From a664233042e1ad343184a0c237c3bd7ac5010779 Mon Sep 17 00:00:00 2001
From: Eric Anholt <eric@anholt.net>
Date: Fri, 20 Nov 2015 17:11:18 -0800
Subject: [PATCH] vc4: Add support for loading sample mask.

---
 src/gallium/drivers/vc4/kernel/vc4_packet.h | 3 +++
 src/gallium/drivers/vc4/vc4_context.c       | 2 ++
 src/gallium/drivers/vc4/vc4_program.c       | 4 ++++
 src/gallium/drivers/vc4/vc4_qir.h           | 1 +
 src/gallium/drivers/vc4/vc4_state.c         | 2 +-
 src/gallium/drivers/vc4/vc4_uniforms.c      | 8 ++++++++
 6 files changed, 19 insertions(+), 1 deletion(-)

diff --git a/src/gallium/drivers/vc4/kernel/vc4_packet.h b/src/gallium/drivers/vc4/kernel/vc4_packet.h
index 771e2b78761..9c12d331f61 100644
--- a/src/gallium/drivers/vc4/kernel/vc4_packet.h
+++ b/src/gallium/drivers/vc4/kernel/vc4_packet.h
@@ -121,6 +121,9 @@ enum vc4_packet {
 #define VC4_PACKET_TILE_COORDINATES_SIZE				3
 #define VC4_PACKET_GEM_HANDLES_SIZE					9
 
+/* Number of multisamples supported. */
+#define VC4_MAX_SAMPLES							4
+
 #define VC4_MASK(high, low) (((1 << ((high) - (low) + 1)) - 1) << (low))
 /* Using the GNU statement expression extension */
 #define VC4_SET_FIELD(value, field)                                       \
diff --git a/src/gallium/drivers/vc4/vc4_context.c b/src/gallium/drivers/vc4/vc4_context.c
index d490fad5167..4f3e2263f56 100644
--- a/src/gallium/drivers/vc4/vc4_context.c
+++ b/src/gallium/drivers/vc4/vc4_context.c
@@ -245,6 +245,8 @@ vc4_context_create(struct pipe_screen *pscreen, void *priv, unsigned flags)
 
         vc4_debug |= saved_shaderdb_flag;
 
+        vc4->sample_mask = (1 << VC4_MAX_SAMPLES) - 1;
+
         return &vc4->base;
 
 fail:
diff --git a/src/gallium/drivers/vc4/vc4_program.c b/src/gallium/drivers/vc4/vc4_program.c
index 197577b6c20..9c6e8647256 100644
--- a/src/gallium/drivers/vc4/vc4_program.c
+++ b/src/gallium/drivers/vc4/vc4_program.c
@@ -1462,6 +1462,10 @@ ntq_emit_intrinsic(struct vc4_compile *c, nir_intrinsic_instr *instr)
                                     instr->const_index[0]);
                 break;
 
+        case nir_intrinsic_load_sample_mask_in:
+                *dest = qir_uniform(c, QUNIFORM_SAMPLE_MASK, 0);
+                break;
+
         case nir_intrinsic_load_input:
                 assert(instr->num_components == 1);
                 if (instr->const_index[0] == VC4_NIR_TLB_COLOR_READ_INPUT) {
diff --git a/src/gallium/drivers/vc4/vc4_qir.h b/src/gallium/drivers/vc4/vc4_qir.h
index ddb35e41fcf..ad243ec1113 100644
--- a/src/gallium/drivers/vc4/vc4_qir.h
+++ b/src/gallium/drivers/vc4/vc4_qir.h
@@ -247,6 +247,7 @@ enum quniform_contents {
         QUNIFORM_STENCIL,
 
         QUNIFORM_ALPHA_REF,
+        QUNIFORM_SAMPLE_MASK,
 };
 
 struct vc4_varying_slot {
diff --git a/src/gallium/drivers/vc4/vc4_state.c b/src/gallium/drivers/vc4/vc4_state.c
index a234ce53b20..2a88188a959 100644
--- a/src/gallium/drivers/vc4/vc4_state.c
+++ b/src/gallium/drivers/vc4/vc4_state.c
@@ -79,7 +79,7 @@ static void
 vc4_set_sample_mask(struct pipe_context *pctx, unsigned sample_mask)
 {
         struct vc4_context *vc4 = vc4_context(pctx);
-        vc4->sample_mask = (uint16_t)sample_mask;
+        vc4->sample_mask = sample_mask & ((1 << VC4_MAX_SAMPLES) - 1);
         vc4->dirty |= VC4_DIRTY_SAMPLE_MASK;
 }
 
diff --git a/src/gallium/drivers/vc4/vc4_uniforms.c b/src/gallium/drivers/vc4/vc4_uniforms.c
index f5ad481f186..5dfdd73f7bd 100644
--- a/src/gallium/drivers/vc4/vc4_uniforms.c
+++ b/src/gallium/drivers/vc4/vc4_uniforms.c
@@ -303,6 +303,10 @@ vc4_write_uniforms(struct vc4_context *vc4, struct vc4_compiled_shader *shader,
                         cl_aligned_f(&uniforms,
                                      vc4->zsa->base.alpha.ref_value);
                         break;
+
+                case QUNIFORM_SAMPLE_MASK:
+                        cl_aligned_u32(&uniforms, vc4->sample_mask);
+                        break;
                 }
 #if 0
                 uint32_t written_val = *((uint32_t *)uniforms - 1);
@@ -363,6 +367,10 @@ vc4_set_shader_uniform_dirty_flags(struct vc4_compiled_shader *shader)
                 case QUNIFORM_ALPHA_REF:
                         dirty |= VC4_DIRTY_ZSA;
                         break;
+
+                case QUNIFORM_SAMPLE_MASK:
+                        dirty |= VC4_DIRTY_SAMPLE_MASK;
+                        break;
                 }
         }
 
-- 
2.30.2