i965/fs: Add support for translating ir_triop_fma into MAD.

[mesa.git] / src / mesa / drivers / dri / i965 / gen7_vs_state.c

diff --git a/src/mesa/drivers/dri/i965/gen7_vs_state.c b/src/mesa/drivers/dri/i965/gen7_vs_state.c
index 6a7add8e562f7bd845496d2c43939931e92aa59a..ce584f675f6e33ca7a8e773c266c568478bb1aff 100644 (file)
--- a/src/mesa/drivers/dri/i965/gen7_vs_state.c
+++ b/src/mesa/drivers/dri/i965/gen7_vs_state.c
@@ -32,14 +32,25 @@
 static void
 upload_vs_state(struct brw_context *brw)
 {
-   struct intel_context *intel = &brw->intel;
-   struct gl_context *ctx = &intel->ctx;
+   struct gl_context *ctx = &brw->ctx;
+   uint32_t floating_point_mode = 0;
+   const int max_threads_shift = brw->is_haswell ?
+      HSW_VS_MAX_THREADS_SHIFT : GEN6_VS_MAX_THREADS_SHIFT;
 
+   gen7_emit_vs_workaround_flush(brw);
+
+   /* BRW_NEW_VS_BINDING_TABLE */
    BEGIN_BATCH(2);
    OUT_BATCH(_3DSTATE_BINDING_TABLE_POINTERS_VS << 16 | (2 - 2));
    OUT_BATCH(brw->vs.bind_bo_offset);
    ADVANCE_BATCH();
 
+   /* CACHE_NEW_SAMPLER */
+   BEGIN_BATCH(2);
+   OUT_BATCH(_3DSTATE_SAMPLER_STATE_POINTERS_VS << 16 | (2 - 2));
+   OUT_BATCH(brw->vs.sampler_offset);
+   ADVANCE_BATCH();
+
    if (brw->vs.push_const_size == 0) {
       /* Disable the push constant buffers. */
       BEGIN_BATCH(7);
@@ -59,25 +70,39 @@ upload_vs_state(struct brw_context *brw)
       /* Pointer to the VS constant buffer.  Covered by the set of
        * state flags from gen6_prepare_wm_contants
        */
-      OUT_BATCH(brw->vs.push_const_offset);
+      OUT_BATCH(brw->vs.push_const_offset | GEN7_MOCS_L3);
       OUT_BATCH(0);
       OUT_BATCH(0);
       OUT_BATCH(0);
       ADVANCE_BATCH();
    }
 
+   /* Use ALT floating point mode for ARB vertex programs, because they
+    * require 0^0 == 1.
+    */
+   if (ctx->Shader.CurrentVertexProgram == NULL)
+      floating_point_mode = GEN6_VS_FLOATING_POINT_MODE_ALT;
+
    BEGIN_BATCH(6);
    OUT_BATCH(_3DSTATE_VS << 16 | (6 - 2));
-   OUT_RELOC(brw->vs.prog_bo, I915_GEM_DOMAIN_INSTRUCTION, 0, 0);
-   OUT_BATCH((0 << GEN6_VS_SAMPLER_COUNT_SHIFT) |
-            GEN6_VS_FLOATING_POINT_MODE_ALT |
-            (brw->vs.nr_surfaces << GEN6_VS_BINDING_TABLE_ENTRY_COUNT_SHIFT));
-   OUT_BATCH(0); /* scratch space base offset */
-   OUT_BATCH((1 << GEN6_VS_DISPATCH_START_GRF_SHIFT) |
-            (brw->vs.prog_data->urb_read_length << GEN6_VS_URB_READ_LENGTH_SHIFT) |
+   OUT_BATCH(brw->vs.prog_offset);
+   OUT_BATCH(floating_point_mode |
+            ((ALIGN(brw->vs.sampler_count, 4)/4) << GEN6_VS_SAMPLER_COUNT_SHIFT));
+
+   if (brw->vs.prog_data->base.total_scratch) {
+      OUT_RELOC(brw->vs.scratch_bo,
+               I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
+               ffs(brw->vs.prog_data->base.total_scratch) - 11);
+   } else {
+      OUT_BATCH(0);
+   }
+
+   OUT_BATCH((brw->vs.prog_data->base.dispatch_grf_start_reg <<
+              GEN6_VS_DISPATCH_START_GRF_SHIFT) |
+            (brw->vs.prog_data->base.urb_read_length << GEN6_VS_URB_READ_LENGTH_SHIFT) |
             (0 << GEN6_VS_URB_ENTRY_READ_OFFSET_SHIFT));
 
-   OUT_BATCH(((brw->vs_max_threads - 1) << GEN6_VS_MAX_THREADS_SHIFT) |
+   OUT_BATCH(((brw->max_vs_threads - 1) << max_threads_shift) |
             GEN6_VS_STATISTICS_ENABLE |
             GEN6_VS_ENABLE);
    ADVANCE_BATCH();
@@ -86,14 +111,11 @@ upload_vs_state(struct brw_context *brw)
 const struct brw_tracked_state gen7_vs_state = {
    .dirty = {
       .mesa  = _NEW_TRANSFORM | _NEW_PROGRAM_CONSTANTS,
-      .brw   = (BRW_NEW_CURBE_OFFSETS |
-                BRW_NEW_NR_VS_SURFACES |
-               BRW_NEW_URB_FENCE |
-               BRW_NEW_CONTEXT |
+      .brw   = (BRW_NEW_CONTEXT |
                BRW_NEW_VERTEX_PROGRAM |
                BRW_NEW_VS_BINDING_TABLE |
                BRW_NEW_BATCH),
-      .cache = CACHE_NEW_VS_PROG
+      .cache = CACHE_NEW_VS_PROG | CACHE_NEW_SAMPLER
    },
    .emit = upload_vs_state,
 };