From: Emma Anholt <emma@anholt.net>
Date: Sun, 27 Jun 2021 04:31:00 +0000 (-0700)
Subject: i915g: Bake the decls and program together.
X-Git-Tag: upstream/21.2.3~1193
X-Git-Url: http://review.tizen.org/git/?a=commitdiff_plain;h=79800a957d532febdb7ba307c424923fde2629b4;p=platform%2Fupstream%2Fmesa.git

i915g: Bake the decls and program together.

Simplifies program upload a bunch, and will let us disasm the program
independently of the whole cmd buffer.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11617>
---

diff --git a/src/gallium/drivers/i915/i915_context.h b/src/gallium/drivers/i915/i915_context.h
index 6c158b7..385e9ca 100644
--- a/src/gallium/drivers/i915/i915_context.h
+++ b/src/gallium/drivers/i915/i915_context.h
@@ -97,9 +97,6 @@ struct i915_fragment_shader {
 
    struct draw_fragment_shader *draw_data;
 
-   uint *decl;
-   uint decl_len;
-
    uint *program;
    uint program_len;
 
diff --git a/src/gallium/drivers/i915/i915_fpc_translate.c b/src/gallium/drivers/i915/i915_fpc_translate.c
index 361d870..1fc5ab1 100644
--- a/src/gallium/drivers/i915/i915_fpc_translate.c
+++ b/src/gallium/drivers/i915/i915_fpc_translate.c
@@ -49,11 +49,8 @@
  * Simple pass-through fragment shader to use when we don't have
  * a real shader (or it fails to compile for some reason).
  */
-static unsigned passthrough_decl[] = {
-   _3DSTATE_PIXEL_SHADER_PROGRAM | ((1 * 3) - 1),
-};
-
 static unsigned passthrough_program[] = {
+   _3DSTATE_PIXEL_SHADER_PROGRAM | ((1 * 3) - 1),
    /* move to output color:
     */
    (A0_MOV | (REG_TYPE_OC << A0_DEST_TYPE_SHIFT) | A0_DEST_CHANNEL_ALL |
@@ -98,12 +95,9 @@ static void
 i915_use_passthrough_shader(struct i915_fragment_shader *fs)
 {
    fs->program = (uint *)MALLOC(sizeof(passthrough_program));
-   fs->decl = (uint *)MALLOC(sizeof(passthrough_decl));
    if (fs->program) {
       memcpy(fs->program, passthrough_program, sizeof(passthrough_program));
-      memcpy(fs->decl, passthrough_decl, sizeof(passthrough_decl));
       fs->program_len = ARRAY_SIZE(passthrough_program);
-      fs->decl_len = ARRAY_SIZE(passthrough_decl);
    }
    fs->num_constants = 0;
 }
@@ -1008,23 +1002,12 @@ i915_fini_compile(struct i915_context *i915, struct i915_fp_compile *p)
 
       /* Copy compilation results to fragment program struct:
        */
-      assert(!ifs->decl);
       assert(!ifs->program);
 
-      ifs->decl = (uint *)MALLOC(decl_size * sizeof(uint));
-      ifs->program = (uint *)MALLOC(program_size * sizeof(uint));
-
-      if (ifs->decl) {
-         ifs->decl_len = decl_size;
-
-         memcpy(ifs->decl, p->declarations, decl_size * sizeof(uint));
-      }
-
-      if (ifs->program) {
-         ifs->program_len = program_size;
-
-         memcpy(ifs->program, p->program, program_size * sizeof(uint));
-      }
+      ifs->program_len = decl_size + program_size;
+      ifs->program = (uint *)MALLOC(ifs->program_len * sizeof(uint));
+      memcpy(ifs->program, p->declarations, decl_size * sizeof(uint));
+      memcpy(&ifs->program[decl_size], p->program, program_size * sizeof(uint));
    }
 
    /* Release the compilation struct:
diff --git a/src/gallium/drivers/i915/i915_state.c b/src/gallium/drivers/i915/i915_state.c
index 93b85b0..e31b96f 100644
--- a/src/gallium/drivers/i915/i915_state.c
+++ b/src/gallium/drivers/i915/i915_state.c
@@ -555,16 +555,12 @@ i915_delete_fs_state(struct pipe_context *pipe, void *shader)
 {
    struct i915_fragment_shader *ifs = (struct i915_fragment_shader *)shader;
 
-   FREE(ifs->decl);
-   ifs->decl = NULL;
-
    FREE(ifs->program);
    ifs->program = NULL;
    FREE((struct tgsi_token *)ifs->state.tokens);
    ifs->state.tokens = NULL;
 
    ifs->program_len = 0;
-   ifs->decl_len = 0;
 
    FREE(ifs);
 }
diff --git a/src/gallium/drivers/i915/i915_state_emit.c b/src/gallium/drivers/i915/i915_state_emit.c
index 8f971f5e..cc55a2c 100644
--- a/src/gallium/drivers/i915/i915_state_emit.c
+++ b/src/gallium/drivers/i915/i915_state_emit.c
@@ -368,45 +368,29 @@ emit_constants(struct i915_context *i915)
 static void
 validate_program(struct i915_context *i915, unsigned *batch_space)
 {
-   uint additional_size = 0;
-
-   additional_size += i915->current.fixup_swizzle ? 3 : 0;
-
    /* we need more batch space if we want to emulate rgba framebuffers */
-   *batch_space = i915->fs->decl_len + i915->fs->program_len + additional_size;
+   *batch_space = i915->fs->program_len + (i915->current.fixup_swizzle ? 3 : 0);
 }
 
 static void
 emit_program(struct i915_context *i915)
 {
-   uint additional_size = 0;
-   uint i;
-
-   /* count how much additional space we'll need */
-   validate_program(i915, &additional_size);
-   additional_size -= i915->fs->decl_len + i915->fs->program_len;
-
    /* we should always have, at least, a pass-through program */
    assert(i915->fs->program_len > 0);
 
-   /* output the declarations */
-   {
-      /* first word has the size, we have to adjust that */
-      uint size = (i915->fs->decl[0]);
-      size += additional_size;
-      OUT_BATCH(size);
-   }
+   /* If we're doing a fixup swizzle, that's 3 more dwords to add. */
+   uint32_t additional_size = 0;
+   if (i915->current.fixup_swizzle)
+      additional_size = 3;
+
+   /* output the program: 1 dword of header, then 3 dwords per decl/instruction */
+   assert(i915->fs->program_len % 3 == 1);
 
-   for (i = 1; i < i915->fs->decl_len; i++)
-      OUT_BATCH(i915->fs->decl[i]);
+   /* first word has the size, adjust it for fixup swizzle */
+   OUT_BATCH(i915->fs->program[0] + additional_size);
 
-   /* output the program */
-   assert(i915->fs->program_len % 3 == 0);
-   for (i = 0; i < i915->fs->program_len; i += 3) {
+   for (int i = 1; i < i915->fs->program_len; i++)
       OUT_BATCH(i915->fs->program[i]);
-      OUT_BATCH(i915->fs->program[i + 1]);
-      OUT_BATCH(i915->fs->program[i + 2]);
-   }
 
    /* we emit an additional mov with swizzle to fake RGBA framebuffers */
    if (i915->current.fixup_swizzle) {