i965/fs: Disable CSE optimization for untyped & typed surface reads

[mesa.git] / src / mesa / drivers / dri / i965 / gen6_multisample_state.c

diff --git a/src/mesa/drivers/dri/i965/gen6_multisample_state.c b/src/mesa/drivers/dri/i965/gen6_multisample_state.c
index fd3dd0e53637bb221c0a2992d59922703b679e77..8444c0c9bae581f5da54a6a94cbc94ea171ebe26 100644 (file)
--- a/src/mesa/drivers/dri/i965/gen6_multisample_state.c
+++ b/src/mesa/drivers/dri/i965/gen6_multisample_state.c
@@ -26,6 +26,7 @@
 #include "brw_context.h"
 #include "brw_defines.h"
 #include "brw_multisample_state.h"
+#include "main/framebuffer.h"
 
 void
 gen6_get_sample_position(struct gl_context *ctx,
@@ -34,19 +35,21 @@ gen6_get_sample_position(struct gl_context *ctx,
 {
    uint8_t bits;
 
-   switch (fb->Visual.samples) {
+   switch (_mesa_geometric_samples(fb)) {
    case 1:
       result[0] = result[1] = 0.5f;
       return;
+   case 2:
+      bits = brw_multisample_positions_1x_2x >> (8 * index);
+      break;
    case 4:
-      bits = brw_multisample_positions_4x[0] >> (8 * index);
+      bits = brw_multisample_positions_4x >> (8 * index);
       break;
    case 8:
       bits = brw_multisample_positions_8x[index >> 2] >> (8 * (index & 3));
       break;
    default:
-      assert(!"Not implemented");
-      return;
+      unreachable("Not implemented");
    }
 
    /* Convert from U0.4 back to a floating point coordinate. */
@@ -54,6 +57,51 @@ gen6_get_sample_position(struct gl_context *ctx,
    result[1] = (bits & 0xf) / 16.0f;
 }
 
+/**
+ * Sample index layout shows the numbering of slots in a rectangular
+ * grid of samples with in a pixel. Sample number layout shows the
+ * rectangular grid of samples roughly corresponding to the real sample
+ * locations with in a pixel. Sample number layout matches the sample
+ * index layout in case of 2X and 4x MSAA, but they are different in
+ * case of 8X MSAA.
+ *
+ * 2X MSAA sample index / number layout
+ *           ---------
+ *           | 0 | 1 |
+ *           ---------
+ *
+ * 4X MSAA sample index / number layout
+ *           ---------
+ *           | 0 | 1 |
+ *           ---------
+ *           | 2 | 3 |
+ *           ---------
+ *
+ * 8X MSAA sample index layout    8x MSAA sample number layout
+ *           ---------                      ---------
+ *           | 0 | 1 |                      | 5 | 2 |
+ *           ---------                      ---------
+ *           | 2 | 3 |                      | 4 | 6 |
+ *           ---------                      ---------
+ *           | 4 | 5 |                      | 0 | 3 |
+ *           ---------                      ---------
+ *           | 6 | 7 |                      | 7 | 1 |
+ *           ---------                      ---------
+ *
+ * A sample map is used to map sample indices to sample numbers.
+ */
+void
+gen6_set_sample_maps(struct gl_context *ctx)
+{
+   uint8_t map_2x[2] = {0, 1};
+   uint8_t map_4x[4] = {0, 1, 2, 3};
+   uint8_t map_8x[8] = {5, 2, 4, 6, 0, 3, 7, 1};
+
+   memcpy(ctx->Const.SampleMap2x, map_2x, sizeof(map_2x));
+   memcpy(ctx->Const.SampleMap4x, map_4x, sizeof(map_4x));
+   memcpy(ctx->Const.SampleMap8x, map_8x, sizeof(map_8x));
+}
+
 /**
  * 3DSTATE_MULTISAMPLE
  */
@@ -74,7 +122,7 @@ gen6_emit_3dstate_multisample(struct brw_context *brw,
       break;
    case 4:
       number_of_multisamples = MS_NUMSAMPLES_4;
-      sample_positions_3210 = brw_multisample_positions_4x[0];
+      sample_positions_3210 = brw_multisample_positions_4x;
       break;
    case 8:
       number_of_multisamples = MS_NUMSAMPLES_8;
@@ -82,13 +130,9 @@ gen6_emit_3dstate_multisample(struct brw_context *brw,
       sample_positions_7654 = brw_multisample_positions_8x[1];
       break;
    default:
-      assert(!"Unrecognized num_samples in gen6_emit_3dstate_multisample");
-      break;
+      unreachable("Unrecognized num_samples in gen6_emit_3dstate_multisample");
    }
 
-   /* 3DSTATE_MULTISAMPLE is nonpipelined. */
-   intel_emit_post_sync_nonzero_flush(brw);
-
    int len = brw->gen >= 7 ? 4 : 3;
    BEGIN_BATCH(len);
    OUT_BATCH(_3DSTATE_MULTISAMPLE << 16 | (len - 2));
@@ -99,16 +143,16 @@ gen6_emit_3dstate_multisample(struct brw_context *brw,
    ADVANCE_BATCH();
 }
 
-
 unsigned
 gen6_determine_sample_mask(struct brw_context *brw)
 {
    struct gl_context *ctx = &brw->ctx;
-   float coverage = 1.0;
+   float coverage = 1.0f;
    float coverage_invert = false;
    unsigned sample_mask = ~0u;
 
-   unsigned num_samples = ctx->DrawBuffer->Visual.samples;
+   /* BRW_NEW_NUM_SAMPLES */
+   unsigned num_samples = brw->num_samples;
 
    if (ctx->Multisample._Enabled) {
       if (ctx->Multisample.SampleCoverage) {
@@ -121,7 +165,7 @@ gen6_determine_sample_mask(struct brw_context *brw)
    }
 
    if (num_samples > 1) {
-      int coverage_int = (int) (num_samples * coverage + 0.5);
+      int coverage_int = (int) (num_samples * coverage + 0.5f);
       uint32_t coverage_bits = (1 << coverage_int) - 1;
       if (coverage_invert)
          coverage_bits ^= (1 << num_samples) - 1;
@@ -131,7 +175,6 @@ gen6_determine_sample_mask(struct brw_context *brw)
    }
 }
 
-
 /**
  * 3DSTATE_SAMPLE_MASK
  */
@@ -144,25 +187,19 @@ gen6_emit_3dstate_sample_mask(struct brw_context *brw, unsigned mask)
    ADVANCE_BATCH();
 }
 
-
-static void upload_multisample_state(struct brw_context *brw)
+static void
+upload_multisample_state(struct brw_context *brw)
 {
-   struct gl_context *ctx = &brw->ctx;
-
-   /* _NEW_BUFFERS, _NEW_MULTISAMPLE */
-   unsigned num_samples = ctx->DrawBuffer->Visual.samples;
-
-   gen6_emit_3dstate_multisample(brw, num_samples);
+   /* BRW_NEW_NUM_SAMPLES */
+   gen6_emit_3dstate_multisample(brw, brw->num_samples);
    gen6_emit_3dstate_sample_mask(brw, gen6_determine_sample_mask(brw));
 }
 
-
 const struct brw_tracked_state gen6_multisample_state = {
    .dirty = {
-      .mesa = _NEW_BUFFERS |
-              _NEW_MULTISAMPLE,
-      .brw = BRW_NEW_CONTEXT,
-      .cache = 0
+      .mesa = _NEW_MULTISAMPLE,
+      .brw = BRW_NEW_CONTEXT |
+             BRW_NEW_NUM_SAMPLES,
    },
    .emit = upload_multisample_state
 };