panfrost: XMLify stencil test
[mesa.git] / src / gallium / drivers / panfrost / pan_cmdstream.c
index 3af812f3e859f62d723baad6ee231840e2602b4a..8f6142f7b4a76415eda630638ce97e7d4bc1120f 100644 (file)
@@ -101,7 +101,7 @@ panfrost_vt_update_primitive_size(struct panfrost_context *ctx,
         struct panfrost_rasterizer *rasterizer = ctx->rasterizer;
 
         if (!panfrost_writes_point_size(ctx)) {
-                bool points = prefix->draw_mode == MALI_POINTS;
+                bool points = prefix->draw_mode == MALI_DRAW_MODE_POINTS;
                 float val = 0.0f;
 
                 if (rasterizer)
@@ -369,7 +369,7 @@ panfrost_translate_compare_func(enum pipe_compare_func in)
                 return MALI_FUNC_GREATER;
 
         case PIPE_FUNC_NOTEQUAL:
-                return MALI_FUNC_NOTEQUAL;
+                return MALI_FUNC_NOT_EQUAL;
 
         case PIPE_FUNC_GEQUAL:
                 return MALI_FUNC_GEQUAL;
@@ -387,28 +387,28 @@ panfrost_translate_stencil_op(enum pipe_stencil_op in)
 {
         switch (in) {
         case PIPE_STENCIL_OP_KEEP:
-                return MALI_STENCIL_KEEP;
+                return MALI_STENCIL_OP_KEEP;
 
         case PIPE_STENCIL_OP_ZERO:
-                return MALI_STENCIL_ZERO;
+                return MALI_STENCIL_OP_ZERO;
 
         case PIPE_STENCIL_OP_REPLACE:
-               return MALI_STENCIL_REPLACE;
+               return MALI_STENCIL_OP_REPLACE;
 
         case PIPE_STENCIL_OP_INCR:
-                return MALI_STENCIL_INCR;
+                return MALI_STENCIL_OP_INCR_SAT;
 
         case PIPE_STENCIL_OP_DECR:
-                return MALI_STENCIL_DECR;
+                return MALI_STENCIL_OP_DECR_SAT;
 
         case PIPE_STENCIL_OP_INCR_WRAP:
-                return MALI_STENCIL_INCR_WRAP;
+                return MALI_STENCIL_OP_INCR_WRAP;
 
         case PIPE_STENCIL_OP_DECR_WRAP:
-                return MALI_STENCIL_DECR_WRAP;
+                return MALI_STENCIL_OP_DECR_WRAP;
 
         case PIPE_STENCIL_OP_INVERT:
-                return MALI_STENCIL_INVERT;
+                return MALI_STENCIL_OP_INVERT;
 
         default:
                 unreachable("Invalid stencil op");
@@ -420,28 +420,28 @@ translate_tex_wrap(enum pipe_tex_wrap w)
 {
         switch (w) {
         case PIPE_TEX_WRAP_REPEAT:
-                return MALI_WRAP_REPEAT;
+                return MALI_WRAP_MODE_REPEAT;
 
         case PIPE_TEX_WRAP_CLAMP:
-                return MALI_WRAP_CLAMP;
+                return MALI_WRAP_MODE_CLAMP;
 
         case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
-                return MALI_WRAP_CLAMP_TO_EDGE;
+                return MALI_WRAP_MODE_CLAMP_TO_EDGE;
 
         case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
-                return MALI_WRAP_CLAMP_TO_BORDER;
+                return MALI_WRAP_MODE_CLAMP_TO_BORDER;
 
         case PIPE_TEX_WRAP_MIRROR_REPEAT:
-                return MALI_WRAP_MIRRORED_REPEAT;
+                return MALI_WRAP_MODE_MIRRORED_REPEAT;
 
         case PIPE_TEX_WRAP_MIRROR_CLAMP:
-                return MALI_WRAP_MIRRORED_CLAMP;
+                return MALI_WRAP_MODE_MIRRORED_CLAMP;
 
         case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
-                return MALI_WRAP_MIRRORED_CLAMP_TO_EDGE;
+                return MALI_WRAP_MODE_MIRRORED_CLAMP_TO_EDGE;
 
         case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
-                return MALI_WRAP_MIRRORED_CLAMP_TO_BORDER;
+                return MALI_WRAP_MODE_MIRRORED_CLAMP_TO_BORDER;
 
         default:
                 unreachable("Invalid wrap");
@@ -519,15 +519,15 @@ void panfrost_sampler_desc_init_bifrost(const struct pipe_sampler_state *cso,
 
 static void
 panfrost_make_stencil_state(const struct pipe_stencil_state *in,
-                            struct mali_stencil_test *out)
+                            void *out)
 {
-        out->ref = 0; /* Gallium gets it from elsewhere */
-
-        out->mask = in->valuemask;
-        out->func = panfrost_translate_compare_func(in->func);
-        out->sfail = panfrost_translate_stencil_op(in->fail_op);
-        out->dpfail = panfrost_translate_stencil_op(in->zfail_op);
-        out->dppass = panfrost_translate_stencil_op(in->zpass_op);
+        pan_pack(out, STENCIL, cfg) {
+                cfg.mask = in->valuemask;
+                cfg.compare_function = panfrost_translate_compare_func(in->func);
+                cfg.stencil_fail = panfrost_translate_stencil_op(in->fail_op);
+                cfg.depth_fail = panfrost_translate_stencil_op(in->zfail_op);
+                cfg.depth_pass = panfrost_translate_stencil_op(in->zpass_op);
+        }
 }
 
 static void
@@ -585,9 +585,9 @@ panfrost_frag_meta_zsa_update(struct panfrost_context *ctx,
                 struct pipe_stencil_state default_stencil = {
                         .enabled = 0,
                         .func = PIPE_FUNC_ALWAYS,
-                        .fail_op = MALI_STENCIL_KEEP,
-                        .zfail_op = MALI_STENCIL_KEEP,
-                        .zpass_op = MALI_STENCIL_KEEP,
+                        .fail_op = PIPE_STENCIL_OP_KEEP,
+                        .zfail_op = PIPE_STENCIL_OP_KEEP,
+                        .zpass_op = PIPE_STENCIL_OP_KEEP,
                         .writemask = 0xFF,
                         .valuemask = 0xFF
                 };
@@ -605,7 +605,10 @@ panfrost_frag_meta_zsa_update(struct panfrost_context *ctx,
                 panfrost_make_stencil_state(&zsa->stencil[0],
                                             &fragmeta->stencil_front);
                 fragmeta->stencil_mask_front = zsa->stencil[0].writemask;
-                fragmeta->stencil_front.ref = ctx->stencil_ref.ref_value[0];
+
+                /* Bottom 8-bits of stencil state is the stencil ref, ref is no
+                 * more than 8-bits. Be extra careful. */
+                fragmeta->stencil_front.opaque[0] |= ctx->stencil_ref.ref_value[0];
 
                 /* If back-stencil is not enabled, use the front values */
 
@@ -613,11 +616,10 @@ panfrost_frag_meta_zsa_update(struct panfrost_context *ctx,
                         panfrost_make_stencil_state(&zsa->stencil[1],
                                                     &fragmeta->stencil_back);
                         fragmeta->stencil_mask_back = zsa->stencil[1].writemask;
-                        fragmeta->stencil_back.ref = ctx->stencil_ref.ref_value[1];
+                        fragmeta->stencil_back.opaque[0] |= ctx->stencil_ref.ref_value[1];
                 } else {
                         fragmeta->stencil_back = fragmeta->stencil_front;
                         fragmeta->stencil_mask_back = fragmeta->stencil_mask_front;
-                        fragmeta->stencil_back.ref = fragmeta->stencil_front.ref;
                 }
 
                 if (zsa->depth.enabled)
@@ -958,128 +960,55 @@ panfrost_emit_shader_meta(struct panfrost_batch *batch,
         postfix->shader = shader_ptr;
 }
 
-static void
-panfrost_mali_viewport_init(struct panfrost_context *ctx,
-                            struct mali_viewport *mvp)
+void
+panfrost_emit_viewport(struct panfrost_batch *batch,
+                       struct mali_vertex_tiler_postfix *tiler_postfix)
 {
+        struct panfrost_context *ctx = batch->ctx;
         const struct pipe_viewport_state *vp = &ctx->pipe_viewport;
-
-        /* Clip bounds are encoded as floats. The viewport itself is encoded as
-         * (somewhat) asymmetric ints. */
-
         const struct pipe_scissor_state *ss = &ctx->scissor;
+        const struct pipe_rasterizer_state *rast = &ctx->rasterizer->base;
+        const struct pipe_framebuffer_state *fb = &ctx->pipe_framebuffer;
 
-        memset(mvp, 0, sizeof(*mvp));
-
-        /* By default, do no viewport clipping, i.e. clip to (-inf, inf) in
-         * each direction. Clipping to the viewport in theory should work, but
-         * in practice causes issues when we're not explicitly trying to
-         * scissor */
-
-        *mvp = (struct mali_viewport) {
-                .clip_minx = -INFINITY,
-                .clip_miny = -INFINITY,
-                .clip_maxx = INFINITY,
-                .clip_maxy = INFINITY,
-        };
-
-        /* Always scissor to the viewport by default. */
+        /* Derive min/max from translate/scale. Note since |x| >= 0 by
+         * definition, we have that -|x| <= |x| hence translate - |scale| <=
+         * translate + |scale|, so the ordering is correct here. */
         float vp_minx = (int) (vp->translate[0] - fabsf(vp->scale[0]));
         float vp_maxx = (int) (vp->translate[0] + fabsf(vp->scale[0]));
-
         float vp_miny = (int) (vp->translate[1] - fabsf(vp->scale[1]));
         float vp_maxy = (int) (vp->translate[1] + fabsf(vp->scale[1]));
-
         float minz = (vp->translate[2] - fabsf(vp->scale[2]));
         float maxz = (vp->translate[2] + fabsf(vp->scale[2]));
 
-        /* Apply the scissor test */
-
-        unsigned minx, miny, maxx, maxy;
-
-        if (ss && ctx->rasterizer && ctx->rasterizer->base.scissor) {
-                minx = MAX2(ss->minx, vp_minx);
-                miny = MAX2(ss->miny, vp_miny);
-                maxx = MIN2(ss->maxx, vp_maxx);
-                maxy = MIN2(ss->maxy, vp_maxy);
-        } else {
-                minx = vp_minx;
-                miny = vp_miny;
-                maxx = vp_maxx;
-                maxy = vp_maxy;
-        }
-
-        /* Hardware needs the min/max to be strictly ordered, so flip if we
-         * need to. The viewport transformation in the vertex shader will
-         * handle the negatives if we don't */
-
-        if (miny > maxy) {
-                unsigned temp = miny;
-                miny = maxy;
-                maxy = temp;
-        }
+        /* Scissor to the intersection of viewport and to the scissor, clamped
+         * to the framebuffer */
 
-        if (minx > maxx) {
-                unsigned temp = minx;
-                minx = maxx;
-                maxx = temp;
-        }
+        unsigned minx = MIN2(fb->width, vp_minx);
+        unsigned maxx = MIN2(fb->width, vp_maxx);
+        unsigned miny = MIN2(fb->height, vp_miny);
+        unsigned maxy = MIN2(fb->height, vp_maxy);
 
-        if (minz > maxz) {
-                float temp = minz;
-                minz = maxz;
-                maxz = temp;
+        if (ss && rast && rast->scissor) {
+                minx = MAX2(ss->minx, minx);
+                miny = MAX2(ss->miny, miny);
+                maxx = MIN2(ss->maxx, maxx);
+                maxy = MIN2(ss->maxy, maxy);
         }
 
-        /* Clamp to the framebuffer size as a last check */
-
-        minx = MIN2(ctx->pipe_framebuffer.width, minx);
-        maxx = MIN2(ctx->pipe_framebuffer.width, maxx);
-
-        miny = MIN2(ctx->pipe_framebuffer.height, miny);
-        maxy = MIN2(ctx->pipe_framebuffer.height, maxy);
-
-        /* Upload */
-
-        mvp->viewport0[0] = minx;
-        mvp->viewport1[0] = MALI_POSITIVE(maxx);
-
-        mvp->viewport0[1] = miny;
-        mvp->viewport1[1] = MALI_POSITIVE(maxy);
+        struct panfrost_transfer T = panfrost_pool_alloc(&batch->pool, MALI_VIEWPORT_LENGTH);
 
-        bool clip_near = true;
-        bool clip_far = true;
+        pan_pack(T.cpu, VIEWPORT, cfg) {
+                cfg.scissor_minimum_x = minx;
+                cfg.scissor_minimum_y = miny;
+                cfg.scissor_maximum_x = maxx - 1;
+                cfg.scissor_maximum_y = maxy - 1;
 
-        if (ctx->rasterizer) {
-                clip_near = ctx->rasterizer->base.depth_clip_near;
-                clip_far = ctx->rasterizer->base.depth_clip_far;
+                cfg.minimum_z = rast->depth_clip_near ? minz : -INFINITY;
+                cfg.maximum_z = rast->depth_clip_far ? maxz : INFINITY;
         }
 
-        mvp->clip_minz = clip_near ? minz : -INFINITY;
-        mvp->clip_maxz = clip_far ? maxz : INFINITY;
-}
-
-void
-panfrost_emit_viewport(struct panfrost_batch *batch,
-                       struct mali_vertex_tiler_postfix *tiler_postfix)
-{
-        struct panfrost_context *ctx = batch->ctx;
-        struct mali_viewport mvp;
-
-        panfrost_mali_viewport_init(batch->ctx,  &mvp);
-
-        /* Update the job, unless we're doing wallpapering (whose lack of
-         * scissor we can ignore, since if we "miss" a tile of wallpaper, it'll
-         * just... be faster :) */
-
-        if (!ctx->wallpaper_batch)
-                panfrost_batch_union_scissor(batch, mvp.viewport0[0],
-                                             mvp.viewport0[1],
-                                             mvp.viewport1[0] + 1,
-                                             mvp.viewport1[1] + 1);
-
-        tiler_postfix->viewport = panfrost_pool_upload(&batch->pool, &mvp,
-                                                            sizeof(mvp));
+        tiler_postfix->viewport = T.gpu;
+        panfrost_batch_union_scissor(batch, minx, miny, maxx, maxy);
 }
 
 static mali_ptr
@@ -1324,12 +1253,20 @@ panfrost_emit_const_buf(struct panfrost_batch *batch,
         unsigned ubo_count = panfrost_ubo_count(ctx, stage);
         assert(ubo_count >= 1);
 
-        size_t sz = sizeof(uint64_t) * ubo_count;
-        uint64_t ubos[PAN_MAX_CONST_BUFFERS];
-        int uniform_count = ss->uniform_count;
+        size_t sz = MALI_UNIFORM_BUFFER_LENGTH * ubo_count;
+        struct panfrost_transfer ubos = panfrost_pool_alloc(&batch->pool, sz);
+        uint64_t *ubo_ptr = (uint64_t *) ubos.cpu;
 
         /* Upload uniforms as a UBO */
-        ubos[0] = MALI_MAKE_UBO(2 + uniform_count, transfer.gpu);
+
+        if (ss->uniform_count) {
+                pan_pack(ubo_ptr, UNIFORM_BUFFER, cfg) {
+                        cfg.entries = ss->uniform_count;
+                        cfg.pointer = transfer.gpu;
+                }
+        } else {
+                *ubo_ptr = 0;
+        }
 
         /* The rest are honest-to-goodness UBOs */
 
@@ -1339,22 +1276,19 @@ panfrost_emit_const_buf(struct panfrost_batch *batch,
                 bool empty = usz == 0;
 
                 if (!enabled || empty) {
-                        /* Stub out disabled UBOs to catch accesses */
-                        ubos[ubo] = MALI_MAKE_UBO(0, 0xDEAD0000);
+                        ubo_ptr[ubo] = 0;
                         continue;
                 }
 
-                mali_ptr gpu = panfrost_map_constant_buffer_gpu(batch, stage,
-                                                                buf, ubo);
-
-                unsigned bytes_per_field = 16;
-                unsigned aligned = ALIGN_POT(usz, bytes_per_field);
-                ubos[ubo] = MALI_MAKE_UBO(aligned / bytes_per_field, gpu);
+                pan_pack(ubo_ptr + ubo, UNIFORM_BUFFER, cfg) {
+                        cfg.entries = DIV_ROUND_UP(usz, 16);
+                        cfg.pointer = panfrost_map_constant_buffer_gpu(batch,
+                                        stage, buf, ubo);
+                }
         }
 
-        mali_ptr ubufs = panfrost_pool_upload(&batch->pool, ubos, sz);
         postfix->uniforms = transfer.gpu;
-        postfix->uniform_buffers = ubufs;
+        postfix->uniform_buffers = ubos.gpu;
 
         buf->dirty_mask = 0;
 }