file mode 100644 (file)
index 0000000..3638a94
--- /dev/null
@@ -0,0 +1,2476 @@
+/*
+ * Copyright (C) 2005 Ben Skeggs.
+ *
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+/**
+ * \file
+ *
+ * \author Ben Skeggs <darktama@iinet.net.au>
+ *
+ * \author Jerome Glisse <j.glisse@gmail.com>
+ *
+ * \todo Depth write, WPOS/FOGC inputs
+ *
+ * \todo FogOption
+ *
+ * \todo Verify results of opcodes for accuracy, I've only checked them in
+ * specific cases.
+ */
+
+#include "glheader.h"
+#include "macros.h"
+#include "enums.h"
+#include "shader/prog_instruction.h"
+#include "shader/prog_parameter.h"
+#include "shader/prog_print.h"
+
+#include "r300_context.h"
+#include "r300_fragprog.h"
+#include "r300_reg.h"
+#include "r300_state.h"
+
+/*
+ * Usefull macros and values
+ */
+#define ERROR(fmt, args...) do {                       \
+               fprintf(stderr, "%s::%s(): " fmt "\n",  \
+                       __FILE__, __FUNCTION__, ##args);        \
+               fp->error = GL_TRUE;                    \
+       } while(0)
+
+#define PFS_INVAL 0xFFFFFFFF
+#define COMPILE_STATE struct r300_pfs_compile_state *cs = fp->cs
+
+#define SWIZZLE_XYZ            0
+#define SWIZZLE_XXX            1
+#define SWIZZLE_YYY            2
+#define SWIZZLE_ZZZ            3
+#define SWIZZLE_WWW            4
+#define SWIZZLE_YZX            5
+#define SWIZZLE_ZXY            6
+#define SWIZZLE_WZY            7
+#define SWIZZLE_111            8
+#define SWIZZLE_000            9
+#define SWIZZLE_HHH            10
+
+#define swizzle(r, x, y, z, w) do_swizzle(fp, r,               \
+                                         ((SWIZZLE_##x<<0)|    \
+                                          (SWIZZLE_##y<<3)|    \
+                                          (SWIZZLE_##z<<6)|    \
+                                          (SWIZZLE_##w<<9)),   \
+                                         0)
+
+#define REG_TYPE_INPUT         0
+#define REG_TYPE_OUTPUT                1
+#define REG_TYPE_TEMP          2
+#define REG_TYPE_CONST         3
+
+#define REG_TYPE_SHIFT         0
+#define REG_INDEX_SHIFT                2
+#define REG_VSWZ_SHIFT         8
+#define REG_SSWZ_SHIFT         13
+#define REG_NEGV_SHIFT         18
+#define REG_NEGS_SHIFT         19
+#define REG_ABS_SHIFT          20
+#define REG_NO_USE_SHIFT       21      // Hack for refcounting
+#define REG_VALID_SHIFT                22      // Does the register contain a defined value?
+#define REG_BUILTIN_SHIFT   23 // Is it a builtin (like all zero/all one)?
+
+#define REG_TYPE_MASK          (0x03 << REG_TYPE_SHIFT)
+#define REG_INDEX_MASK         (0x3F << REG_INDEX_SHIFT)
+#define REG_VSWZ_MASK          (0x1F << REG_VSWZ_SHIFT)
+#define REG_SSWZ_MASK          (0x1F << REG_SSWZ_SHIFT)
+#define REG_NEGV_MASK          (0x01 << REG_NEGV_SHIFT)
+#define REG_NEGS_MASK          (0x01 << REG_NEGS_SHIFT)
+#define REG_ABS_MASK           (0x01 << REG_ABS_SHIFT)
+#define REG_NO_USE_MASK                (0x01 << REG_NO_USE_SHIFT)
+#define REG_VALID_MASK         (0x01 << REG_VALID_SHIFT)
+#define REG_BUILTIN_MASK       (0x01 << REG_BUILTIN_SHIFT)
+
+#define REG(type, index, vswz, sswz, nouse, valid, builtin)    \
+       (((type << REG_TYPE_SHIFT) & REG_TYPE_MASK) |                   \
+        ((index << REG_INDEX_SHIFT) & REG_INDEX_MASK) |                \
+        ((nouse << REG_NO_USE_SHIFT) & REG_NO_USE_MASK) |              \
+        ((valid << REG_VALID_SHIFT) & REG_VALID_MASK) |                \
+        ((builtin << REG_BUILTIN_SHIFT) & REG_BUILTIN_MASK) |  \
+        ((vswz << REG_VSWZ_SHIFT) & REG_VSWZ_MASK) |                   \
+        ((sswz << REG_SSWZ_SHIFT) & REG_SSWZ_MASK))
+#define REG_GET_TYPE(reg)                                              \
+       ((reg & REG_TYPE_MASK) >> REG_TYPE_SHIFT)
+#define REG_GET_INDEX(reg)                                             \
+       ((reg & REG_INDEX_MASK) >> REG_INDEX_SHIFT)
+#define REG_GET_VSWZ(reg)                                              \
+       ((reg & REG_VSWZ_MASK) >> REG_VSWZ_SHIFT)
+#define REG_GET_SSWZ(reg)                                              \
+       ((reg & REG_SSWZ_MASK) >> REG_SSWZ_SHIFT)
+#define REG_GET_NO_USE(reg)                                            \
+       ((reg & REG_NO_USE_MASK) >> REG_NO_USE_SHIFT)
+#define REG_GET_VALID(reg)                                             \
+       ((reg & REG_VALID_MASK) >> REG_VALID_SHIFT)
+#define REG_GET_BUILTIN(reg)                                           \
+       ((reg & REG_BUILTIN_MASK) >> REG_BUILTIN_SHIFT)
+#define REG_SET_TYPE(reg, type)                                                \
+       reg = ((reg & ~REG_TYPE_MASK) |                                 \
+              ((type << REG_TYPE_SHIFT) & REG_TYPE_MASK))
+#define REG_SET_INDEX(reg, index)                                      \
+       reg = ((reg & ~REG_INDEX_MASK) |                                \
+              ((index << REG_INDEX_SHIFT) & REG_INDEX_MASK))
+#define REG_SET_VSWZ(reg, vswz)                                                \
+       reg = ((reg & ~REG_VSWZ_MASK) |                                 \
+              ((vswz << REG_VSWZ_SHIFT) & REG_VSWZ_MASK))
+#define REG_SET_SSWZ(reg, sswz)                                                \
+       reg = ((reg & ~REG_SSWZ_MASK) |                                 \
+              ((sswz << REG_SSWZ_SHIFT) & REG_SSWZ_MASK))
+#define REG_SET_NO_USE(reg, nouse)                                     \
+       reg = ((reg & ~REG_NO_USE_MASK) |                               \
+              ((nouse << REG_NO_USE_SHIFT) & REG_NO_USE_MASK))
+#define REG_SET_VALID(reg, valid)                                      \
+       reg = ((reg & ~REG_VALID_MASK) |                                \
+              ((valid << REG_VALID_SHIFT) & REG_VALID_MASK))
+#define REG_SET_BUILTIN(reg, builtin)                                  \
+       reg = ((reg & ~REG_BUILTIN_MASK) |                              \
+              ((builtin << REG_BUILTIN_SHIFT) & REG_BUILTIN_MASK))
+#define REG_ABS(reg)                                                   \
+       reg = (reg | REG_ABS_MASK)
+#define REG_NEGV(reg)                                                  \
+       reg = (reg | REG_NEGV_MASK)
+#define REG_NEGS(reg)                                                  \
+       reg = (reg | REG_NEGS_MASK)
+
+/*
+ * Datas structures for fragment program generation
+ */
+
+/* description of r300 native hw instructions */
+static const struct {
+       const char *name;
+       int argc;
+       int v_op;
+       int s_op;
+} r300_fpop[] = {
+       /* *INDENT-OFF* */
+       {"MAD", 3, R300_FPI0_OUTC_MAD, R300_FPI2_OUTA_MAD},
+       {"DP3", 2, R300_FPI0_OUTC_DP3, R300_FPI2_OUTA_DP4},
+       {"DP4", 2, R300_FPI0_OUTC_DP4, R300_FPI2_OUTA_DP4},
+       {"MIN", 2, R300_FPI0_OUTC_MIN, R300_FPI2_OUTA_MIN},
+       {"MAX", 2, R300_FPI0_OUTC_MAX, R300_FPI2_OUTA_MAX},
+       {"CMP", 3, R300_FPI0_OUTC_CMP, R300_FPI2_OUTA_CMP},
+       {"FRC", 1, R300_FPI0_OUTC_FRC, R300_FPI2_OUTA_FRC},
+       {"EX2", 1, R300_FPI0_OUTC_REPL_ALPHA, R300_FPI2_OUTA_EX2},
+       {"LG2", 1, R300_FPI0_OUTC_REPL_ALPHA, R300_FPI2_OUTA_LG2},
+       {"RCP", 1, R300_FPI0_OUTC_REPL_ALPHA, R300_FPI2_OUTA_RCP},
+       {"RSQ", 1, R300_FPI0_OUTC_REPL_ALPHA, R300_FPI2_OUTA_RSQ},
+       {"REPL_ALPHA", 1, R300_FPI0_OUTC_REPL_ALPHA, PFS_INVAL},
+       {"CMPH", 3, R300_FPI0_OUTC_CMPH, PFS_INVAL},
+       /* *INDENT-ON* */
+};
+
+/* vector swizzles r300 can support natively, with a couple of
+ * cases we handle specially
+ *
+ * REG_VSWZ/REG_SSWZ is an index into this table
+ */
+
+/* mapping from SWIZZLE_* to r300 native values for scalar insns */
+#define SWIZZLE_HALF 6
+
+#define MAKE_SWZ3(x, y, z) (MAKE_SWIZZLE4(SWIZZLE_##x, \
+                                         SWIZZLE_##y, \
+                                         SWIZZLE_##z, \
+                                         SWIZZLE_ZERO))
+/* native swizzles */
+static const struct r300_pfs_swizzle {
+       GLuint hash;            /* swizzle value this matches */
+       GLuint base;            /* base value for hw swizzle */
+       GLuint stride;          /* difference in base between arg0/1/2 */
+       GLuint flags;
+} v_swiz[] = {
+       /* *INDENT-OFF* */
+       {MAKE_SWZ3(X, Y, Z), R300_FPI0_ARGC_SRC0C_XYZ, 4, SLOT_SRC_VECTOR},
+       {MAKE_SWZ3(X, X, X), R300_FPI0_ARGC_SRC0C_XXX, 4, SLOT_SRC_VECTOR},
+       {MAKE_SWZ3(Y, Y, Y), R300_FPI0_ARGC_SRC0C_YYY, 4, SLOT_SRC_VECTOR},
+       {MAKE_SWZ3(Z, Z, Z), R300_FPI0_ARGC_SRC0C_ZZZ, 4, SLOT_SRC_VECTOR},
+       {MAKE_SWZ3(W, W, W), R300_FPI0_ARGC_SRC0A, 1, SLOT_SRC_SCALAR},
+       {MAKE_SWZ3(Y, Z, X), R300_FPI0_ARGC_SRC0C_YZX, 1, SLOT_SRC_VECTOR},
+       {MAKE_SWZ3(Z, X, Y), R300_FPI0_ARGC_SRC0C_ZXY, 1, SLOT_SRC_VECTOR},
+       {MAKE_SWZ3(W, Z, Y), R300_FPI0_ARGC_SRC0CA_WZY, 1, SLOT_SRC_BOTH},
+       {MAKE_SWZ3(ONE, ONE, ONE), R300_FPI0_ARGC_ONE, 0, 0},
+       {MAKE_SWZ3(ZERO, ZERO, ZERO), R300_FPI0_ARGC_ZERO, 0, 0},
+       {MAKE_SWZ3(HALF, HALF, HALF), R300_FPI0_ARGC_HALF, 0, 0},
+       {PFS_INVAL, 0, 0, 0},
+       /* *INDENT-ON* */
+};
+
+/* used during matching of non-native swizzles */
+#define SWZ_X_MASK (7 << 0)
+#define SWZ_Y_MASK (7 << 3)
+#define SWZ_Z_MASK (7 << 6)
+#define SWZ_W_MASK (7 << 9)
+static const struct {
+       GLuint hash;            /* used to mask matching swizzle components */
+       int mask;               /* actual outmask */
+       int count;              /* count of components matched */
+} s_mask[] = {
+       /* *INDENT-OFF* */
+       {SWZ_X_MASK | SWZ_Y_MASK | SWZ_Z_MASK, 1 | 2 | 4, 3},
+       {SWZ_X_MASK | SWZ_Y_MASK, 1 | 2, 2},
+       {SWZ_X_MASK | SWZ_Z_MASK, 1 | 4, 2},
+       {SWZ_Y_MASK | SWZ_Z_MASK, 2 | 4, 2},
+       {SWZ_X_MASK, 1, 1},
+       {SWZ_Y_MASK, 2, 1},
+       {SWZ_Z_MASK, 4, 1},
+       {PFS_INVAL, PFS_INVAL, PFS_INVAL}
+       /* *INDENT-ON* */
+};
+
+static const struct {
+       int base;               /* hw value of swizzle */
+       int stride;             /* difference between SRC0/1/2 */
+       GLuint flags;
+} s_swiz[] = {
+       /* *INDENT-OFF* */
+       {R300_FPI2_ARGA_SRC0C_X, 3, SLOT_SRC_VECTOR},
+       {R300_FPI2_ARGA_SRC0C_Y, 3, SLOT_SRC_VECTOR},
+       {R300_FPI2_ARGA_SRC0C_Z, 3, SLOT_SRC_VECTOR},
+       {R300_FPI2_ARGA_SRC0A, 1, SLOT_SRC_SCALAR},
+       {R300_FPI2_ARGA_ZERO, 0, 0},
+       {R300_FPI2_ARGA_ONE, 0, 0},
+       {R300_FPI2_ARGA_HALF, 0, 0}
+       /* *INDENT-ON* */
+};
+
+/* boiler-plate reg, for convenience */
+static const GLuint undef = REG(REG_TYPE_TEMP,
+                               0,
+                               SWIZZLE_XYZ,
+                               SWIZZLE_W,
+                               GL_FALSE,
+                               GL_FALSE,
+                               GL_FALSE);
+
+/* constant one source */
+static const GLuint pfs_one = REG(REG_TYPE_CONST,
+                                 0,
+                                 SWIZZLE_111,
+                                 SWIZZLE_ONE,
+                                 GL_FALSE,
+                                 GL_TRUE,
+                                 GL_TRUE);
+
+/* constant half source */
+static const GLuint pfs_half = REG(REG_TYPE_CONST,
+                                  0,
+                                  SWIZZLE_HHH,
+                                  SWIZZLE_HALF,
+                                  GL_FALSE,
+                                  GL_TRUE,
+                                  GL_TRUE);
+
+/* constant zero source */
+static const GLuint pfs_zero = REG(REG_TYPE_CONST,
+                                  0,
+                                  SWIZZLE_000,
+                                  SWIZZLE_ZERO,
+                                  GL_FALSE,
+                                  GL_TRUE,
+                                  GL_TRUE);
+
+/*
+ * Common functions prototypes
+ */
+static void dump_program(struct r300_fragment_program *fp);
+static void emit_arith(struct r300_fragment_program *fp, int op,
+                      GLuint dest, int mask,
+                      GLuint src0, GLuint src1, GLuint src2, int flags);
+
+/**
+ * Get an R300 temporary that can be written to in the given slot.
+ */
+static int get_hw_temp(struct r300_fragment_program *fp, int slot)
+{
+       COMPILE_STATE;
+       int r;
+
+       for (r = 0; r < PFS_NUM_TEMP_REGS; ++r) {
+               if (cs->hwtemps[r].free >= 0 && cs->hwtemps[r].free <= slot)
+                       break;
+       }
+
+       if (r >= PFS_NUM_TEMP_REGS) {
+               ERROR("Out of hardware temps\n");
+               return 0;
+       }
+       // Reserved is used to avoid the following scenario:
+       //  R300 temporary X is first assigned to Mesa temporary Y during vector ops
+       //  R300 temporary X is then assigned to Mesa temporary Z for further vector ops
+       //  Then scalar ops on Mesa temporary Z are emitted and move back in time
+       //  to overwrite the value of temporary Y.
+       // End scenario.
+       cs->hwtemps[r].reserved = cs->hwtemps[r].free;
+       cs->hwtemps[r].free = -1;
+
+       // Reset to some value that won't mess things up when the user
+       // tries to read from a temporary that hasn't been assigned a value yet.
+       // In the normal case, vector_valid and scalar_valid should be set to
+       // a sane value by the first emit that writes to this temporary.
+       cs->hwtemps[r].vector_valid = 0;
+       cs->hwtemps[r].scalar_valid = 0;
+
+       if (r > fp->max_temp_idx)
+               fp->max_temp_idx = r;
+
+       return r;
+}
+
+/**
+ * Get an R300 temporary that will act as a TEX destination register.
+ */
+static int get_hw_temp_tex(struct r300_fragment_program *fp)
+{
+       COMPILE_STATE;
+       int r;
+
+       for (r = 0; r < PFS_NUM_TEMP_REGS; ++r) {
+               if (cs->used_in_node & (1 << r))
+                       continue;
+
+               // Note: Be very careful here
+               if (cs->hwtemps[r].free >= 0 && cs->hwtemps[r].free <= 0)
+                       break;
+       }
+
+       if (r >= PFS_NUM_TEMP_REGS)
+               return get_hw_temp(fp, 0);      /* Will cause an indirection */
+
+       cs->hwtemps[r].reserved = cs->hwtemps[r].free;
+       cs->hwtemps[r].free = -1;
+
+       // Reset to some value that won't mess things up when the user
+       // tries to read from a temporary that hasn't been assigned a value yet.
+       // In the normal case, vector_valid and scalar_valid should be set to
+       // a sane value by the first emit that writes to this temporary.
+       cs->hwtemps[r].vector_valid = cs->nrslots;
+       cs->hwtemps[r].scalar_valid = cs->nrslots;
+
+       if (r > fp->max_temp_idx)
+               fp->max_temp_idx = r;
+
+       return r;
+}
+
+/**
+ * Mark the given hardware register as free.
+ */
+static void free_hw_temp(struct r300_fragment_program *fp, int idx)
+{
+       COMPILE_STATE;
+
+       // Be very careful here. Consider sequences like
+       //  MAD r0, r1,r2,r3
+       //  TEX r4, ...
+       // The TEX instruction may be moved in front of the MAD instruction
+       // due to the way nodes work. We don't want to alias r1 and r4 in
+       // this case.
+       // I'm certain the register allocation could be further sanitized,
+       // but it's tricky because of stuff that can happen inside emit_tex
+       // and emit_arith.
+       cs->hwtemps[idx].free = cs->nrslots + 1;
+}
+
+/**
+ * Create a new Mesa temporary register.
+ */
+static GLuint get_temp_reg(struct r300_fragment_program *fp)
+{
+       COMPILE_STATE;
+       GLuint r = undef;
+       GLuint index;
+
+       index = ffs(~cs->temp_in_use);
+       if (!index) {
+               ERROR("Out of program temps\n");
+               return r;
+       }
+
+       cs->temp_in_use |= (1 << --index);
+       cs->temps[index].refcount = 0xFFFFFFFF;
+       cs->temps[index].reg = -1;
+
+       REG_SET_TYPE(r, REG_TYPE_TEMP);
+       REG_SET_INDEX(r, index);
+       REG_SET_VALID(r, GL_TRUE);
+       return r;
+}
+
+/**
+ * Create a new Mesa temporary register that will act as the destination
+ * register for a texture read.
+ */
+static GLuint get_temp_reg_tex(struct r300_fragment_program *fp)
+{
+       COMPILE_STATE;
+       GLuint r = undef;
+       GLuint index;
+
+       index = ffs(~cs->temp_in_use);
+       if (!index) {
+               ERROR("Out of program temps\n");
+               return r;
+       }
+
+       cs->temp_in_use |= (1 << --index);
+       cs->temps[index].refcount = 0xFFFFFFFF;
+       cs->temps[index].reg = get_hw_temp_tex(fp);
+
+       REG_SET_TYPE(r, REG_TYPE_TEMP);
+       REG_SET_INDEX(r, index);
+       REG_SET_VALID(r, GL_TRUE);
+       return r;
+}
+
+/**
+ * Free a Mesa temporary and the associated R300 temporary.
+ */
+static void free_temp(struct r300_fragment_program *fp, GLuint r)
+{
+       COMPILE_STATE;
+       GLuint index = REG_GET_INDEX(r);
+
+       if (!(cs->temp_in_use & (1 << index)))
+               return;
+
+       if (REG_GET_TYPE(r) == REG_TYPE_TEMP) {
+               free_hw_temp(fp, cs->temps[index].reg);
+               cs->temps[index].reg = -1;
+               cs->temp_in_use &= ~(1 << index);
+       } else if (REG_GET_TYPE(r) == REG_TYPE_INPUT) {
+               free_hw_temp(fp, cs->inputs[index].reg);
+               cs->inputs[index].reg = -1;
+       }
+}
+
+/**
+ * Emit a hardware constant/parameter.
+ *
+ * \p cp Stable pointer to an array of 4 floats.
+ *  The pointer must be stable in the sense that it remains to be valid
+ *  and hold the contents of the constant/parameter throughout the lifetime
+ *  of the fragment program (actually, up until the next time the fragment
+ *  program is translated).
+ */
+static GLuint emit_const4fv(struct r300_fragment_program *fp,
+                           const GLfloat * cp)
+{
+       GLuint reg = undef;
+       int index;
+
+       for (index = 0; index < fp->const_nr; ++index) {
+               if (fp->constant[index] == cp)
+                       break;
+       }
+
+       if (index >= fp->const_nr) {
+               if (index >= PFS_NUM_CONST_REGS) {
+                       ERROR("Out of hw constants!\n");
+                       return reg;
+               }
+
+               fp->const_nr++;
+               fp->constant[index] = cp;
+       }
+
+       REG_SET_TYPE(reg, REG_TYPE_CONST);
+       REG_SET_INDEX(reg, index);
+       REG_SET_VALID(reg, GL_TRUE);
+       return reg;
+}
+
+static inline GLuint negate(GLuint r)
+{
+       REG_NEGS(r);
+       REG_NEGV(r);
+       return r;
+}
+
+/* Hack, to prevent clobbering sources used multiple times when
+ * emulating non-native instructions
+ */
+static inline GLuint keep(GLuint r)
+{
+       REG_SET_NO_USE(r, GL_TRUE);
+       return r;
+}
+
+static inline GLuint absolute(GLuint r)
+{
+       REG_ABS(r);
+       return r;
+}
+
+static int swz_native(struct r300_fragment_program *fp,
+                     GLuint src, GLuint * r, GLuint arbneg)
+{
+       /* Native swizzle, handle negation */
+       src = (src & ~REG_NEGS_MASK) | (((arbneg >> 3) & 1) << REG_NEGS_SHIFT);
+
+       if ((arbneg & 0x7) == 0x0) {
+               src = src & ~REG_NEGV_MASK;
+               *r = src;
+       } else if ((arbneg & 0x7) == 0x7) {
+               src |= REG_NEGV_MASK;
+               *r = src;
+       } else {
+               if (!REG_GET_VALID(*r))
+                       *r = get_temp_reg(fp);
+               src |= REG_NEGV_MASK;
+               emit_arith(fp,
+                          PFS_OP_MAD,
+                          *r, arbneg & 0x7, keep(src), pfs_one, pfs_zero, 0);
+               src = src & ~REG_NEGV_MASK;
+               emit_arith(fp,
+                          PFS_OP_MAD,
+                          *r,
+                          (arbneg ^ 0x7) | WRITEMASK_W,
+                          src, pfs_one, pfs_zero, 0);
+       }
+
+       return 3;
+}
+
+static int swz_emit_partial(struct r300_fragment_program *fp,
+                           GLuint src,
+                           GLuint * r, int mask, int mc, GLuint arbneg)
+{
+       GLuint tmp;
+       GLuint wmask = 0;
+
+       if (!REG_GET_VALID(*r))
+               *r = get_temp_reg(fp);
+
+       /* A partial match, VSWZ/mask define what parts of the
+        * desired swizzle we match
+        */
+       if (mc + s_mask[mask].count == 3) {
+               wmask = WRITEMASK_W;
+               src |= ((arbneg >> 3) & 1) << REG_NEGS_SHIFT;
+       }
+
+       tmp = arbneg & s_mask[mask].mask;
+       if (tmp) {
+               tmp = tmp ^ s_mask[mask].mask;
+               if (tmp) {
+                       emit_arith(fp,
+                                  PFS_OP_MAD,
+                                  *r,
+                                  arbneg & s_mask[mask].mask,
+                                  keep(src) | REG_NEGV_MASK,
+                                  pfs_one, pfs_zero, 0);
+                       if (!wmask) {
+                               REG_SET_NO_USE(src, GL_TRUE);
+                       } else {
+                               REG_SET_NO_USE(src, GL_FALSE);
+                       }
+                       emit_arith(fp,
+                                  PFS_OP_MAD,
+                                  *r, tmp | wmask, src, pfs_one, pfs_zero, 0);
+               } else {
+                       if (!wmask) {
+                               REG_SET_NO_USE(src, GL_TRUE);
+                       } else {
+                               REG_SET_NO_USE(src, GL_FALSE);
+                       }
+                       emit_arith(fp,
+                                  PFS_OP_MAD,
+                                  *r,
+                                  (arbneg & s_mask[mask].mask) | wmask,
+                                  src | REG_NEGV_MASK, pfs_one, pfs_zero, 0);
+               }
+       } else {
+               if (!wmask) {
+                       REG_SET_NO_USE(src, GL_TRUE);
+               } else {
+                       REG_SET_NO_USE(src, GL_FALSE);
+               }
+               emit_arith(fp, PFS_OP_MAD,
+                          *r,
+                          s_mask[mask].mask | wmask,
+                          src, pfs_one, pfs_zero, 0);
+       }
+
+       return s_mask[mask].count;
+}
+
+static GLuint do_swizzle(struct r300_fragment_program *fp,
+                        GLuint src, GLuint arbswz, GLuint arbneg)
+{
+       GLuint r = undef;
+       GLuint vswz;
+       int c_mask = 0;
+       int v_match = 0;
+
+       /* If swizzling from something without an XYZW native swizzle,
+        * emit result to a temp, and do new swizzle from the temp.
+        */
+#if 0
+       if (REG_GET_VSWZ(src) != SWIZZLE_XYZ || REG_GET_SSWZ(src) != SWIZZLE_W) {
+               GLuint temp = get_temp_reg(fp);
+               emit_arith(fp,
+                          PFS_OP_MAD,
+                          temp, WRITEMASK_XYZW, src, pfs_one, pfs_zero, 0);
+               src = temp;
+       }
+#endif
+
+       if (REG_GET_VSWZ(src) != SWIZZLE_XYZ || REG_GET_SSWZ(src) != SWIZZLE_W) {
+               GLuint vsrcswz =
+                   (v_swiz[REG_GET_VSWZ(src)].
+                    hash & (SWZ_X_MASK | SWZ_Y_MASK | SWZ_Z_MASK)) |
+                   REG_GET_SSWZ(src) << 9;
+               GLint i;
+
+               GLuint newswz = 0;
+               GLuint offset;
+               for (i = 0; i < 4; ++i) {
+                       offset = GET_SWZ(arbswz, i);
+
+                       newswz |=
+                           (offset <= 3) ? GET_SWZ(vsrcswz,
+                                                   offset) << i *
+                           3 : offset << i * 3;
+               }
+
+               arbswz = newswz & (SWZ_X_MASK | SWZ_Y_MASK | SWZ_Z_MASK);
+               REG_SET_SSWZ(src, GET_SWZ(newswz, 3));
+       } else {
+               /* set scalar swizzling */
+               REG_SET_SSWZ(src, GET_SWZ(arbswz, 3));
+
+       }
+       do {
+               vswz = REG_GET_VSWZ(src);
+               do {
+                       int chash;
+
+                       REG_SET_VSWZ(src, vswz);
+                       chash = v_swiz[REG_GET_VSWZ(src)].hash &
+                           s_mask[c_mask].hash;
+
+                       if (chash == (arbswz & s_mask[c_mask].hash)) {
+                               if (s_mask[c_mask].count == 3) {
+                                       v_match += swz_native(fp,
+                                                             src, &r, arbneg);
+                               } else {
+                                       v_match += swz_emit_partial(fp,
+                                                                   src,
+                                                                   &r,
+                                                                   c_mask,
+                                                                   v_match,
+                                                                   arbneg);
+                               }
+
+                               if (v_match == 3)
+                                       return r;
+
+                               /* Fill with something invalid.. all 0's was
+                                * wrong before, matched SWIZZLE_X.  So all
+                                * 1's will be okay for now
+                                */
+                               arbswz |= (PFS_INVAL & s_mask[c_mask].hash);
+                       }
+               } while (v_swiz[++vswz].hash != PFS_INVAL);
+               REG_SET_VSWZ(src, SWIZZLE_XYZ);
+       } while (s_mask[++c_mask].hash != PFS_INVAL);
+
+       ERROR("should NEVER get here\n");
+       return r;
+}
+
+static GLuint t_src(struct r300_fragment_program *fp,
+                   struct prog_src_register fpsrc)
+{
+       GLuint r = undef;
+
+       switch (fpsrc.File) {
+       case PROGRAM_TEMPORARY:
+               REG_SET_INDEX(r, fpsrc.Index);
+               REG_SET_VALID(r, GL_TRUE);
+               REG_SET_TYPE(r, REG_TYPE_TEMP);
+               break;
+       case PROGRAM_INPUT:
+               REG_SET_INDEX(r, fpsrc.Index);
+               REG_SET_VALID(r, GL_TRUE);
+               REG_SET_TYPE(r, REG_TYPE_INPUT);
+               break;
+       case PROGRAM_LOCAL_PARAM:
+               r = emit_const4fv(fp,
+                                 fp->mesa_program.Base.LocalParams[fpsrc.
+                                                                   Index]);
+               break;
+       case PROGRAM_ENV_PARAM:
+               r = emit_const4fv(fp,
+                                 fp->ctx->FragmentProgram.Parameters[fpsrc.
+                                                                     Index]);
+               break;
+       case PROGRAM_STATE_VAR:
+       case PROGRAM_NAMED_PARAM:
+               r = emit_const4fv(fp,
+                                 fp->mesa_program.Base.Parameters->
+                                 ParameterValues[fpsrc.Index]);
+               break;
+       default:
+               ERROR("unknown SrcReg->File %x\n", fpsrc.File);
+               return r;
+       }
+
+       /* no point swizzling ONE/ZERO/HALF constants... */
+       if (REG_GET_VSWZ(r) < SWIZZLE_111 || REG_GET_SSWZ(r) < SWIZZLE_ZERO)
+               r = do_swizzle(fp, r, fpsrc.Swizzle, fpsrc.NegateBase);
+       return r;
+}
+
+static GLuint t_scalar_src(struct r300_fragment_program *fp,
+                          struct prog_src_register fpsrc)
+{
+       struct prog_src_register src = fpsrc;
+       int sc = GET_SWZ(fpsrc.Swizzle, 0);     /* X */
+
+       src.Swizzle = ((sc << 0) | (sc << 3) | (sc << 6) | (sc << 9));
+
+       return t_src(fp, src);
+}
+
+static GLuint t_dst(struct r300_fragment_program *fp,
+                   struct prog_dst_register dest)
+{
+       GLuint r = undef;
+
+       switch (dest.File) {
+       case PROGRAM_TEMPORARY:
+               REG_SET_INDEX(r, dest.Index);
+               REG_SET_VALID(r, GL_TRUE);
+               REG_SET_TYPE(r, REG_TYPE_TEMP);
+               return r;
+       case PROGRAM_OUTPUT:
+               REG_SET_TYPE(r, REG_TYPE_OUTPUT);
+               switch (dest.Index) {
+               case FRAG_RESULT_COLR:
+               case FRAG_RESULT_DEPR:
+                       REG_SET_INDEX(r, dest.Index);
+                       REG_SET_VALID(r, GL_TRUE);
+                       return r;
+               default:
+                       ERROR("Bad DstReg->Index 0x%x\n", dest.Index);
+                       return r;
+               }
+       default:
+               ERROR("Bad DstReg->File 0x%x\n", dest.File);
+               return r;
+       }
+}
+
+static int t_hw_src(struct r300_fragment_program *fp, GLuint src, GLboolean tex)
+{
+       COMPILE_STATE;
+       int idx;
+       int index = REG_GET_INDEX(src);
+
+       switch (REG_GET_TYPE(src)) {
+       case REG_TYPE_TEMP:
+               /* NOTE: if reg==-1 here, a source is being read that
+                *       hasn't been written to. Undefined results.
+                */
+               if (cs->temps[index].reg == -1)
+                       cs->temps[index].reg = get_hw_temp(fp, cs->nrslots);
+
+               idx = cs->temps[index].reg;
+
+               if (!REG_GET_NO_USE(src) && (--cs->temps[index].refcount == 0))
+                       free_temp(fp, src);
+               break;
+       case REG_TYPE_INPUT:
+               idx = cs->inputs[index].reg;
+
+               if (!REG_GET_NO_USE(src) && (--cs->inputs[index].refcount == 0))
+                       free_hw_temp(fp, cs->inputs[index].reg);
+               break;
+       case REG_TYPE_CONST:
+               return (index | SRC_CONST);
+       default:
+               ERROR("Invalid type for source reg\n");
+               return (0 | SRC_CONST);
+       }
+
+       if (!tex)
+               cs->used_in_node |= (1 << idx);
+
+       return idx;
+}
+
+static int t_hw_dst(struct r300_fragment_program *fp,
+                   GLuint dest, GLboolean tex, int slot)
+{
+       COMPILE_STATE;
+       int idx;
+       GLuint index = REG_GET_INDEX(dest);
+       assert(REG_GET_VALID(dest));
+
+       switch (REG_GET_TYPE(dest)) {
+       case REG_TYPE_TEMP:
+               if (cs->temps[REG_GET_INDEX(dest)].reg == -1) {
+                       if (!tex) {
+                               cs->temps[index].reg = get_hw_temp(fp, slot);
+                       } else {
+                               cs->temps[index].reg = get_hw_temp_tex(fp);
+                       }
+               }
+               idx = cs->temps[index].reg;
+
+               if (!REG_GET_NO_USE(dest) && (--cs->temps[index].refcount == 0))
+                       free_temp(fp, dest);
+
+               cs->dest_in_node |= (1 << idx);
+               cs->used_in_node |= (1 << idx);
+               break;
+       case REG_TYPE_OUTPUT:
+               switch (index) {
+               case FRAG_RESULT_COLR:
+                       fp->node[fp->cur_node].flags |=
+                           R300_PFS_NODE_OUTPUT_COLOR;
+                       break;
+               case FRAG_RESULT_DEPR:
+                       fp->node[fp->cur_node].flags |=
+                           R300_PFS_NODE_OUTPUT_DEPTH;
+                       break;
+               }
+               return index;
+               break;
+       default:
+               ERROR("invalid dest reg type %d\n", REG_GET_TYPE(dest));
+               return 0;
+       }
+
+       return idx;
+}
+
+static void emit_nop(struct r300_fragment_program *fp)
+{
+       COMPILE_STATE;
+
+       if (cs->nrslots >= PFS_MAX_ALU_INST) {
+               ERROR("Out of ALU instruction slots\n");
+               return;
+       }
+
+       fp->alu.inst[cs->nrslots].inst0 = NOP_INST0;
+       fp->alu.inst[cs->nrslots].inst1 = NOP_INST1;
+       fp->alu.inst[cs->nrslots].inst2 = NOP_INST2;
+       fp->alu.inst[cs->nrslots].inst3 = NOP_INST3;
+       cs->nrslots++;
+}
+
+static void emit_tex(struct r300_fragment_program *fp,
+                    struct prog_instruction *fpi, int opcode)
+{
+       COMPILE_STATE;
+       GLuint coord = t_src(fp, fpi->SrcReg[0]);
+       GLuint dest = undef, rdest = undef;
+       GLuint din, uin;
+       int unit = fpi->TexSrcUnit;
+       int hwsrc, hwdest;
+       GLuint tempreg = 0;
+
+       uin = cs->used_in_node;
+       din = cs->dest_in_node;
+
+       /* Resolve source/dest to hardware registers */
+       if (opcode != R300_FPITX_OP_KIL) {
+               if (fpi->TexSrcTarget == TEXTURE_RECT_INDEX) {
+                       /**
+                        * Hardware uses [0..1]x[0..1] range for rectangle textures
+                        * instead of [0..Width]x[0..Height].
+                        * Add a scaling instruction.
+                        *
+                        * \todo Refactor this once we have proper rewriting/optimization
+                        * support for programs.
+                        */
+                       gl_state_index tokens[STATE_LENGTH] = {
+                               STATE_INTERNAL, STATE_R300_TEXRECT_FACTOR, 0, 0,
+                               0
+                       };
+                       int factor_index;
+                       GLuint factorreg;
+
+                       tokens[2] = unit;
+                       factor_index =
+                           _mesa_add_state_reference(fp->mesa_program.Base.
+                                                     Parameters, tokens);
+                       factorreg =
+                           emit_const4fv(fp,
+                                         fp->mesa_program.Base.Parameters->
+                                         ParameterValues[factor_index]);
+                       tempreg = keep(get_temp_reg(fp));
+
+                       emit_arith(fp, PFS_OP_MAD, tempreg, WRITEMASK_XYZW,
+                                  coord, factorreg, pfs_zero, 0);
+
+                       /* Ensure correct node indirection */
+                       uin = cs->used_in_node;
+                       din = cs->dest_in_node;
+
+                       hwsrc = t_hw_src(fp, tempreg, GL_TRUE);
+               } else {
+                       hwsrc = t_hw_src(fp, coord, GL_TRUE);
+               }
+
+               dest = t_dst(fp, fpi->DstReg);
+
+               /* r300 doesn't seem to be able to do TEX->output reg */
+               if (REG_GET_TYPE(dest) == REG_TYPE_OUTPUT) {
+                       rdest = dest;
+                       dest = get_temp_reg_tex(fp);
+               } else if (fpi->DstReg.WriteMask != WRITEMASK_XYZW) {
+                       /* in case write mask isn't XYZW */
+                       rdest = dest;
+                       dest = get_temp_reg_tex(fp);
+               }
+               hwdest =
+                   t_hw_dst(fp, dest, GL_TRUE,
+                            fp->node[fp->cur_node].alu_offset);
+
+               /* Use a temp that hasn't been used in this node, rather
+                * than causing an indirection
+                */
+               if (uin & (1 << hwdest)) {
+                       free_hw_temp(fp, hwdest);
+                       hwdest = get_hw_temp_tex(fp);
+                       cs->temps[REG_GET_INDEX(dest)].reg = hwdest;
+               }
+       } else {
+               hwdest = 0;
+               unit = 0;
+               hwsrc = t_hw_src(fp, coord, GL_TRUE);
+       }
+
+       /* Indirection if source has been written in this node, or if the
+        * dest has been read/written in this node
+        */
+       if ((REG_GET_TYPE(coord) != REG_TYPE_CONST &&
+            (din & (1 << hwsrc))) || (uin & (1 << hwdest))) {
+
+               /* Finish off current node */
+               if (fp->node[fp->cur_node].alu_offset == cs->nrslots)
+                       emit_nop(fp);
+
+               fp->node[fp->cur_node].alu_end =
+                   cs->nrslots - fp->node[fp->cur_node].alu_offset - 1;
+               assert(fp->node[fp->cur_node].alu_end >= 0);
+
+               if (++fp->cur_node >= PFS_MAX_TEX_INDIRECT) {
+                       ERROR("too many levels of texture indirection\n");
+                       return;
+               }
+
+               /* Start new node */
+               fp->node[fp->cur_node].tex_offset = fp->tex.length;
+               fp->node[fp->cur_node].alu_offset = cs->nrslots;
+               fp->node[fp->cur_node].tex_end = -1;
+               fp->node[fp->cur_node].alu_end = -1;
+               fp->node[fp->cur_node].flags = 0;
+               cs->used_in_node = 0;
+               cs->dest_in_node = 0;
+       }
+
+       if (fp->cur_node == 0)
+               fp->first_node_has_tex = 1;
+
+       fp->tex.inst[fp->tex.length++] = 0 | (hwsrc << R300_FPITX_SRC_SHIFT)
+           | (hwdest << R300_FPITX_DST_SHIFT)
+           | (unit << R300_FPITX_IMAGE_SHIFT)
+           /* not entirely sure about this */
+           | (opcode << R300_FPITX_OPCODE_SHIFT);
+
+       cs->dest_in_node |= (1 << hwdest);
+       if (REG_GET_TYPE(coord) != REG_TYPE_CONST)
+               cs->used_in_node |= (1 << hwsrc);
+
+       fp->node[fp->cur_node].tex_end++;
+
+       /* Copy from temp to output if needed */
+       if (REG_GET_VALID(rdest)) {
+               emit_arith(fp, PFS_OP_MAD, rdest, fpi->DstReg.WriteMask, dest,
+                          pfs_one, pfs_zero, 0);
+               free_temp(fp, dest);
+       }
+
+       /* Free temp register */
+       if (tempreg != 0)
+               free_temp(fp, tempreg);
+}
+
+/**
+ * Returns the first slot where we could possibly allow writing to dest,
+ * according to register allocation.
+ */
+static int get_earliest_allowed_write(struct r300_fragment_program *fp,
+                                     GLuint dest, int mask)
+{
+       COMPILE_STATE;
+       int idx;
+       int pos;
+       GLuint index = REG_GET_INDEX(dest);
+       assert(REG_GET_VALID(dest));
+
+       switch (REG_GET_TYPE(dest)) {
+       case REG_TYPE_TEMP:
+               if (cs->temps[index].reg == -1)
+                       return 0;
+
+               idx = cs->temps[index].reg;
+               break;
+       case REG_TYPE_OUTPUT:
+               return 0;
+       default:
+               ERROR("invalid dest reg type %d\n", REG_GET_TYPE(dest));
+               return 0;
+       }
+
+       pos = cs->hwtemps[idx].reserved;
+       if (mask & WRITEMASK_XYZ) {
+               if (pos < cs->hwtemps[idx].vector_lastread)
+                       pos = cs->hwtemps[idx].vector_lastread;
+       }
+       if (mask & WRITEMASK_W) {
+               if (pos < cs->hwtemps[idx].scalar_lastread)
+                       pos = cs->hwtemps[idx].scalar_lastread;
+       }
+
+       return pos;
+}
+
+/**
+ * Allocates a slot for an ALU instruction that can consist of
+ * a vertex part or a scalar part or both.
+ *
+ * Sources from src (src[0] to src[argc-1]) are added to the slot in the
+ * appropriate position (vector and/or scalar), and their positions are
+ * recorded in the srcpos array.
+ *
+ * This function emits instruction code for the source fetch and the
+ * argument selection. It does not emit instruction code for the
+ * opcode or the destination selection.
+ *
+ * @return the index of the slot
+ */
+static int find_and_prepare_slot(struct r300_fragment_program *fp,
+                                GLboolean emit_vop,
+                                GLboolean emit_sop,
+                                int argc, GLuint * src, GLuint dest, int mask)
+{
+       COMPILE_STATE;
+       int hwsrc[3];
+       int srcpos[3];
+       unsigned int used;
+       int tempused;
+       int tempvsrc[3];
+       int tempssrc[3];
+       int pos;
+       int regnr;
+       int i, j;
+
+       // Determine instruction slots, whether sources are required on
+       // vector or scalar side, and the smallest slot number where
+       // all source registers are available
+       used = 0;
+       if (emit_vop)
+               used |= SLOT_OP_VECTOR;
+       if (emit_sop)
+               used |= SLOT_OP_SCALAR;
+
+       pos = get_earliest_allowed_write(fp, dest, mask);
+
+       if (fp->node[fp->cur_node].alu_offset > pos)
+               pos = fp->node[fp->cur_node].alu_offset;
+       for (i = 0; i < argc; ++i) {
+               if (!REG_GET_BUILTIN(src[i])) {
+                       if (emit_vop)
+                               used |= v_swiz[REG_GET_VSWZ(src[i])].flags << i;
+                       if (emit_sop)
+                               used |= s_swiz[REG_GET_SSWZ(src[i])].flags << i;
+               }
+
+               hwsrc[i] = t_hw_src(fp, src[i], GL_FALSE);      /* Note: sideeffects wrt refcounting! */
+               regnr = hwsrc[i] & 31;
+
+               if (REG_GET_TYPE(src[i]) == REG_TYPE_TEMP) {
+                       if (used & (SLOT_SRC_VECTOR << i)) {
+                               if (cs->hwtemps[regnr].vector_valid > pos)
+                                       pos = cs->hwtemps[regnr].vector_valid;
+                       }
+                       if (used & (SLOT_SRC_SCALAR << i)) {
+                               if (cs->hwtemps[regnr].scalar_valid > pos)
+                                       pos = cs->hwtemps[regnr].scalar_valid;
+                       }
+               }
+       }
+
+       // Find a slot that fits
+       for (;; ++pos) {
+               if (cs->slot[pos].used & used & SLOT_OP_BOTH)
+                       continue;
+
+               if (pos >= cs->nrslots) {
+                       if (cs->nrslots >= PFS_MAX_ALU_INST) {
+                               ERROR("Out of ALU instruction slots\n");
+                               return -1;
+                       }
+
+                       fp->alu.inst[pos].inst0 = NOP_INST0;
+                       fp->alu.inst[pos].inst1 = NOP_INST1;
+                       fp->alu.inst[pos].inst2 = NOP_INST2;
+                       fp->alu.inst[pos].inst3 = NOP_INST3;
+
+                       cs->nrslots++;
+               }
+               // Note: When we need both parts (vector and scalar) of a source,
+               // we always try to put them into the same position. This makes the
+               // code easier to read, and it is optimal (i.e. one doesn't gain
+               // anything by splitting the parts).
+               // It also avoids headaches with swizzles that access both parts (i.e WXY)
+               tempused = cs->slot[pos].used;
+               for (i = 0; i < 3; ++i) {
+                       tempvsrc[i] = cs->slot[pos].vsrc[i];
+                       tempssrc[i] = cs->slot[pos].ssrc[i];
+               }
+
+               for (i = 0; i < argc; ++i) {
+                       int flags = (used >> i) & SLOT_SRC_BOTH;
+
+                       if (!flags) {
+                               srcpos[i] = 0;
+                               continue;
+                       }
+
+                       for (j = 0; j < 3; ++j) {
+                               if ((tempused >> j) & flags & SLOT_SRC_VECTOR) {
+                                       if (tempvsrc[j] != hwsrc[i])
+                                               continue;
+                               }
+
+                               if ((tempused >> j) & flags & SLOT_SRC_SCALAR) {
+                                       if (tempssrc[j] != hwsrc[i])
+                                               continue;
+                               }
+
+                               break;
+                       }
+
+                       if (j == 3)
+                               break;
+
+                       srcpos[i] = j;
+                       tempused |= flags << j;
+                       if (flags & SLOT_SRC_VECTOR)
+                               tempvsrc[j] = hwsrc[i];
+                       if (flags & SLOT_SRC_SCALAR)
+                               tempssrc[j] = hwsrc[i];
+               }
+
+               if (i == argc)
+                       break;
+       }
+
+       // Found a slot, reserve it
+       cs->slot[pos].used = tempused | (used & SLOT_OP_BOTH);
+       for (i = 0; i < 3; ++i) {
+               cs->slot[pos].vsrc[i] = tempvsrc[i];
+               cs->slot[pos].ssrc[i] = tempssrc[i];
+       }
+
+       for (i = 0; i < argc; ++i) {
+               if (REG_GET_TYPE(src[i]) == REG_TYPE_TEMP) {
+                       int regnr = hwsrc[i] & 31;
+
+                       if (used & (SLOT_SRC_VECTOR << i)) {
+                               if (cs->hwtemps[regnr].vector_lastread < pos)
+                                       cs->hwtemps[regnr].vector_lastread =
+                                           pos;
+                       }
+                       if (used & (SLOT_SRC_SCALAR << i)) {
+                               if (cs->hwtemps[regnr].scalar_lastread < pos)
+                                       cs->hwtemps[regnr].scalar_lastread =
+                                           pos;
+                       }
+               }
+       }
+
+       // Emit the source fetch code
+       fp->alu.inst[pos].inst1 &= ~R300_FPI1_SRC_MASK;
+       fp->alu.inst[pos].inst1 |=
+           ((cs->slot[pos].vsrc[0] << R300_FPI1_SRC0C_SHIFT) |
+            (cs->slot[pos].vsrc[1] << R300_FPI1_SRC1C_SHIFT) |
+            (cs->slot[pos].vsrc[2] << R300_FPI1_SRC2C_SHIFT));
+
+       fp->alu.inst[pos].inst3 &= ~R300_FPI3_SRC_MASK;
+       fp->alu.inst[pos].inst3 |=
+           ((cs->slot[pos].ssrc[0] << R300_FPI3_SRC0A_SHIFT) |
+            (cs->slot[pos].ssrc[1] << R300_FPI3_SRC1A_SHIFT) |
+            (cs->slot[pos].ssrc[2] << R300_FPI3_SRC2A_SHIFT));
+
+       // Emit the argument selection code
+       if (emit_vop) {
+               int swz[3];
+
+               for (i = 0; i < 3; ++i) {
+                       if (i < argc) {
+                               swz[i] = (v_swiz[REG_GET_VSWZ(src[i])].base +
+                                         (srcpos[i] *
+                                          v_swiz[REG_GET_VSWZ(src[i])].
+                                          stride)) | ((src[i] & REG_NEGV_MASK)
+                                                      ? ARG_NEG : 0) | ((src[i]
+                                                                         &
+                                                                         REG_ABS_MASK)
+                                                                        ?
+                                                                        ARG_ABS
+                                                                        : 0);
+                       } else {
+                               swz[i] = R300_FPI0_ARGC_ZERO;
+                       }
+               }
+
+               fp->alu.inst[pos].inst0 &=
+                   ~(R300_FPI0_ARG0C_MASK | R300_FPI0_ARG1C_MASK |
+                     R300_FPI0_ARG2C_MASK);
+               fp->alu.inst[pos].inst0 |=
+                   (swz[0] << R300_FPI0_ARG0C_SHIFT) | (swz[1] <<
+                                                        R300_FPI0_ARG1C_SHIFT)
+                   | (swz[2] << R300_FPI0_ARG2C_SHIFT);
+       }
+
+       if (emit_sop) {
+               int swz[3];
+
+               for (i = 0; i < 3; ++i) {
+                       if (i < argc) {
+                               swz[i] = (s_swiz[REG_GET_SSWZ(src[i])].base +
+                                         (srcpos[i] *
+                                          s_swiz[REG_GET_SSWZ(src[i])].
+                                          stride)) | ((src[i] & REG_NEGV_MASK)
+                                                      ? ARG_NEG : 0) | ((src[i]
+                                                                         &
+                                                                         REG_ABS_MASK)
+                                                                        ?
+                                                                        ARG_ABS
+                                                                        : 0);
+                       } else {
+                               swz[i] = R300_FPI2_ARGA_ZERO;
+                       }
+               }
+
+               fp->alu.inst[pos].inst2 &=
+                   ~(R300_FPI2_ARG0A_MASK | R300_FPI2_ARG1A_MASK |
+                     R300_FPI2_ARG2A_MASK);
+               fp->alu.inst[pos].inst2 |=
+                   (swz[0] << R300_FPI2_ARG0A_SHIFT) | (swz[1] <<
+                                                        R300_FPI2_ARG1A_SHIFT)
+                   | (swz[2] << R300_FPI2_ARG2A_SHIFT);
+       }
+
+       return pos;
+}
+
+/**
+ * Append an ALU instruction to the instruction list.
+ */
+static void emit_arith(struct r300_fragment_program *fp,
+                      int op,
+                      GLuint dest,
+                      int mask,
+                      GLuint src0, GLuint src1, GLuint src2, int flags)
+{
+       COMPILE_STATE;
+       GLuint src[3] = { src0, src1, src2 };
+       int hwdest;
+       GLboolean emit_vop, emit_sop;
+       int vop, sop, argc;
+       int pos;
+
+       vop = r300_fpop[op].v_op;
+       sop = r300_fpop[op].s_op;
+       argc = r300_fpop[op].argc;
+
+       if (REG_GET_TYPE(dest) == REG_TYPE_OUTPUT &&
+           REG_GET_INDEX(dest) == FRAG_RESULT_DEPR) {
+               if (mask & WRITEMASK_Z) {
+                       mask = WRITEMASK_W;
+               } else {
+                       return;
+               }
+       }
+
+       emit_vop = GL_FALSE;
+       emit_sop = GL_FALSE;
+       if ((mask & WRITEMASK_XYZ) || vop == R300_FPI0_OUTC_DP3)
+               emit_vop = GL_TRUE;
+       if ((mask & WRITEMASK_W) || vop == R300_FPI0_OUTC_REPL_ALPHA)
+               emit_sop = GL_TRUE;
+
+       pos =
+           find_and_prepare_slot(fp, emit_vop, emit_sop, argc, src, dest,
+                                 mask);
+       if (pos < 0)
+               return;
+
+       hwdest = t_hw_dst(fp, dest, GL_FALSE, pos);     /* Note: Side effects wrt register allocation */
+
+       if (flags & PFS_FLAG_SAT) {
+               vop |= R300_FPI0_OUTC_SAT;
+               sop |= R300_FPI2_OUTA_SAT;
+       }
+
+       /* Throw the pieces together and get FPI0/1 */
+       if (emit_vop) {
+               fp->alu.inst[pos].inst0 |= vop;
+
+               fp->alu.inst[pos].inst1 |= hwdest << R300_FPI1_DSTC_SHIFT;
+
+               if (REG_GET_TYPE(dest) == REG_TYPE_OUTPUT) {
+                       if (REG_GET_INDEX(dest) == FRAG_RESULT_COLR) {
+                               fp->alu.inst[pos].inst1 |=
+                                   (mask & WRITEMASK_XYZ) <<
+                                   R300_FPI1_DSTC_OUTPUT_MASK_SHIFT;
+                       } else
+                               assert(0);
+               } else {
+                       fp->alu.inst[pos].inst1 |=
+                           (mask & WRITEMASK_XYZ) <<
+                           R300_FPI1_DSTC_REG_MASK_SHIFT;
+
+                       cs->hwtemps[hwdest].vector_valid = pos + 1;
+               }
+       }
+
+       /* And now FPI2/3 */
+       if (emit_sop) {
+               fp->alu.inst[pos].inst2 |= sop;
+
+               if (mask & WRITEMASK_W) {
+                       if (REG_GET_TYPE(dest) == REG_TYPE_OUTPUT) {
+                               if (REG_GET_INDEX(dest) == FRAG_RESULT_COLR) {
+                                       fp->alu.inst[pos].inst3 |=
+                                           (hwdest << R300_FPI3_DSTA_SHIFT) |
+                                           R300_FPI3_DSTA_OUTPUT;
+                               } else if (REG_GET_INDEX(dest) ==
+                                          FRAG_RESULT_DEPR) {
+                                       fp->alu.inst[pos].inst3 |=
+                                           R300_FPI3_DSTA_DEPTH;
+                               } else
+                                       assert(0);
+                       } else {
+                               fp->alu.inst[pos].inst3 |=
+                                   (hwdest << R300_FPI3_DSTA_SHIFT) |
+                                   R300_FPI3_DSTA_REG;
+
+                               cs->hwtemps[hwdest].scalar_valid = pos + 1;
+                       }
+               }
+       }
+
+       return;
+}
+
+#if 0
+static GLuint get_attrib(struct r300_fragment_program *fp, GLuint attr)
+{
+       struct gl_fragment_program *mp = &fp->mesa_program;
+       GLuint r = undef;
+
+       if (!(mp->Base.InputsRead & (1 << attr))) {
+               ERROR("Attribute %d was not provided!\n", attr);
+               return undef;
+       }
+
+       REG_SET_TYPE(r, REG_TYPE_INPUT);
+       REG_SET_INDEX(r, attr);
+       REG_SET_VALID(r, GL_TRUE);
+       return r;
+}
+#endif
+
+static GLfloat SinCosConsts[2][4] = {
+       {
+        1.273239545,           // 4/PI
+        -0.405284735,          // -4/(PI*PI)
+        3.141592654,           // PI
+        0.2225                 // weight
+        },
+       {
+        0.75,
+        0.0,
+        0.159154943,           // 1/(2*PI)
+        6.283185307            // 2*PI
+        }
+};
+
+/**
+ * Emit a LIT instruction.
+ * \p flags may be PFS_FLAG_SAT
+ *
+ * Definition of LIT (from ARB_fragment_program):
+ * tmp = VectorLoad(op0);
+ * if (tmp.x < 0) tmp.x = 0;
+ * if (tmp.y < 0) tmp.y = 0;
+ * if (tmp.w < -(128.0-epsilon)) tmp.w = -(128.0-epsilon);
+ * else if (tmp.w > 128-epsilon) tmp.w = 128-epsilon;
+ * result.x = 1.0;
+ * result.y = tmp.x;
+ * result.z = (tmp.x > 0) ? RoughApproxPower(tmp.y, tmp.w) : 0.0;
+ * result.w = 1.0;
+ *
+ * The longest path of computation is the one leading to result.z,
+ * consisting of 5 operations. This implementation of LIT takes
+ * 5 slots. So unless there's some special undocumented opcode,
+ * this implementation is potentially optimal. Unfortunately,
+ * emit_arith is a bit too conservative because it doesn't understand
+ * partial writes to the vector component.
+ */
+static const GLfloat LitConst[4] =
+    { 127.999999, 127.999999, 127.999999, -127.999999 };
+
+static void emit_lit(struct r300_fragment_program *fp,
+                    GLuint dest, int mask, GLuint src, int flags)
+{
+       COMPILE_STATE;
+       GLuint cnst;
+       int needTemporary;
+       GLuint temp;
+
+       cnst = emit_const4fv(fp, LitConst);
+
+       needTemporary = 0;
+       if ((mask & WRITEMASK_XYZW) != WRITEMASK_XYZW) {
+               needTemporary = 1;
+       } else if (REG_GET_TYPE(dest) == REG_TYPE_OUTPUT) {
+               // LIT is typically followed by DP3/DP4, so there's no point
+               // in creating special code for this case
+               needTemporary = 1;
+       }
+
+       if (needTemporary) {
+               temp = keep(get_temp_reg(fp));
+       } else {
+               temp = keep(dest);
+       }
+
+       // Note: The order of emit_arith inside the slots is relevant,
+       // because emit_arith only looks at scalar vs. vector when resolving
+       // dependencies, and it does not consider individual vector components,
+       // so swizzling between the two parts can create fake dependencies.
+
+       // First slot
+       emit_arith(fp, PFS_OP_MAX, temp, WRITEMASK_XY,
+                  keep(src), pfs_zero, undef, 0);
+       emit_arith(fp, PFS_OP_MAX, temp, WRITEMASK_W, src, cnst, undef, 0);
+
+       // Second slot
+       emit_arith(fp, PFS_OP_MIN, temp, WRITEMASK_Z,
+                  swizzle(temp, W, W, W, W), cnst, undef, 0);
+       emit_arith(fp, PFS_OP_LG2, temp, WRITEMASK_W,
+                  swizzle(temp, Y, Y, Y, Y), undef, undef, 0);
+
+       // Third slot
+       // If desired, we saturate the y result here.
+       // This does not affect the use as a condition variable in the CMP later
+       emit_arith(fp, PFS_OP_MAD, temp, WRITEMASK_W,
+                  temp, swizzle(temp, Z, Z, Z, Z), pfs_zero, 0);
+       emit_arith(fp, PFS_OP_MAD, temp, WRITEMASK_Y,
+                  swizzle(temp, X, X, X, X), pfs_one, pfs_zero, flags);
+
+       // Fourth slot
+       emit_arith(fp, PFS_OP_MAD, temp, WRITEMASK_X,
+                  pfs_one, pfs_one, pfs_zero, 0);
+       emit_arith(fp, PFS_OP_EX2, temp, WRITEMASK_W, temp, undef, undef, 0);
+
+       // Fifth slot
+       emit_arith(fp, PFS_OP_CMP, temp, WRITEMASK_Z,
+                  pfs_zero, swizzle(temp, W, W, W, W),
+                  negate(swizzle(temp, Y, Y, Y, Y)), flags);
+       emit_arith(fp, PFS_OP_MAD, temp, WRITEMASK_W, pfs_one, pfs_one,
+                  pfs_zero, 0);
+
+       if (needTemporary) {
+               emit_arith(fp, PFS_OP_MAD, dest, mask,
+                          temp, pfs_one, pfs_zero, flags);
+               free_temp(fp, temp);
+       } else {
+               // Decrease refcount of the destination
+               t_hw_dst(fp, dest, GL_FALSE, cs->nrslots);
+       }
+}
+
+static GLboolean parse_program(struct r300_fragment_program *fp)
+{
+       struct gl_fragment_program *mp = &fp->mesa_program;
+       const struct prog_instruction *inst = mp->Base.Instructions;
+       struct prog_instruction *fpi;
+       GLuint src[3], dest, temp[2];
+       int flags, mask = 0;
+       int const_sin[2];
+
+       if (!inst || inst[0].Opcode == OPCODE_END) {
+               ERROR("empty program?\n");
+               return GL_FALSE;
+       }
+
+       for (fpi = mp->Base.Instructions; fpi->Opcode != OPCODE_END; fpi++) {
+               if (fpi->SaturateMode == SATURATE_ZERO_ONE)
+                       flags = PFS_FLAG_SAT;
+               else
+                       flags = 0;
+
+               if (fpi->Opcode != OPCODE_KIL) {
+                       dest = t_dst(fp, fpi->DstReg);
+                       mask = fpi->DstReg.WriteMask;
+               }
+
+               switch (fpi->Opcode) {
+               case OPCODE_ABS:
+                       src[0] = t_src(fp, fpi->SrcReg[0]);
+                       emit_arith(fp, PFS_OP_MAD, dest, mask,
+                                  absolute(src[0]), pfs_one, pfs_zero, flags);
+                       break;
+               case OPCODE_ADD:
+                       src[0] = t_src(fp, fpi->SrcReg[0]);
+                       src[1] = t_src(fp, fpi->SrcReg[1]);
+                       emit_arith(fp, PFS_OP_MAD, dest, mask,
+                                  src[0], pfs_one, src[1], flags);
+                       break;
+               case OPCODE_CMP:
+                       src[0] = t_src(fp, fpi->SrcReg[0]);
+                       src[1] = t_src(fp, fpi->SrcReg[1]);
+                       src[2] = t_src(fp, fpi->SrcReg[2]);
+                       /* ARB_f_p - if src0.c < 0.0 ? src1.c : src2.c
+                        *    r300 - if src2.c < 0.0 ? src1.c : src0.c
+                        */
+                       emit_arith(fp, PFS_OP_CMP, dest, mask,
+                                  src[2], src[1], src[0], flags);
+                       break;
+               case OPCODE_COS:
+                       /*
+                        * cos using a parabola (see SIN):
+                        * cos(x):
+                        *   x = (x/(2*PI))+0.75
+                        *   x = frac(x)
+                        *   x = (x*2*PI)-PI
+                        *   result = sin(x)
+                        */
+                       temp[0] = get_temp_reg(fp);
+                       const_sin[0] = emit_const4fv(fp, SinCosConsts[0]);
+                       const_sin[1] = emit_const4fv(fp, SinCosConsts[1]);
+                       src[0] = t_scalar_src(fp, fpi->SrcReg[0]);
+
+                       /* add 0.5*PI and do range reduction */
+
+                       emit_arith(fp, PFS_OP_MAD, temp[0], WRITEMASK_X,
+                                  swizzle(src[0], X, X, X, X),
+                                  swizzle(const_sin[1], Z, Z, Z, Z),
+                                  swizzle(const_sin[1], X, X, X, X), 0);
+
+                       emit_arith(fp, PFS_OP_FRC, temp[0], WRITEMASK_X,
+                                  swizzle(temp[0], X, X, X, X),
+                                  undef, undef, 0);
+
+                       emit_arith(fp, PFS_OP_MAD, temp[0], WRITEMASK_Z, swizzle(temp[0], X, X, X, X), swizzle(const_sin[1], W, W, W, W),       //2*PI
+                                  negate(swizzle(const_sin[0], Z, Z, Z, Z)),   //-PI
+                                  0);
+
+                       /* SIN */
+
+                       emit_arith(fp, PFS_OP_MAD, temp[0],
+                                  WRITEMASK_X | WRITEMASK_Y, swizzle(temp[0],
+                                                                     Z, Z, Z,
+                                                                     Z),
+                                  const_sin[0], pfs_zero, 0);
+
+                       emit_arith(fp, PFS_OP_MAD, temp[0], WRITEMASK_X,
+                                  swizzle(temp[0], Y, Y, Y, Y),
+                                  absolute(swizzle(temp[0], Z, Z, Z, Z)),
+                                  swizzle(temp[0], X, X, X, X), 0);
+
+                       emit_arith(fp, PFS_OP_MAD, temp[0], WRITEMASK_Y,
+                                  swizzle(temp[0], X, X, X, X),
+                                  absolute(swizzle(temp[0], X, X, X, X)),
+                                  negate(swizzle(temp[0], X, X, X, X)), 0);
+
+                       emit_arith(fp, PFS_OP_MAD, dest, mask,
+                                  swizzle(temp[0], Y, Y, Y, Y),
+                                  swizzle(const_sin[0], W, W, W, W),
+                                  swizzle(temp[0], X, X, X, X), flags);
+
+                       free_temp(fp, temp[0]);
+                       break;
+               case OPCODE_DP3:
+                       src[0] = t_src(fp, fpi->SrcReg[0]);
+                       src[1] = t_src(fp, fpi->SrcReg[1]);
+                       emit_arith(fp, PFS_OP_DP3, dest, mask,
+                                  src[0], src[1], undef, flags);
+                       break;
+               case OPCODE_DP4:
+                       src[0] = t_src(fp, fpi->SrcReg[0]);
+                       src[1] = t_src(fp, fpi->SrcReg[1]);
+                       emit_arith(fp, PFS_OP_DP4, dest, mask,
+                                  src[0], src[1], undef, flags);
+                       break;
+               case OPCODE_DPH:
+                       src[0] = t_src(fp, fpi->SrcReg[0]);
+                       src[1] = t_src(fp, fpi->SrcReg[1]);
+                       /* src0.xyz1 -> temp
+                        * DP4 dest, temp, src1
+                        */
+#if 0
+                       temp[0] = get_temp_reg(fp);
+                       src[0].s_swz = SWIZZLE_ONE;
+                       emit_arith(fp, PFS_OP_MAD, temp[0], mask,
+                                  src[0], pfs_one, pfs_zero, 0);
+                       emit_arith(fp, PFS_OP_DP4, dest, mask,
+                                  temp[0], src[1], undef, flags);
+                       free_temp(fp, temp[0]);
+#else
+                       emit_arith(fp, PFS_OP_DP4, dest, mask,
+                                  swizzle(src[0], X, Y, Z, ONE), src[1],
+                                  undef, flags);
+#endif
+                       break;
+               case OPCODE_DST:
+                       src[0] = t_src(fp, fpi->SrcReg[0]);
+                       src[1] = t_src(fp, fpi->SrcReg[1]);
+                       /* dest.y = src0.y * src1.y */
+                       if (mask & WRITEMASK_Y)
+                               emit_arith(fp, PFS_OP_MAD, dest, WRITEMASK_Y,
+                                          keep(src[0]), keep(src[1]),
+                                          pfs_zero, flags);
+                       /* dest.z = src0.z */
+                       if (mask & WRITEMASK_Z)
+                               emit_arith(fp, PFS_OP_MAD, dest, WRITEMASK_Z,
+                                          src[0], pfs_one, pfs_zero, flags);
+                       /* result.x = 1.0
+                        * result.w = src1.w */
+                       if (mask & WRITEMASK_XW) {
+                               REG_SET_VSWZ(src[1], SWIZZLE_111);      /*Cheat */
+                               emit_arith(fp, PFS_OP_MAD, dest,
+                                          mask & WRITEMASK_XW,
+                                          src[1], pfs_one, pfs_zero, flags);
+                       }
+                       break;
+               case OPCODE_EX2:
+                       src[0] = t_scalar_src(fp, fpi->SrcReg[0]);
+                       emit_arith(fp, PFS_OP_EX2, dest, mask,
+                                  src[0], undef, undef, flags);
+                       break;
+               case OPCODE_FLR:
+                       src[0] = t_src(fp, fpi->SrcReg[0]);
+                       temp[0] = get_temp_reg(fp);
+                       /* FRC temp, src0
+                        * MAD dest, src0, 1.0, -temp
+                        */
+                       emit_arith(fp, PFS_OP_FRC, temp[0], mask,
+                                  keep(src[0]), undef, undef, 0);
+                       emit_arith(fp, PFS_OP_MAD, dest, mask,
+                                  src[0], pfs_one, negate(temp[0]), flags);
+                       free_temp(fp, temp[0]);
+                       break;
+               case OPCODE_FRC:
+                       src[0] = t_src(fp, fpi->SrcReg[0]);
+                       emit_arith(fp, PFS_OP_FRC, dest, mask,
+                                  src[0], undef, undef, flags);
+                       break;
+               case OPCODE_KIL:
+                       emit_tex(fp, fpi, R300_FPITX_OP_KIL);
+                       break;
+               case OPCODE_LG2:
+                       src[0] = t_scalar_src(fp, fpi->SrcReg[0]);
+                       emit_arith(fp, PFS_OP_LG2, dest, mask,
+                                  src[0], undef, undef, flags);
+                       break;
+               case OPCODE_LIT:
+                       src[0] = t_src(fp, fpi->SrcReg[0]);
+                       emit_lit(fp, dest, mask, src[0], flags);
+                       break;
+               case OPCODE_LRP:
+                       src[0] = t_src(fp, fpi->SrcReg[0]);
+                       src[1] = t_src(fp, fpi->SrcReg[1]);
+                       src[2] = t_src(fp, fpi->SrcReg[2]);
+                       /* result = tmp0tmp1 + (1 - tmp0)tmp2
+                        *        = tmp0tmp1 + tmp2 + (-tmp0)tmp2
+                        *     MAD temp, -tmp0, tmp2, tmp2
+                        *     MAD result, tmp0, tmp1, temp
+                        */
+                       temp[0] = get_temp_reg(fp);
+                       emit_arith(fp, PFS_OP_MAD, temp[0], mask,
+                                  negate(keep(src[0])), keep(src[2]), src[2],
+                                  0);
+                       emit_arith(fp, PFS_OP_MAD, dest, mask,
+                                  src[0], src[1], temp[0], flags);
+                       free_temp(fp, temp[0]);
+                       break;
+               case OPCODE_MAD:
+                       src[0] = t_src(fp, fpi->SrcReg[0]);
+                       src[1] = t_src(fp, fpi->SrcReg[1]);
+                       src[2] = t_src(fp, fpi->SrcReg[2]);
+                       emit_arith(fp, PFS_OP_MAD, dest, mask,
+                                  src[0], src[1], src[2], flags);
+                       break;
+               case OPCODE_MAX:
+                       src[0] = t_src(fp, fpi->SrcReg[0]);
+                       src[1] = t_src(fp, fpi->SrcReg[1]);
+                       emit_arith(fp, PFS_OP_MAX, dest, mask,
+                                  src[0], src[1], undef, flags);
+                       break;
+               case OPCODE_MIN:
+                       src[0] = t_src(fp, fpi->SrcReg[0]);
+                       src[1] = t_src(fp, fpi->SrcReg[1]);
+                       emit_arith(fp, PFS_OP_MIN, dest, mask,
+                                  src[0], src[1], undef, flags);
+                       break;
+               case OPCODE_MOV:
+               case OPCODE_SWZ:
+                       src[0] = t_src(fp, fpi->SrcReg[0]);
+                       emit_arith(fp, PFS_OP_MAD, dest, mask,
+                                  src[0], pfs_one, pfs_zero, flags);
+                       break;
+               case OPCODE_MUL:
+                       src[0] = t_src(fp, fpi->SrcReg[0]);
+                       src[1] = t_src(fp, fpi->SrcReg[1]);
+                       emit_arith(fp, PFS_OP_MAD, dest, mask,
+                                  src[0], src[1], pfs_zero, flags);
+                       break;
+               case OPCODE_POW:
+                       src[0] = t_scalar_src(fp, fpi->SrcReg[0]);
+                       src[1] = t_scalar_src(fp, fpi->SrcReg[1]);
+                       temp[0] = get_temp_reg(fp);
+                       emit_arith(fp, PFS_OP_LG2, temp[0], WRITEMASK_W,
+                                  src[0], undef, undef, 0);
+                       emit_arith(fp, PFS_OP_MAD, temp[0], WRITEMASK_W,
+                                  temp[0], src[1], pfs_zero, 0);
+                       emit_arith(fp, PFS_OP_EX2, dest, fpi->DstReg.WriteMask,
+                                  temp[0], undef, undef, 0);
+                       free_temp(fp, temp[0]);
+                       break;
+               case OPCODE_RCP:
+                       src[0] = t_scalar_src(fp, fpi->SrcReg[0]);
+                       emit_arith(fp, PFS_OP_RCP, dest, mask,
+                                  src[0], undef, undef, flags);
+                       break;
+               case OPCODE_RSQ:
+                       src[0] = t_scalar_src(fp, fpi->SrcReg[0]);
+                       emit_arith(fp, PFS_OP_RSQ, dest, mask,
+                                  absolute(src[0]), pfs_zero, pfs_zero, flags);
+                       break;
+               case OPCODE_SCS:
+                       /*
+                        * scs using a parabola :
+                        * scs(x):
+                        *   result.x = sin(-abs(x)+0.5*PI)  (cos)
+                        *   result.y = sin(x)               (sin)
+                        *
+                        */
+                       temp[0] = get_temp_reg(fp);
+                       temp[1] = get_temp_reg(fp);
+                       const_sin[0] = emit_const4fv(fp, SinCosConsts[0]);
+                       const_sin[1] = emit_const4fv(fp, SinCosConsts[1]);
+                       src[0] = t_scalar_src(fp, fpi->SrcReg[0]);
+
+                       /* x = -abs(x)+0.5*PI */
+                       emit_arith(fp, PFS_OP_MAD, temp[0], WRITEMASK_Z, swizzle(const_sin[0], Z, Z, Z, Z),     //PI
+                                  pfs_half,
+                                  negate(abs
+                                         (swizzle(keep(src[0]), X, X, X, X))),
+                                  0);
+
+                       /* C*x (sin) */
+                       emit_arith(fp, PFS_OP_MAD, temp[0], WRITEMASK_W,
+                                  swizzle(const_sin[0], Y, Y, Y, Y),
+                                  swizzle(keep(src[0]), X, X, X, X),
+                                  pfs_zero, 0);
+
+                       /* B*x, C*x (cos) */
+                       emit_arith(fp, PFS_OP_MAD, temp[0],
+                                  WRITEMASK_X | WRITEMASK_Y, swizzle(temp[0],
+                                                                     Z, Z, Z,
+                                                                     Z),
+                                  const_sin[0], pfs_zero, 0);
+
+                       /* B*x (sin) */
+                       emit_arith(fp, PFS_OP_MAD, temp[1], WRITEMASK_W,
+                                  swizzle(const_sin[0], X, X, X, X),
+                                  keep(src[0]), pfs_zero, 0);
+
+                       /* y = B*x + C*x*abs(x) (sin) */
+                       emit_arith(fp, PFS_OP_MAD, temp[1], WRITEMASK_Z,
+                                  absolute(src[0]),
+                                  swizzle(temp[0], W, W, W, W),
+                                  swizzle(temp[1], W, W, W, W), 0);
+
+                       /* y = B*x + C*x*abs(x) (cos) */
+                       emit_arith(fp, PFS_OP_MAD, temp[1], WRITEMASK_W,
+                                  swizzle(temp[0], Y, Y, Y, Y),
+                                  absolute(swizzle(temp[0], Z, Z, Z, Z)),
+                                  swizzle(temp[0], X, X, X, X), 0);
+
+                       /* y*abs(y) - y (cos), y*abs(y) - y (sin) */
+                       emit_arith(fp, PFS_OP_MAD, temp[0],
+                                  WRITEMASK_X | WRITEMASK_Y, swizzle(temp[1],
+                                                                     W, Z, Y,
+                                                                     X),
+                                  absolute(swizzle(temp[1], W, Z, Y, X)),
+                                  negate(swizzle(temp[1], W, Z, Y, X)), 0);
+
+                       /* dest.xy = mad(temp.xy, P, temp2.wz) */
+                       emit_arith(fp, PFS_OP_MAD, dest,
+                                  mask & (WRITEMASK_X | WRITEMASK_Y), temp[0],
+                                  swizzle(const_sin[0], W, W, W, W),
+                                  swizzle(temp[1], W, Z, Y, X), flags);
+
+                       free_temp(fp, temp[0]);
+                       free_temp(fp, temp[1]);
+                       break;
+               case OPCODE_SGE:
+                       src[0] = t_src(fp, fpi->SrcReg[0]);
+                       src[1] = t_src(fp, fpi->SrcReg[1]);
+                       temp[0] = get_temp_reg(fp);
+                       /* temp = src0 - src1
+                        * dest.c = (temp.c < 0.0) ? 0 : 1
+                        */
+                       emit_arith(fp, PFS_OP_MAD, temp[0], mask,
+                                  src[0], pfs_one, negate(src[1]), 0);
+                       emit_arith(fp, PFS_OP_CMP, dest, mask,
+                                  pfs_one, pfs_zero, temp[0], 0);
+                       free_temp(fp, temp[0]);
+                       break;
+               case OPCODE_SIN:
+                       /*
+                        *  using a parabola:
+                        * sin(x) = 4/pi * x + -4/(pi*pi) * x * abs(x)
+                        * extra precision is obtained by weighting against
+                        * itself squared.
+                        */
+
+                       temp[0] = get_temp_reg(fp);
+                       const_sin[0] = emit_const4fv(fp, SinCosConsts[0]);
+                       const_sin[1] = emit_const4fv(fp, SinCosConsts[1]);
+                       src[0] = t_scalar_src(fp, fpi->SrcReg[0]);
+
+                       /* do range reduction */
+
+                       emit_arith(fp, PFS_OP_MAD, temp[0], WRITEMASK_X,
+                                  swizzle(keep(src[0]), X, X, X, X),
+                                  swizzle(const_sin[1], Z, Z, Z, Z),
+                                  pfs_half, 0);
+
+                       emit_arith(fp, PFS_OP_FRC, temp[0], WRITEMASK_X,
+                                  swizzle(temp[0], X, X, X, X),
+                                  undef, undef, 0);
+
+                       emit_arith(fp, PFS_OP_MAD, temp[0], WRITEMASK_Z, swizzle(temp[0], X, X, X, X), swizzle(const_sin[1], W, W, W, W),       //2*PI
+                                  negate(swizzle(const_sin[0], Z, Z, Z, Z)),   //PI
+                                  0);
+
+                       /* SIN */
+
+                       emit_arith(fp, PFS_OP_MAD, temp[0],
+                                  WRITEMASK_X | WRITEMASK_Y, swizzle(temp[0],
+                                                                     Z, Z, Z,
+                                                                     Z),
+                                  const_sin[0], pfs_zero, 0);
+
+                       emit_arith(fp, PFS_OP_MAD, temp[0], WRITEMASK_X,
+                                  swizzle(temp[0], Y, Y, Y, Y),
+                                  absolute(swizzle(temp[0], Z, Z, Z, Z)),
+                                  swizzle(temp[0], X, X, X, X), 0);
+
+                       emit_arith(fp, PFS_OP_MAD, temp[0], WRITEMASK_Y,
+                                  swizzle(temp[0], X, X, X, X),
+                                  absolute(swizzle(temp[0], X, X, X, X)),
+                                  negate(swizzle(temp[0], X, X, X, X)), 0);
+
+                       emit_arith(fp, PFS_OP_MAD, dest, mask,
+                                  swizzle(temp[0], Y, Y, Y, Y),
+                                  swizzle(const_sin[0], W, W, W, W),
+                                  swizzle(temp[0], X, X, X, X), flags);
+
+                       free_temp(fp, temp[0]);
+                       break;
+               case OPCODE_SLT:
+                       src[0] = t_src(fp, fpi->SrcReg[0]);
+                       src[1] = t_src(fp, fpi->SrcReg[1]);
+                       temp[0] = get_temp_reg(fp);
+                       /* temp = src0 - src1
+                        * dest.c = (temp.c < 0.0) ? 1 : 0
+                        */
+                       emit_arith(fp, PFS_OP_MAD, temp[0], mask,
+                                  src[0], pfs_one, negate(src[1]), 0);
+                       emit_arith(fp, PFS_OP_CMP, dest, mask,
+                                  pfs_zero, pfs_one, temp[0], 0);
+                       free_temp(fp, temp[0]);
+                       break;
+               case OPCODE_SUB:
+                       src[0] = t_src(fp, fpi->SrcReg[0]);
+                       src[1] = t_src(fp, fpi->SrcReg[1]);
+                       emit_arith(fp, PFS_OP_MAD, dest, mask,
+                                  src[0], pfs_one, negate(src[1]), flags);
+                       break;
+               case OPCODE_TEX:
+                       emit_tex(fp, fpi, R300_FPITX_OP_TEX);
+                       break;
+               case OPCODE_TXB:
+                       emit_tex(fp, fpi, R300_FPITX_OP_TXB);
+                       break;
+               case OPCODE_TXP:
+                       emit_tex(fp, fpi, R300_FPITX_OP_TXP);
+                       break;
+               case OPCODE_XPD:{
+                               src[0] = t_src(fp, fpi->SrcReg[0]);
+                               src[1] = t_src(fp, fpi->SrcReg[1]);
+                               temp[0] = get_temp_reg(fp);
+                               /* temp = src0.zxy * src1.yzx */
+                               emit_arith(fp, PFS_OP_MAD, temp[0],
+                                          WRITEMASK_XYZ, swizzle(keep(src[0]),
+                                                                 Z, X, Y, W),
+                                          swizzle(keep(src[1]), Y, Z, X, W),
+                                          pfs_zero, 0);
+                               /* dest.xyz = src0.yzx * src1.zxy - temp
+                                * dest.w       = undefined
+                                * */
+                               emit_arith(fp, PFS_OP_MAD, dest,
+                                          mask & WRITEMASK_XYZ, swizzle(src[0],
+                                                                        Y, Z,
+                                                                        X, W),
+                                          swizzle(src[1], Z, X, Y, W),
+                                          negate(temp[0]), flags);
+                               /* cleanup */
+                               free_temp(fp, temp[0]);
+                               break;
+                       }
+               default:
+                       ERROR("unknown fpi->Opcode %d\n", fpi->Opcode);
+                       break;
+               }
+
+               if (fp->error)
+                       return GL_FALSE;
+
+       }
+
+       return GL_TRUE;
+}
+
+static void insert_wpos(struct gl_program *prog)
+{
+       static gl_state_index tokens[STATE_LENGTH] = {
+               STATE_INTERNAL, STATE_R300_WINDOW_DIMENSION, 0, 0, 0
+       };
+       struct prog_instruction *fpi;
+       GLuint window_index;
+       int i = 0;
+       GLuint tempregi = prog->NumTemporaries;
+       /* should do something else if no temps left... */
+       prog->NumTemporaries++;
+
+       fpi = _mesa_alloc_instructions(prog->NumInstructions + 3);
+       _mesa_init_instructions(fpi, prog->NumInstructions + 3);
+
+       /* perspective divide */
+       fpi[i].Opcode = OPCODE_RCP;
+
+       fpi[i].DstReg.File = PROGRAM_TEMPORARY;
+       fpi[i].DstReg.Index = tempregi;
+       fpi[i].DstReg.WriteMask = WRITEMASK_W;
+       fpi[i].DstReg.CondMask = COND_TR;
+
+       fpi[i].SrcReg[0].File = PROGRAM_INPUT;
+       fpi[i].SrcReg[0].Index = FRAG_ATTRIB_WPOS;
+       fpi[i].SrcReg[0].Swizzle = SWIZZLE_WWWW;
+       i++;
+
+       fpi[i].Opcode = OPCODE_MUL;
+
+       fpi[i].DstReg.File = PROGRAM_TEMPORARY;
+       fpi[i].DstReg.Index = tempregi;
+       fpi[i].DstReg.WriteMask = WRITEMASK_XYZ;
+       fpi[i].DstReg.CondMask = COND_TR;
+
+       fpi[i].SrcReg[0].File = PROGRAM_INPUT;
+       fpi[i].SrcReg[0].Index = FRAG_ATTRIB_WPOS;
+       fpi[i].SrcReg[0].Swizzle = SWIZZLE_XYZW;
+
+       fpi[i].SrcReg[1].File = PROGRAM_TEMPORARY;
+       fpi[i].SrcReg[1].Index = tempregi;
+       fpi[i].SrcReg[1].Swizzle = SWIZZLE_WWWW;
+       i++;
+
+       /* viewport transformation */
+       window_index = _mesa_add_state_reference(prog->Parameters, tokens);
+
+       fpi[i].Opcode = OPCODE_MAD;
+
+       fpi[i].DstReg.File = PROGRAM_TEMPORARY;
+       fpi[i].DstReg.Index = tempregi;
+       fpi[i].DstReg.WriteMask = WRITEMASK_XYZ;
+       fpi[i].DstReg.CondMask = COND_TR;
+
+       fpi[i].SrcReg[0].File = PROGRAM_TEMPORARY;
+       fpi[i].SrcReg[0].Index = tempregi;
+       fpi[i].SrcReg[0].Swizzle =
+           MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO);
+
+       fpi[i].SrcReg[1].File = PROGRAM_STATE_VAR;
+       fpi[i].SrcReg[1].Index = window_index;
+       fpi[i].SrcReg[1].Swizzle =
+           MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO);
+
+       fpi[i].SrcReg[2].File = PROGRAM_STATE_VAR;
+       fpi[i].SrcReg[2].Index = window_index;
+       fpi[i].SrcReg[2].Swizzle =
+           MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO);
+       i++;
+
+       _mesa_copy_instructions(&fpi[i], prog->Instructions,
+                               prog->NumInstructions);
+
+       free(prog->Instructions);
+
+       prog->Instructions = fpi;
+
+       prog->NumInstructions += i;
+       fpi = &prog->Instructions[prog->NumInstructions - 1];
+
+       assert(fpi->Opcode == OPCODE_END);
+
+       for (fpi = &prog->Instructions[3]; fpi->Opcode != OPCODE_END; fpi++) {
+               for (i = 0; i < 3; i++)
+                       if (fpi->SrcReg[i].File == PROGRAM_INPUT &&
+                           fpi->SrcReg[i].Index == FRAG_ATTRIB_WPOS) {
+                               fpi->SrcReg[i].File = PROGRAM_TEMPORARY;
+                               fpi->SrcReg[i].Index = tempregi;
+                       }
+       }
+}
+
+/* - Init structures
+ * - Determine what hwregs each input corresponds to
+ */
+static void init_program(r300ContextPtr r300, struct r300_fragment_program *fp)
+{
+       struct r300_pfs_compile_state *cs = NULL;
+       struct gl_fragment_program *mp = &fp->mesa_program;
+       struct prog_instruction *fpi;
+       GLuint InputsRead = mp->Base.InputsRead;
+       GLuint temps_used = 0;  /* for fp->temps[] */
+       int i, j;
+
+       /* New compile, reset tracking data */
+       fp->optimization =
+           driQueryOptioni(&r300->radeon.optionCache, "fp_optimization");
+       fp->translated = GL_FALSE;
+       fp->error = GL_FALSE;
+       fp->cs = cs = &(R300_CONTEXT(fp->ctx)->state.pfs_compile);
+       fp->tex.length = 0;
+       fp->cur_node = 0;
+       fp->first_node_has_tex = 0;
+       fp->const_nr = 0;
+       fp->max_temp_idx = 0;
+       fp->node[0].alu_end = -1;
+       fp->node[0].tex_end = -1;
+
+       _mesa_memset(cs, 0, sizeof(*fp->cs));
+       for (i = 0; i < PFS_MAX_ALU_INST; i++) {
+               for (j = 0; j < 3; j++) {
+                       cs->slot[i].vsrc[j] = SRC_CONST;
+                       cs->slot[i].ssrc[j] = SRC_CONST;
+               }
+       }
+
+       /* Work out what temps the Mesa inputs correspond to, this must match
+        * what setup_rs_unit does, which shouldn't be a problem as rs_unit
+        * configures itself based on the fragprog's InputsRead
+        *
+        * NOTE: this depends on get_hw_temp() allocating registers in order,
+        * starting from register 0.
+        */
+
+       /* Texcoords come first */
+       for (i = 0; i < fp->ctx->Const.MaxTextureUnits; i++) {
+               if (InputsRead & (FRAG_BIT_TEX0 << i)) {
+                       cs->inputs[FRAG_ATTRIB_TEX0 + i].refcount = 0;
+                       cs->inputs[FRAG_ATTRIB_TEX0 + i].reg =
+                           get_hw_temp(fp, 0);
+               }
+       }
+       InputsRead &= ~FRAG_BITS_TEX_ANY;
+
+       /* fragment position treated as a texcoord */
+       if (InputsRead & FRAG_BIT_WPOS) {
+               cs->inputs[FRAG_ATTRIB_WPOS].refcount = 0;
+               cs->inputs[FRAG_ATTRIB_WPOS].reg = get_hw_temp(fp, 0);
+               insert_wpos(&mp->Base);
+       }
+       InputsRead &= ~FRAG_BIT_WPOS;
+
+       /* Then primary colour */
+       if (InputsRead & FRAG_BIT_COL0) {
+               cs->inputs[FRAG_ATTRIB_COL0].refcount = 0;
+               cs->inputs[FRAG_ATTRIB_COL0].reg = get_hw_temp(fp, 0);
+       }
+       InputsRead &= ~FRAG_BIT_COL0;
+
+       /* Secondary color */
+       if (InputsRead & FRAG_BIT_COL1) {
+               cs->inputs[FRAG_ATTRIB_COL1].refcount = 0;
+               cs->inputs[FRAG_ATTRIB_COL1].reg = get_hw_temp(fp, 0);
+       }
+       InputsRead &= ~FRAG_BIT_COL1;
+
+       /* Anything else */
+       if (InputsRead) {
+               WARN_ONCE("Don't know how to handle inputs 0x%x\n", InputsRead);
+               /* force read from hwreg 0 for now */
+               for (i = 0; i < 32; i++)
+                       if (InputsRead & (1 << i))
+                               cs->inputs[i].reg = 0;
+       }
+
+       /* Pre-parse the mesa program, grabbing refcounts on input/temp regs.
+        * That way, we can free up the reg when it's no longer needed
+        */
+       if (!mp->Base.Instructions) {
+               ERROR("No instructions found in program\n");
+               return;
+       }
+
+       for (fpi = mp->Base.Instructions; fpi->Opcode != OPCODE_END; fpi++) {
+               int idx;
+
+               for (i = 0; i < 3; i++) {
+                       idx = fpi->SrcReg[i].Index;
+                       switch (fpi->SrcReg[i].File) {
+                       case PROGRAM_TEMPORARY:
+                               if (!(temps_used & (1 << idx))) {
+                                       cs->temps[idx].reg = -1;
+                                       cs->temps[idx].refcount = 1;
+                                       temps_used |= (1 << idx);
+                               } else
+                                       cs->temps[idx].refcount++;
+                               break;
+                       case PROGRAM_INPUT:
+                               cs->inputs[idx].refcount++;
+                               break;
+                       default:
+                               break;
+                       }
+               }
+
+               idx = fpi->DstReg.Index;
+               if (fpi->DstReg.File == PROGRAM_TEMPORARY) {
+                       if (!(temps_used & (1 << idx))) {
+                               cs->temps[idx].reg = -1;
+                               cs->temps[idx].refcount = 1;
+                               temps_used |= (1 << idx);
+                       } else
+                               cs->temps[idx].refcount++;
+               }
+       }
+       cs->temp_in_use = temps_used;
+}
+
+static void update_params(struct r300_fragment_program *fp)
+{
+       struct gl_fragment_program *mp = &fp->mesa_program;
+
+       /* Ask Mesa nicely to fill in ParameterValues for us */
+       if (mp->Base.Parameters)
+               _mesa_load_state_parameters(fp->ctx, mp->Base.Parameters);
+}
+
+void r500TranslateFragmentShader(r300ContextPtr r300,
+                                struct r300_fragment_program *fp)
+{
+       struct r300_pfs_compile_state *cs = NULL;
+
+       if (!fp->translated) {
+
+               init_program(r300, fp);
+               cs = fp->cs;
+
+               if (parse_program(fp) == GL_FALSE) {
+                       dump_program(fp);
+                       return;
+               }
+
+               /* Finish off */
+               fp->node[fp->cur_node].alu_end =
+                   cs->nrslots - fp->node[fp->cur_node].alu_offset - 1;
+               if (fp->node[fp->cur_node].tex_end < 0)
+                       fp->node[fp->cur_node].tex_end = 0;
+               fp->alu_offset = 0;
+               fp->alu_end = cs->nrslots - 1;
+               fp->tex_offset = 0;
+               fp->tex_end = fp->tex.length ? fp->tex.length - 1 : 0;
+               assert(fp->node[fp->cur_node].alu_end >= 0);
+               assert(fp->alu_end >= 0);
+
+               fp->translated = GL_TRUE;
+               if (RADEON_DEBUG & DEBUG_PIXEL)
+                       dump_program(fp);
+               r300UpdateStateParameters(fp->ctx, _NEW_PROGRAM);
+       }
+
+       update_params(fp);
+}
+
+/* just some random things... */
+static void dump_program(struct r300_fragment_program *fp)
+{
+       int n, i, j;
+       static int pc = 0;
+
+       fprintf(stderr, "pc=%d*************************************\n", pc++);
+
+       fprintf(stderr, "Mesa program:\n");
+       fprintf(stderr, "-------------\n");
+       _mesa_print_program(&fp->mesa_program.Base);
+       fflush(stdout);
+
+       fprintf(stderr, "Hardware program\n");
+       fprintf(stderr, "----------------\n");
+
+       for (n = 0; n < (fp->cur_node + 1); n++) {
+               fprintf(stderr, "NODE %d: alu_offset: %d, tex_offset: %d, "
+                       "alu_end: %d, tex_end: %d\n", n,
+                       fp->node[n].alu_offset,
+                       fp->node[n].tex_offset,
+                       fp->node[n].alu_end, fp->node[n].tex_end);
+
+               if (fp->tex.length) {
+                       fprintf(stderr, "  TEX:\n");
+                       for (i = fp->node[n].tex_offset;
+                            i <= fp->node[n].tex_offset + fp->node[n].tex_end;
+                            ++i) {
+                               const char *instr;
+
+                               switch ((fp->tex.
+                                        inst[i] >> R300_FPITX_OPCODE_SHIFT) &
+                                       15) {
+                               case R300_FPITX_OP_TEX:
+                                       instr = "TEX";
+                                       break;
+                               case R300_FPITX_OP_KIL:
+                                       instr = "KIL";
+                                       break;
+                               case R300_FPITX_OP_TXP:
+                                       instr = "TXP";
+                                       break;
+                               case R300_FPITX_OP_TXB:
+                                       instr = "TXB";
+                                       break;
+                               default:
+                                       instr = "UNKNOWN";
+                               }
+
+                               fprintf(stderr,
+                                       "    %s t%i, %c%i, texture[%i]   (%08x)\n",
+                                       instr,
+                                       (fp->tex.
+                                        inst[i] >> R300_FPITX_DST_SHIFT) & 31,
+                                       (fp->tex.
+                                        inst[i] & R300_FPITX_SRC_CONST) ? 'c' :
+                                       't',
+                                       (fp->tex.
+                                        inst[i] >> R300_FPITX_SRC_SHIFT) & 31,
+                                       (fp->tex.
+                                        inst[i] & R300_FPITX_IMAGE_MASK) >>
+                                       R300_FPITX_IMAGE_SHIFT,
+                                       fp->tex.inst[i]);
+                       }
+               }
+
+               for (i = fp->node[n].alu_offset;
+                    i <= fp->node[n].alu_offset + fp->node[n].alu_end; ++i) {
+                       char srcc[3][10], dstc[20];
+                       char srca[3][10], dsta[20];
+                       char argc[3][20];
+                       char arga[3][20];
+                       char flags[5], tmp[10];
+
+                       for (j = 0; j < 3; ++j) {
+                               int regc = fp->alu.inst[i].inst1 >> (j * 6);
+                               int rega = fp->alu.inst[i].inst3 >> (j * 6);
+
+                               sprintf(srcc[j], "%c%i",
+                                       (regc & 32) ? 'c' : 't', regc & 31);
+                               sprintf(srca[j], "%c%i",
+                                       (rega & 32) ? 'c' : 't', rega & 31);
+                       }
+
+                       dstc[0] = 0;
+                       sprintf(flags, "%s%s%s",
+                               (fp->alu.inst[i].
+                                inst1 & R300_FPI1_DSTC_REG_X) ? "x" : "",
+                               (fp->alu.inst[i].
+                                inst1 & R300_FPI1_DSTC_REG_Y) ? "y" : "",
+                               (fp->alu.inst[i].
+                                inst1 & R300_FPI1_DSTC_REG_Z) ? "z" : "");
+                       if (flags[0] != 0) {
+                               sprintf(dstc, "t%i.%s ",
+                                       (fp->alu.inst[i].
+                                        inst1 >> R300_FPI1_DSTC_SHIFT) & 31,
+                                       flags);
+                       }
+                       sprintf(flags, "%s%s%s",
+                               (fp->alu.inst[i].
+                                inst1 & R300_FPI1_DSTC_OUTPUT_X) ? "x" : "",
+                               (fp->alu.inst[i].
+                                inst1 & R300_FPI1_DSTC_OUTPUT_Y) ? "y" : "",
+                               (fp->alu.inst[i].
+                                inst1 & R300_FPI1_DSTC_OUTPUT_Z) ? "z" : "");
+                       if (flags[0] != 0) {
+                               sprintf(tmp, "o%i.%s",
+                                       (fp->alu.inst[i].
+                                        inst1 >> R300_FPI1_DSTC_SHIFT) & 31,
+                                       flags);
+                               strcat(dstc, tmp);
+                       }
+
+                       dsta[0] = 0;
+                       if (fp->alu.inst[i].inst3 & R300_FPI3_DSTA_REG) {
+                               sprintf(dsta, "t%i.w ",
+                                       (fp->alu.inst[i].
+                                        inst3 >> R300_FPI3_DSTA_SHIFT) & 31);
+                       }
+                       if (fp->alu.inst[i].inst3 & R300_FPI3_DSTA_OUTPUT) {
+                               sprintf(tmp, "o%i.w ",
+                                       (fp->alu.inst[i].
+                                        inst3 >> R300_FPI3_DSTA_SHIFT) & 31);
+                               strcat(dsta, tmp);
+                       }
+                       if (fp->alu.inst[i].inst3 & R300_FPI3_DSTA_DEPTH) {
+                               strcat(dsta, "Z");
+                       }
+
+                       fprintf(stderr,
+                               "%3i: xyz: %3s %3s %3s -> %-20s (%08x)\n"
+                               "       w: %3s %3s %3s -> %-20s (%08x)\n", i,
+                               srcc[0], srcc[1], srcc[2], dstc,
+                               fp->alu.inst[i].inst1, srca[0], srca[1],
+                               srca[2], dsta, fp->alu.inst[i].inst3);
+
+                       for (j = 0; j < 3; ++j) {
+                               int regc = fp->alu.inst[i].inst0 >> (j * 7);
+                               int rega = fp->alu.inst[i].inst2 >> (j * 7);
+                               int d;
+                               char buf[20];
+
+                               d = regc & 31;
+                               if (d < 12) {
+                                       switch (d % 4) {
+                                       case R300_FPI0_ARGC_SRC0C_XYZ:
+                                               sprintf(buf, "%s.xyz",
+                                                       srcc[d / 4]);
+                                               break;
+                                       case R300_FPI0_ARGC_SRC0C_XXX:
+                                               sprintf(buf, "%s.xxx",
+                                                       srcc[d / 4]);
+                                               break;
+                                       case R300_FPI0_ARGC_SRC0C_YYY:
+                                               sprintf(buf, "%s.yyy",
+                                                       srcc[d / 4]);
+                                               break;
+                                       case R300_FPI0_ARGC_SRC0C_ZZZ:
+                                               sprintf(buf, "%s.zzz",
+                                                       srcc[d / 4]);
+                                               break;
+                                       }
+                               } else if (d < 15) {
+                                       sprintf(buf, "%s.www", srca[d - 12]);
+                               } else if (d == 20) {
+                                       sprintf(buf, "0.0");
+                               } else if (d == 21) {
+                                       sprintf(buf, "1.0");
+                               } else if (d == 22) {
+                                       sprintf(buf, "0.5");
+                               } else if (d >= 23 && d < 32) {
+                                       d -= 23;
+                                       switch (d / 3) {
+                                       case 0:
+                                               sprintf(buf, "%s.yzx",
+                                                       srcc[d % 3]);
+                                               break;
+                                       case 1:
+                                               sprintf(buf, "%s.zxy",
+                                                       srcc[d % 3]);
+                                               break;
+                                       case 2:
+                                               sprintf(buf, "%s.Wzy",
+                                                       srcc[d % 3]);
+                                               break;
+                                       }
+                               } else {
+                                       sprintf(buf, "%i", d);
+                               }
+
+                               sprintf(argc[j], "%s%s%s%s",
+                                       (regc & 32) ? "-" : "",
+                                       (regc & 64) ? "|" : "",
+                                       buf, (regc & 64) ? "|" : "");
+
+                               d = rega & 31;
+                               if (d < 9) {
+                                       sprintf(buf, "%s.%c", srcc[d / 3],
+                                               'x' + (char)(d % 3));
+                               } else if (d < 12) {
+                                       sprintf(buf, "%s.w", srca[d - 9]);
+                               } else if (d == 16) {
+                                       sprintf(buf, "0.0");
+                               } else if (d == 17) {
+                                       sprintf(buf, "1.0");
+                               } else if (d == 18) {
+                                       sprintf(buf, "0.5");
+                               } else {
+                                       sprintf(buf, "%i", d);
+                               }
+
+                               sprintf(arga[j], "%s%s%s%s",
+                                       (rega & 32) ? "-" : "",
+                                       (rega & 64) ? "|" : "",
+                                       buf, (rega & 64) ? "|" : "");
+                       }
+
+                       fprintf(stderr, "     xyz: %8s %8s %8s    op: %08x\n"
+                               "       w: %8s %8s %8s    op: %08x\n",
+                               argc[0], argc[1], argc[2],
+                               fp->alu.inst[i].inst0, arga[0], arga[1],
+                               arga[2], fp->alu.inst[i].inst2);
+               }
+       }
+}
diff --git a/src/mesa/drivers/dri/r300/r500_fragprog.h b/src/mesa/drivers/dri/r300/r500_fragprog.h
new file mode 100644 (file)
index 0000000..72fca77
--- /dev/null
@@ -0,0 +1,104 @@
+/*
+ * Copyright (C) 2005 Ben Skeggs.
+ *
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+/*
+ * Authors:
+ *   Ben Skeggs <darktama@iinet.net.au>
+ *   Jerome Glisse <j.glisse@gmail.com>
+ */
+#ifndef __R300_FRAGPROG_H_
+#define __R300_FRAGPROG_H_
+
+#include "glheader.h"
+#include "macros.h"
+#include "enums.h"
+#include "shader/program.h"
+#include "shader/prog_instruction.h"
+
+#include "r300_context.h"
+
+typedef struct r300_fragment_program_swizzle {
+       GLuint length;
+       GLuint src[4];
+       GLuint inst[8];
+} r300_fragment_program_swizzle_t;
+
+/* supported hw opcodes */
+#define PFS_OP_MAD 0
+#define PFS_OP_DP3 1
+#define PFS_OP_DP4 2
+#define PFS_OP_MIN 3
+#define PFS_OP_MAX 4
+#define PFS_OP_CMP 5
+#define PFS_OP_FRC 6
+#define PFS_OP_EX2 7
+#define PFS_OP_LG2 8
+#define PFS_OP_RCP 9
+#define PFS_OP_RSQ 10
+#define PFS_OP_REPL_ALPHA 11
+#define PFS_OP_CMPH 12
+#define MAX_PFS_OP 12
+
+#define PFS_FLAG_SAT   (1 << 0)
+#define PFS_FLAG_ABS   (1 << 1)
+
+#define ARG_NEG                        (1 << 5)
+#define ARG_ABS                        (1 << 6)
+#define ARG_MASK               (127 << 0)
+#define ARG_STRIDE             7
+#define SRC_CONST              (1 << 5)
+#define SRC_MASK               (63 << 0)
+#define SRC_STRIDE             6
+
+#define NOP_INST0 (                                             \
+               (R300_FPI0_OUTC_MAD) |                           \
+               (R300_FPI0_ARGC_ZERO << R300_FPI0_ARG0C_SHIFT) | \
+               (R300_FPI0_ARGC_ZERO << R300_FPI0_ARG1C_SHIFT) | \
+               (R300_FPI0_ARGC_ZERO << R300_FPI0_ARG2C_SHIFT))
+#define NOP_INST1 (                                         \
+               ((0 | SRC_CONST) << R300_FPI1_SRC0C_SHIFT) | \
+               ((0 | SRC_CONST) << R300_FPI1_SRC1C_SHIFT) | \
+               ((0 | SRC_CONST) << R300_FPI1_SRC2C_SHIFT))
+#define NOP_INST2 ( \
+               (R300_FPI2_OUTA_MAD) |                           \
+               (R300_FPI2_ARGA_ZERO << R300_FPI2_ARG0A_SHIFT) | \
+               (R300_FPI2_ARGA_ZERO << R300_FPI2_ARG1A_SHIFT) | \
+               (R300_FPI2_ARGA_ZERO << R300_FPI2_ARG2A_SHIFT))
+#define NOP_INST3 (                                         \
+               ((0 | SRC_CONST) << R300_FPI3_SRC0A_SHIFT) | \
+               ((0 | SRC_CONST) << R300_FPI3_SRC1A_SHIFT) | \
+               ((0 | SRC_CONST) << R300_FPI3_SRC2A_SHIFT))
+
+#define DRI_CONF_FP_OPTIMIZATION_SPEED   0
+#define DRI_CONF_FP_OPTIMIZATION_QUALITY 1
+
+struct r300_fragment_program;
+
+extern void r300TranslateFragmentShader(r300ContextPtr r300,
+                                       struct r300_fragment_program *fp);
+
+#endif