vc4: Add support for texel fetches from MSAA resources.

This is the core of ARB_texture_multisample.  Most of the piglit tests for
GL_ARB_texture_multisample require GL 3.0, but exposing support for this
lets us use the gallium blitter for multisample resolves.  We can
sometimes multisample resolve using just the RCL, but that requires that
the blit is 1:1, unflipped, and aligned to tile boundaries.

1 files changed, 90 insertions, 11 deletions

diff --git a/src/gallium/drivers/vc4/vc4_program.c b/src/gallium/drivers/vc4/vc4_program.c
index dda2d84..31968bb 100644
--- a/src/gallium/drivers/vc4/vc4_program.c
+++ b/src/gallium/drivers/vc4/vc4_program.c
@@ -294,6 +294,76 @@ ntq_umul(struct vc4_compile *c, struct qreg src0, struct qreg src1)
                                         qir_uniform_ui(c, 24)));
 }
 
+static struct qreg
+ntq_scale_depth_texture(struct vc4_compile *c, struct qreg src)
+{
+        struct qreg depthf = qir_ITOF(c, qir_SHR(c, src,
+                                                 qir_uniform_ui(c, 8)));
+        return qir_FMUL(c, depthf, qir_uniform_f(c, 1.0f/0xffffff));
+}
+
+/**
+ * Emits a lowered TXF_MS from an MSAA texture.
+ *
+ * The addressing math has been lowered in NIR, and now we just need to read
+ * it like a UBO.
+ */
+static void
+ntq_emit_txf(struct vc4_compile *c, nir_tex_instr *instr)
+{
+        uint32_t tile_width = 32;
+        uint32_t tile_height = 32;
+        uint32_t tile_size = (tile_height * tile_width *
+                              VC4_MAX_SAMPLES * sizeof(uint32_t));
+
+        unsigned unit = instr->sampler_index;
+        uint32_t w = align(c->key->tex[unit].msaa_width, tile_width);
+        uint32_t w_tiles = w / tile_width;
+        uint32_t h = align(c->key->tex[unit].msaa_height, tile_height);
+        uint32_t h_tiles = h / tile_height;
+        uint32_t size = w_tiles * h_tiles * tile_size;
+
+        struct qreg addr;
+        assert(instr->num_srcs == 1);
+        assert(instr->src[0].src_type == nir_tex_src_coord);
+        addr = ntq_get_src(c, instr->src[0].src, 0);
+
+        /* Perform the clamping required by kernel validation. */
+        addr = qir_MAX(c, addr, qir_uniform_ui(c, 0));
+        addr = qir_MIN(c, addr,  qir_uniform_ui(c, size - 4));
+
+        qir_TEX_DIRECT(c, addr, qir_uniform(c, QUNIFORM_TEXTURE_MSAA_ADDR, unit));
+
+        struct qreg tex = qir_TEX_RESULT(c);
+        c->num_texture_samples++;
+
+        struct qreg texture_output[4];
+        enum pipe_format format = c->key->tex[unit].format;
+        if (util_format_is_depth_or_stencil(format)) {
+                struct qreg scaled = ntq_scale_depth_texture(c, tex);
+                for (int i = 0; i < 4; i++)
+                        texture_output[i] = scaled;
+        } else {
+                struct qreg tex_result_unpacked[4];
+                for (int i = 0; i < 4; i++)
+                        tex_result_unpacked[i] = qir_UNPACK_8_F(c, tex, i);
+
+                const uint8_t *format_swiz =
+                        vc4_get_format_swizzle(c->key->tex[unit].format);
+                for (int i = 0; i < 4; i++) {
+                        texture_output[i] =
+                                get_swizzled_channel(c, tex_result_unpacked,
+                                                     format_swiz[i]);
+                }
+        }
+
+        struct qreg *dest = ntq_get_dest(c, &instr->dest);
+        for (int i = 0; i < 4; i++) {
+                dest[i] = get_swizzled_channel(c, texture_output,
+                                               c->key->tex[unit].swizzle[i]);
+        }
+}
+
 static void
 ntq_emit_tex(struct vc4_compile *c, nir_tex_instr *instr)
 {
@@ -301,6 +371,11 @@ ntq_emit_tex(struct vc4_compile *c, nir_tex_instr *instr)
         bool is_txb = false, is_txl = false, has_proj = false;
         unsigned unit = instr->sampler_index;
 
+        if (instr->op == nir_texop_txf) {
+                ntq_emit_txf(c, instr);
+                return;
+        }
+
         for (unsigned i = 0; i < instr->num_srcs; i++) {
                 switch (instr->src[i].src_type) {
                 case nir_tex_src_coord:
@@ -396,11 +471,7 @@ ntq_emit_tex(struct vc4_compile *c, nir_tex_instr *instr)
 
         struct qreg unpacked[4];
         if (util_format_is_depth_or_stencil(format)) {
-                struct qreg depthf = qir_ITOF(c, qir_SHR(c, tex,
-                                                         qir_uniform_ui(c, 8)));
-                struct qreg normalized = qir_FMUL(c, depthf,
-                                                  qir_uniform_f(c, 1.0f/0xffffff));
-
+                struct qreg normalized = ntq_scale_depth_texture(c, tex);
                 struct qreg depth_output;
 
                 struct qreg one = qir_uniform_f(c, 1.0f);
@@ -1712,6 +1783,7 @@ vc4_shader_ntq(struct vc4_context *vc4, enum qstage stage,
                 nir_lower_clip_vs(c->s, c->key->ucp_enables);
 
         vc4_nir_lower_io(c);
+        vc4_nir_lower_txf_ms(c);
         nir_lower_idiv(c->s);
         nir_lower_load_const_to_scalar(c->s);
 
@@ -1947,12 +2019,19 @@ vc4_setup_shared_key(struct vc4_context *vc4, struct vc4_key *key,
                 struct pipe_sampler_state *sampler_state =
                         texstate->samplers[i];
 
-                if (sampler) {
-                        key->tex[i].format = sampler->format;
-                        key->tex[i].swizzle[0] = sampler->swizzle_r;
-                        key->tex[i].swizzle[1] = sampler->swizzle_g;
-                        key->tex[i].swizzle[2] = sampler->swizzle_b;
-                        key->tex[i].swizzle[3] = sampler->swizzle_a;
+                if (!sampler)
+                        continue;
+
+                key->tex[i].format = sampler->format;
+                key->tex[i].swizzle[0] = sampler->swizzle_r;
+                key->tex[i].swizzle[1] = sampler->swizzle_g;
+                key->tex[i].swizzle[2] = sampler->swizzle_b;
+                key->tex[i].swizzle[3] = sampler->swizzle_a;
+
+                if (sampler->texture->nr_samples) {
+                        key->tex[i].msaa_width = sampler->texture->width0;
+                        key->tex[i].msaa_height = sampler->texture->height0;
+                } else if (sampler){
                         key->tex[i].compare_mode = sampler_state->compare_mode;
                         key->tex[i].compare_func = sampler_state->compare_func;
                         key->tex[i].wrap_s = sampler_state->wrap_s;