X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fdrivers%2Fsvga%2Fsvga_tgsi.c;h=7cbd5166900e70fbe5911b1f42b70688ae72a1e3;hb=b1772651b728fd231b0c5885415b35f43a307c7f;hp=81eea1a145aff7c99e89f4ded22cbd4c7c08718b;hpb=178407f33c413cbe7434597b2129abde90041b6b;p=mesa.git

diff --git a/src/gallium/drivers/svga/svga_tgsi.c b/src/gallium/drivers/svga/svga_tgsi.c
index 81eea1a145a..7cbd5166900 100644
--- a/src/gallium/drivers/svga/svga_tgsi.c
+++ b/src/gallium/drivers/svga/svga_tgsi.c
@@ -30,11 +30,14 @@
 #include "tgsi/tgsi_parse.h"
 #include "tgsi/tgsi_dump.h"
 #include "tgsi/tgsi_scan.h"
+#include "util/u_math.h"
 #include "util/u_memory.h"
+#include "util/u_bitmask.h"
 
 #include "svgadump/svga_shader_dump.h"
 
 #include "svga_context.h"
+#include "svga_shader.h"
 #include "svga_tgsi.h"
 #include "svga_tgsi_emit.h"
 #include "svga_debug.h"
@@ -47,26 +50,19 @@
  */
 static char err_buf[128];
 
-#if 0
-static void svga_destroy_shader_emitter( struct svga_shader_emitter *emit )
-{
-   if (emit->buf != err_buf)
-      FREE(emit->buf);
-}
-#endif
-
 
-static boolean svga_shader_expand( struct svga_shader_emitter *emit )
+static boolean
+svga_shader_expand(struct svga_shader_emitter *emit)
 {
    char *new_buf;
    unsigned newsize = emit->size * 2;
 
-   if(emit->buf != err_buf)
+   if (emit->buf != err_buf)
       new_buf = REALLOC(emit->buf, emit->size, newsize);
    else
       new_buf = NULL;
 
-   if (new_buf == NULL) {
+   if (!new_buf) {
       emit->ptr = err_buf;
       emit->buf = err_buf;
       emit->size = sizeof(err_buf);
@@ -77,190 +73,196 @@ static boolean svga_shader_expand( struct svga_shader_emitter *emit )
    emit->ptr = new_buf + (emit->ptr - emit->buf);
    emit->buf = new_buf;
    return TRUE;
-}   
+}
+
 
-static INLINE boolean reserve(  struct svga_shader_emitter *emit,
-                                unsigned nr_dwords )
+static inline boolean
+reserve(struct svga_shader_emitter *emit, unsigned nr_dwords)
 {
    if (emit->ptr - emit->buf + nr_dwords * sizeof(unsigned) >= emit->size) {
-      if (!svga_shader_expand( emit ))
+      if (!svga_shader_expand(emit)) {
          return FALSE;
+      }
    }
 
    return TRUE;
 }
 
-boolean svga_shader_emit_dword( struct svga_shader_emitter *emit,
-                                unsigned dword )
+
+boolean
+svga_shader_emit_dword(struct svga_shader_emitter * emit, unsigned dword)
 {
    if (!reserve(emit, 1))
       return FALSE;
 
-   *(unsigned *)emit->ptr = dword;
+   *(unsigned *) emit->ptr = dword;
    emit->ptr += sizeof dword;
    return TRUE;
 }
 
-boolean svga_shader_emit_dwords( struct svga_shader_emitter *emit,
-                                 const unsigned *dwords,
-                                 unsigned nr )
+
+boolean
+svga_shader_emit_dwords(struct svga_shader_emitter * emit,
+                        const unsigned *dwords, unsigned nr)
 {
    if (!reserve(emit, nr))
       return FALSE;
 
-   memcpy( emit->ptr, dwords, nr * sizeof *dwords );
+   memcpy(emit->ptr, dwords, nr * sizeof *dwords);
    emit->ptr += nr * sizeof *dwords;
    return TRUE;
 }
 
-boolean svga_shader_emit_opcode( struct svga_shader_emitter *emit,
-                                 unsigned opcode )
+
+boolean
+svga_shader_emit_opcode(struct svga_shader_emitter * emit, unsigned opcode)
 {
    SVGA3dShaderInstToken *here;
 
    if (!reserve(emit, 1))
       return FALSE;
 
-   here = (SVGA3dShaderInstToken *)emit->ptr;
+   here = (SVGA3dShaderInstToken *) emit->ptr;
    here->value = opcode;
 
    if (emit->insn_offset) {
-      SVGA3dShaderInstToken *prev = (SVGA3dShaderInstToken *)(emit->buf + 
-                                                              emit->insn_offset);
+      SVGA3dShaderInstToken *prev =
+         (SVGA3dShaderInstToken *) (emit->buf + emit->insn_offset);
       prev->size = (here - prev) - 1;
    }
-   
+
    emit->insn_offset = emit->ptr - emit->buf;
    emit->ptr += sizeof(unsigned);
    return TRUE;
 }
 
-#define SVGA3D_PS_2X (SVGA3D_PS_20 | 1)
-#define SVGA3D_VS_2X (SVGA3D_VS_20 | 1)
 
-static boolean svga_shader_emit_header( struct svga_shader_emitter *emit )
+static boolean
+svga_shader_emit_header(struct svga_shader_emitter *emit)
 {
    SVGA3dShaderVersion header;
 
-   memset( &header, 0, sizeof header );
+   memset(&header, 0, sizeof header);
 
    switch (emit->unit) {
    case PIPE_SHADER_FRAGMENT:
-      header.value = emit->use_sm30 ? SVGA3D_PS_30 : SVGA3D_PS_2X;
+      header.value = SVGA3D_PS_30;
       break;
    case PIPE_SHADER_VERTEX:
-      header.value = emit->use_sm30 ? SVGA3D_VS_30 : SVGA3D_VS_2X;
+      header.value = SVGA3D_VS_30;
       break;
    }
- 
-   return svga_shader_emit_dword( emit, header.value );
-}
-
-
 
+   return svga_shader_emit_dword(emit, header.value);
+}
 
 
-/* Parse TGSI shader and translate to SVGA/DX9 serialized
- * representation.  
+/**
+ * Parse TGSI shader and translate to SVGA/DX9 serialized
+ * representation.
  *
  * In this function SVGA shader is emitted to an in-memory buffer that
  * can be dynamically grown.  Once we've finished and know how large
  * it is, it will be copied to a hardware buffer for upload.
  */
-static struct svga_shader_result *
-svga_tgsi_translate( const struct svga_shader *shader,
-                     union svga_compile_key key,
-                     unsigned unit )
+struct svga_shader_variant *
+svga_tgsi_vgpu9_translate(struct svga_context *svga,
+                          const struct svga_shader *shader,
+                          const struct svga_compile_key *key, unsigned unit)
 {
-   struct svga_shader_result *result = NULL;
+   struct svga_shader_variant *variant = NULL;
    struct svga_shader_emitter emit;
-   int ret = 0;
+
+   SVGA_STATS_TIME_PUSH(svga_sws(svga), SVGA_STATS_TIME_TGSIVGPU9TRANSLATE);
 
    memset(&emit, 0, sizeof(emit));
 
-   emit.use_sm30 = shader->use_sm30;
    emit.size = 1024;
    emit.buf = MALLOC(emit.size);
    if (emit.buf == NULL) {
-      ret = PIPE_ERROR_OUT_OF_MEMORY;
       goto fail;
    }
 
    emit.ptr = emit.buf;
    emit.unit = unit;
-   emit.key = key;
+   emit.key = *key;
 
-   tgsi_scan_shader( shader->tokens, &emit.info);
+   tgsi_scan_shader(shader->tokens, &emit.info);
 
    emit.imm_start = emit.info.file_max[TGSI_FILE_CONSTANT] + 1;
-   
+
    if (unit == PIPE_SHADER_FRAGMENT)
-      emit.imm_start += key.fkey.num_unnormalized_coords;
+      emit.imm_start += key->num_unnormalized_coords;
 
    if (unit == PIPE_SHADER_VERTEX) {
-      emit.imm_start += key.vkey.need_prescale ? 2 : 0;
-      emit.imm_start += key.vkey.num_zero_stride_vertex_elements;
+      emit.imm_start += key->vs.need_prescale ? 2 : 0;
    }
 
-   emit.nr_hw_const = (emit.imm_start + emit.info.file_max[TGSI_FILE_IMMEDIATE] + 1);
+   emit.nr_hw_float_const =
+      (emit.imm_start + emit.info.file_max[TGSI_FILE_IMMEDIATE] + 1);
 
    emit.nr_hw_temp = emit.info.file_max[TGSI_FILE_TEMPORARY] + 1;
-   emit.in_main_func = TRUE;
-
-   if (!svga_shader_emit_header( &emit ))
-      goto fail;
 
-   if (!svga_shader_emit_instructions( &emit, shader->tokens ))
+   if (emit.nr_hw_temp >= SVGA3D_TEMPREG_MAX) {
+      debug_printf("svga: too many temporary registers (%u)\n",
+                   emit.nr_hw_temp);
       goto fail;
-   
-   result = CALLOC_STRUCT(svga_shader_result);
-   if (result == NULL)
-      goto fail;
-
-   result->shader = shader;
-   result->tokens = (const unsigned *)emit.buf;
-   result->nr_tokens = (emit.ptr - emit.buf) / sizeof(unsigned);
-   memcpy(&result->key, &key, sizeof key);
-
-   return result;
+   }
 
-fail:
-   FREE(result);
-   FREE(emit.buf);
-   return NULL;
-}
+   emit.in_main_func = TRUE;
 
+   if (!svga_shader_emit_header(&emit)) {
+      debug_printf("svga: emit header failed\n");
+      goto fail;
+   }
 
+   if (!svga_shader_emit_instructions(&emit, shader->tokens)) {
+      debug_printf("svga: emit instructions failed\n");
+      goto fail;
+   }
 
+   variant = svga_new_shader_variant(svga);
+   if (!variant)
+      goto fail;
 
-struct svga_shader_result *
-svga_translate_fragment_program( const struct svga_fragment_shader *fs,
-                                 const struct svga_fs_compile_key *fkey )
-{
-   union svga_compile_key key;
-   memcpy(&key.fkey, fkey, sizeof *fkey);
+   variant->shader = shader;
+   variant->tokens = (const unsigned *) emit.buf;
+   variant->nr_tokens = (emit.ptr - emit.buf) / sizeof(unsigned);
+   memcpy(&variant->key, key, sizeof(*key));
+   variant->id = UTIL_BITMASK_INVALID_INDEX;
 
-   return svga_tgsi_translate( &fs->base, 
-                               key,
-                               PIPE_SHADER_FRAGMENT );
-}
+   variant->pstipple_sampler_unit = emit.pstipple_sampler_unit;
 
-struct svga_shader_result *
-svga_translate_vertex_program( const struct svga_vertex_shader *vs,
-                               const struct svga_vs_compile_key *vkey )
-{
-   union svga_compile_key key;
-   memcpy(&key.vkey, vkey, sizeof *vkey);
+   /* If there was exactly one write to a fragment shader output register
+    * and it came from a constant buffer, we know all fragments will have
+    * the same color (except for blending).
+    */
+   variant->constant_color_output =
+      emit.constant_color_output && emit.num_output_writes == 1;
 
-   return svga_tgsi_translate( &vs->base, 
-                               key,
-                               PIPE_SHADER_VERTEX );
-}
+#if 0
+   if (!svga_shader_verify(variant->tokens, variant->nr_tokens) ||
+       SVGA_DEBUG & DEBUG_TGSI) {
+      debug_printf("#####################################\n");
+      debug_printf("Shader %u below\n", shader->id);
+      tgsi_dump(shader->tokens, 0);
+      if (SVGA_DEBUG & DEBUG_TGSI) {
+         debug_printf("Shader %u compiled below\n", shader->id);
+         svga_shader_dump(variant->tokens, variant->nr_tokens, FALSE);
+      }
+      debug_printf("#####################################\n");
+   }
+#endif
 
+   goto done;
 
-void svga_destroy_shader_result( struct svga_shader_result *result )
-{
-   FREE((unsigned *)result->tokens);
-   FREE(result);
+fail:
+   FREE(variant);
+   if (emit.buf != err_buf)
+      FREE(emit.buf);
+   variant = NULL;
+
+done:
+   SVGA_STATS_TIME_POP(svga_sws(svga));
+   return variant;
 }
-