tgsi: add support for texture offsets to the TGSI IR. (v2)

This adds tokens for texture offsets, to store 4 * swizzled vec 3
for use in TXF and other opcodes.

It also contains TGSI exec changes for softpipe to use this code,
along with GLSL->TGSI support for TXF.

v2: add some more comments, add back padding I removed.

Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Brian Paul <brianp@vmware.com>

8 files changed, 132 insertions, 9 deletions

diff --git a/src/gallium/auxiliary/tgsi/tgsi_build.c b/src/gallium/auxiliary/tgsi/tgsi_build.c
index 269940e..6ec2b0d 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_build.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_build.c
@@ -631,6 +631,7 @@ tgsi_default_instruction_texture( void )
    struct tgsi_instruction_texture instruction_texture;
 
    instruction_texture.Texture = TGSI_TEXTURE_UNKNOWN;
+   instruction_texture.NumOffsets = 0;
    instruction_texture.Padding = 0;
 
    return instruction_texture;
@@ -639,6 +640,7 @@ tgsi_default_instruction_texture( void )
 static struct tgsi_instruction_texture
 tgsi_build_instruction_texture(
    unsigned texture,
+   unsigned num_offsets,
    struct tgsi_token *prev_token,
    struct tgsi_instruction *instruction,
    struct tgsi_header *header )
@@ -646,6 +648,7 @@ tgsi_build_instruction_texture(
    struct tgsi_instruction_texture instruction_texture;
 
    instruction_texture.Texture = texture;
+   instruction_texture.NumOffsets = num_offsets;
    instruction_texture.Padding = 0;
    instruction->Texture = 1;
 
@@ -654,6 +657,43 @@ tgsi_build_instruction_texture(
    return instruction_texture;
 }
 
+
+static struct tgsi_texture_offset
+tgsi_default_texture_offset( void )
+{
+   struct tgsi_texture_offset texture_offset;
+
+   texture_offset.Index = 0;
+   texture_offset.File = 0;
+   texture_offset.SwizzleX = 0;
+   texture_offset.SwizzleY = 0;
+   texture_offset.SwizzleZ = 0;
+   texture_offset.Padding = 0;
+
+   return texture_offset;
+}
+
+static struct tgsi_texture_offset
+tgsi_build_texture_offset(
+   int index, int file, int swizzle_x, int swizzle_y, int swizzle_z,
+   struct tgsi_token *prev_token,
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header )
+{
+   struct tgsi_texture_offset texture_offset;
+
+   texture_offset.Index = index;
+   texture_offset.File = file;
+   texture_offset.SwizzleX = swizzle_x;
+   texture_offset.SwizzleY = swizzle_y;
+   texture_offset.SwizzleZ = swizzle_z;
+   texture_offset.Padding = 0;
+
+   instruction_grow( instruction, header );
+
+   return texture_offset;
+}
+
 static struct tgsi_src_register
 tgsi_default_src_register( void )
 {
@@ -825,6 +865,9 @@ tgsi_default_full_instruction( void )
    full_instruction.Predicate = tgsi_default_instruction_predicate();
    full_instruction.Label = tgsi_default_instruction_label();
    full_instruction.Texture = tgsi_default_instruction_texture();
+   for( i = 0;  i < TGSI_FULL_MAX_TEX_OFFSETS; i++ ) {
+      full_instruction.TexOffsets[i] = tgsi_default_texture_offset();
+   }
    for( i = 0;  i < TGSI_FULL_MAX_DST_REGISTERS; i++ ) {
       full_instruction.Dst[i] = tgsi_default_full_dst_register();
    }
@@ -908,12 +951,31 @@ tgsi_build_full_instruction(
 
       *instruction_texture = tgsi_build_instruction_texture(
          full_inst->Texture.Texture,
+	 full_inst->Texture.NumOffsets,
          prev_token,
          instruction,
          header   );
       prev_token = (struct tgsi_token  *) instruction_texture;
-   }
 
+      for (i = 0; i < full_inst->Texture.NumOffsets; i++) {
+         struct tgsi_texture_offset *texture_offset;
+	
+         if ( maxsize <= size )
+            return 0;
+	 texture_offset = (struct tgsi_texture_offset *)&tokens[size];
+         size++;
+         *texture_offset = tgsi_build_texture_offset(
+            full_inst->TexOffsets[i].Index,
+            full_inst->TexOffsets[i].File,
+            full_inst->TexOffsets[i].SwizzleX,
+            full_inst->TexOffsets[i].SwizzleY,
+            full_inst->TexOffsets[i].SwizzleZ,
+            prev_token,
+            instruction,
+            header);
+         prev_token = (struct tgsi_token *) texture_offset;
+      }
+   }
    for( i = 0;  i <   full_inst->Instruction.NumDstRegs; i++ ) {
       const struct tgsi_full_dst_register *reg = &full_inst->Dst[i];
       struct tgsi_dst_register *dst_register;
diff --git a/src/gallium/auxiliary/tgsi/tgsi_dump.c b/src/gallium/auxiliary/tgsi/tgsi_dump.c
index c126620..7e2825e 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_dump.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_dump.c
@@ -693,6 +693,17 @@ iter_instruction(
    if (inst->Instruction.Texture) {
       TXT( ", " );
       ENM( inst->Texture.Texture, tgsi_texture_names );
+      for (i = 0; i < inst->Texture.NumOffsets; i++) {
+         TXT( ", " );
+         ENM( inst->TexOffsets[i].File, tgsi_file_names);
+         CHR( '[' );
+         SID( inst->TexOffsets[i].Index );
+         CHR( ']' );
+         CHR( '.' );
+         ENM( inst->TexOffsets[i].SwizzleX, tgsi_swizzle_names);
+         ENM( inst->TexOffsets[i].SwizzleY, tgsi_swizzle_names);
+         ENM( inst->TexOffsets[i].SwizzleZ, tgsi_swizzle_names);
+      }
    }
 
    switch (inst->Instruction.Opcode) {
diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.c b/src/gallium/auxiliary/tgsi/tgsi_exec.c
index 38dc1ef..fd118c5 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_exec.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_exec.c
@@ -1929,11 +1929,28 @@ exec_txf(struct tgsi_exec_machine *mach,
 	 const struct tgsi_full_instruction *inst)
 {
    struct tgsi_sampler *sampler;
-   const uint unit = inst->Src[1].Register.Index;
+   const uint unit = inst->Src[2].Register.Index;
    union tgsi_exec_channel r[4];
+   union tgsi_exec_channel offset[3];
    uint chan;
    float rgba[NUM_CHANNELS][QUAD_SIZE];
    int j;
+   int8_t offsets[3];
+
+   if (inst->Texture.NumOffsets == 1) {
+      union tgsi_exec_channel index;
+      index.i[0] = index.i[1] = index.i[2] = index.i[3] = inst->TexOffsets[0].Index;
+      fetch_src_file_channel(mach, inst->TexOffsets[0].File,
+                             inst->TexOffsets[0].SwizzleX, &index, &ZeroVec, &offset[0]);
+      fetch_src_file_channel(mach, inst->TexOffsets[0].File,
+                             inst->TexOffsets[0].SwizzleY, &index, &ZeroVec, &offset[1]);
+      fetch_src_file_channel(mach, inst->TexOffsets[0].File,
+                             inst->TexOffsets[0].SwizzleZ, &index, &ZeroVec, &offset[2]);
+     offsets[0] = offset[0].i[0];
+     offsets[1] = offset[1].i[0];
+     offsets[2] = offset[2].i[0];
+   } else
+     offsets[0] = offsets[1] = offsets[2] = 0;
 
    IFETCH(&r[3], 0, CHAN_W);
 
@@ -1959,7 +1976,8 @@ exec_txf(struct tgsi_exec_machine *mach,
    }      
 
    sampler = mach->Samplers[unit];
-   sampler->get_texel(sampler, r[0].i, r[1].i, r[2].i, r[3].i, rgba);
+   sampler->get_texel(sampler, r[0].i, r[1].i, r[2].i, r[3].i,
+		      offsets, rgba);
 
    for (j = 0; j < QUAD_SIZE; j++) {
       r[0].f[j] = rgba[0][j];
diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.h b/src/gallium/auxiliary/tgsi/tgsi_exec.h
index 3f6964c..a750715 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_exec.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_exec.h
@@ -94,7 +94,8 @@ struct tgsi_sampler
 		    int dims[4]);
    void (*get_texel)(struct tgsi_sampler *sampler, const int i[QUAD_SIZE],
 		     const int j[QUAD_SIZE], const int k[QUAD_SIZE],
-		     const int lod[QUAD_SIZE], float rgba[NUM_CHANNELS][QUAD_SIZE]);
+		     const int lod[QUAD_SIZE], const int8_t offset[3],
+		     float rgba[NUM_CHANNELS][QUAD_SIZE]);
 };
 
 #define TGSI_EXEC_NUM_TEMPS       128
diff --git a/src/gallium/auxiliary/tgsi/tgsi_parse.c b/src/gallium/auxiliary/tgsi/tgsi_parse.c
index fb36f9d..e1902eb 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_parse.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_parse.c
@@ -188,6 +188,9 @@ tgsi_parse_token(
 
       if (inst->Instruction.Texture) {
          next_token( ctx, &inst->Texture);
+         for( i = 0; i < inst->Texture.NumOffsets; i++ ) {
+            next_token( ctx, &inst->TexOffsets[i] );
+         }
       }
 
       assert( inst->Instruction.NumDstRegs <= TGSI_FULL_MAX_DST_REGISTERS );
diff --git a/src/gallium/auxiliary/tgsi/tgsi_parse.h b/src/gallium/auxiliary/tgsi/tgsi_parse.h
index b7a3c9b..f7b7e6e 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_parse.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_parse.h
@@ -86,6 +86,7 @@ struct tgsi_full_property
 
 #define TGSI_FULL_MAX_DST_REGISTERS 2
 #define TGSI_FULL_MAX_SRC_REGISTERS 5 /* SAMPLE_D has 5 */
+#define TGSI_FULL_MAX_TEX_OFFSETS 4
 
 struct tgsi_full_instruction
 {
@@ -95,6 +96,7 @@ struct tgsi_full_instruction
    struct tgsi_instruction_texture     Texture;
    struct tgsi_full_dst_register       Dst[TGSI_FULL_MAX_DST_REGISTERS];
    struct tgsi_full_src_register       Src[TGSI_FULL_MAX_SRC_REGISTERS];
+   struct tgsi_texture_offset          TexOffsets[TGSI_FULL_MAX_TEX_OFFSETS];
 };
 
 union tgsi_full_token
diff --git a/src/gallium/auxiliary/tgsi/tgsi_ureg.c b/src/gallium/auxiliary/tgsi/tgsi_ureg.c
index a920741..cada435 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_ureg.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_ureg.c
@@ -54,6 +54,7 @@ union tgsi_any_token {
    struct tgsi_instruction_predicate insn_predicate;
    struct tgsi_instruction_label insn_label;
    struct tgsi_instruction_texture insn_texture;
+   struct tgsi_texture_offset insn_texture_offset;
    struct tgsi_src_register src;
    struct tgsi_dimension dim;
    struct tgsi_dst_register dst;
@@ -997,7 +998,7 @@ ureg_fixup_label(struct ureg_program *ureg,
 void
 ureg_emit_texture(struct ureg_program *ureg,
                   unsigned extended_token,
-                  unsigned target )
+                  unsigned target, unsigned num_offsets)
 {
    union tgsi_any_token *out, *insn;
 
@@ -1008,6 +1009,20 @@ ureg_emit_texture(struct ureg_program *ureg,
 
    out[0].value = 0;
    out[0].insn_texture.Texture = target;
+   out[0].insn_texture.NumOffsets = num_offsets;
+}
+
+void
+ureg_emit_texture_offset(struct ureg_program *ureg,
+                         const struct tgsi_texture_offset *offset)
+{
+   union tgsi_any_token *out;
+
+   out = get_tokens( ureg, DOMAIN_INSN, 1);
+
+   out[0].value = 0;
+   out[0].insn_texture_offset = *offset;
+   
 }
 
 
@@ -1074,6 +1089,8 @@ ureg_tex_insn(struct ureg_program *ureg,
               const struct ureg_dst *dst,
               unsigned nr_dst,
               unsigned target,
+              const struct tgsi_texture_offset *texoffsets,
+              unsigned nr_offset,
               const struct ureg_src *src,
               unsigned nr_src )
 {
@@ -1106,7 +1123,10 @@ ureg_tex_insn(struct ureg_program *ureg,
                          nr_dst,
                          nr_src);
 
-   ureg_emit_texture( ureg, insn.extended_token, target );
+   ureg_emit_texture( ureg, insn.extended_token, target, nr_offset );
+
+   for (i = 0; i < nr_offset; i++)
+      ureg_emit_texture_offset( ureg, &texoffsets[i]);
 
    for (i = 0; i < nr_dst; i++)
       ureg_emit_dst( ureg, dst[i] );
diff --git a/src/gallium/auxiliary/tgsi/tgsi_ureg.h b/src/gallium/auxiliary/tgsi/tgsi_ureg.h
index e3a4915..8f5f22e 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_ureg.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_ureg.h
@@ -451,6 +451,8 @@ ureg_tex_insn(struct ureg_program *ureg,
               const struct ureg_dst *dst,
               unsigned nr_dst,
               unsigned target,
+              const struct tgsi_texture_offset *texoffsets,
+              unsigned nr_offset,
               const struct ureg_src *src,
               unsigned nr_src );
 
@@ -493,7 +495,11 @@ ureg_emit_label(struct ureg_program *ureg,
 void
 ureg_emit_texture(struct ureg_program *ureg,
                   unsigned insn_token,
-                  unsigned target );
+                  unsigned target, unsigned num_offsets);
+
+void
+ureg_emit_texture_offset(struct ureg_program *ureg,
+                         const struct tgsi_texture_offset *offset);
 
 void 
 ureg_emit_dst( struct ureg_program *ureg,
@@ -677,7 +683,7 @@ static INLINE void ureg_##op( struct ureg_program *ureg,                \
                          dst.PredSwizzleW,                              \
                          1,                                             \
                          2);                                            \
-   ureg_emit_texture( ureg, insn.extended_token, target );              \
+   ureg_emit_texture( ureg, insn.extended_token, target, 0 );		\
    ureg_emit_dst( ureg, dst );                                          \
    ureg_emit_src( ureg, src0 );                                         \
    ureg_emit_src( ureg, src1 );                                         \
@@ -732,7 +738,7 @@ static INLINE void ureg_##op( struct ureg_program *ureg,                \
                          dst.PredSwizzleW,                              \
                          1,                                             \
                          4);                                            \
-   ureg_emit_texture( ureg, insn.extended_token, target );              \
+   ureg_emit_texture( ureg, insn.extended_token, target, 0 );		\
    ureg_emit_dst( ureg, dst );                                          \
    ureg_emit_src( ureg, src0 );                                         \
    ureg_emit_src( ureg, src1 );                                         \