util: Move gallium's PIPE_FORMAT utils to /util/format/
[mesa.git] / src / gallium / drivers / freedreno / a3xx / fd3_texture.c
index f28919f474e38a4145f0a3a08d148821587aa758..9fc91273b6d89d3c362ac3dace79c251aeb31eaf 100644 (file)
@@ -1,5 +1,3 @@
-/* -*- mode: C; c-file-style: "k&r"; tab-width 4; indent-tabs-mode: t; -*- */
-
 /*
  * Copyright (C) 2013 Rob Clark <robclark@freedesktop.org>
  *
 #include "util/u_string.h"
 #include "util/u_memory.h"
 #include "util/u_inlines.h"
-#include "util/u_format.h"
+#include "util/format/u_format.h"
 
 #include "fd3_texture.h"
-#include "fd3_util.h"
+#include "fd3_format.h"
 
 static enum a3xx_tex_clamp
-tex_clamp(unsigned wrap)
+tex_clamp(unsigned wrap, bool clamp_to_edge, bool *needs_border)
 {
-       /* hardware probably supports more, but we can't coax all the
-        * wrap/clamp modes out of the GLESv2 blob driver.
-        *
-        * TODO once we have basics working, go back and just try
-        * different values and see what happens
-        */
+       /* Hardware does not support _CLAMP, but we emulate it: */
+       if (wrap == PIPE_TEX_WRAP_CLAMP) {
+               wrap = (clamp_to_edge) ?
+                       PIPE_TEX_WRAP_CLAMP_TO_EDGE : PIPE_TEX_WRAP_CLAMP_TO_BORDER;
+       }
+
        switch (wrap) {
        case PIPE_TEX_WRAP_REPEAT:
                return A3XX_TEX_REPEAT;
-       case PIPE_TEX_WRAP_CLAMP:
        case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
                return A3XX_TEX_CLAMP_TO_EDGE;
        case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
+               *needs_border = true;
                return A3XX_TEX_CLAMP_TO_BORDER;
-       case PIPE_TEX_WRAP_MIRROR_CLAMP:
-       case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
        case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
+               /* only works for PoT.. need to emulate otherwise! */
                return A3XX_TEX_MIRROR_CLAMP;
        case PIPE_TEX_WRAP_MIRROR_REPEAT:
                return A3XX_TEX_MIRROR_REPEAT;
+       case PIPE_TEX_WRAP_MIRROR_CLAMP:
+       case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
+               /* these two we could perhaps emulate, but we currently
+                * just don't advertise PIPE_CAP_TEXTURE_MIRROR_CLAMP
+                */
        default:
                DBG("invalid wrap: %u", wrap);
                return 0;
@@ -65,13 +67,13 @@ tex_clamp(unsigned wrap)
 }
 
 static enum a3xx_tex_filter
-tex_filter(unsigned filter)
+tex_filter(unsigned filter, bool aniso)
 {
        switch (filter) {
        case PIPE_TEX_FILTER_NEAREST:
                return A3XX_TEX_NEAREST;
        case PIPE_TEX_FILTER_LINEAR:
-               return A3XX_TEX_LINEAR;
+               return aniso ? A3XX_TEX_ANISO : A3XX_TEX_LINEAR;
        default:
                DBG("invalid filter: %u", filter);
                return 0;
@@ -83,7 +85,9 @@ fd3_sampler_state_create(struct pipe_context *pctx,
                const struct pipe_sampler_state *cso)
 {
        struct fd3_sampler_stateobj *so = CALLOC_STRUCT(fd3_sampler_stateobj);
+       unsigned aniso = util_last_bit(MIN2(cso->max_anisotropy >> 1, 8));
        bool miplinear = false;
+       bool clamp_to_edge;
 
        if (!so)
                return NULL;
@@ -93,29 +97,102 @@ fd3_sampler_state_create(struct pipe_context *pctx,
 
        so->base = *cso;
 
+       /*
+        * For nearest filtering, _CLAMP means _CLAMP_TO_EDGE;  for linear
+        * filtering, _CLAMP means _CLAMP_TO_BORDER while additionally
+        * clamping the texture coordinates to [0.0, 1.0].
+        *
+        * The clamping will be taken care of in the shaders.  There are two
+        * filters here, but let the minification one has a say.
+        */
+       clamp_to_edge = (cso->min_img_filter == PIPE_TEX_FILTER_NEAREST);
+       if (!clamp_to_edge) {
+               so->saturate_s = (cso->wrap_s == PIPE_TEX_WRAP_CLAMP);
+               so->saturate_t = (cso->wrap_t == PIPE_TEX_WRAP_CLAMP);
+               so->saturate_r = (cso->wrap_r == PIPE_TEX_WRAP_CLAMP);
+       }
+
+       so->needs_border = false;
        so->texsamp0 =
                        COND(!cso->normalized_coords, A3XX_TEX_SAMP_0_UNNORM_COORDS) |
+                       COND(!cso->seamless_cube_map, A3XX_TEX_SAMP_0_CUBEMAPSEAMLESSFILTOFF) |
                        COND(miplinear, A3XX_TEX_SAMP_0_MIPFILTER_LINEAR) |
-                       A3XX_TEX_SAMP_0_XY_MAG(tex_filter(cso->mag_img_filter)) |
-                       A3XX_TEX_SAMP_0_XY_MIN(tex_filter(cso->min_img_filter)) |
-                       A3XX_TEX_SAMP_0_WRAP_S(tex_clamp(cso->wrap_s)) |
-                       A3XX_TEX_SAMP_0_WRAP_T(tex_clamp(cso->wrap_t)) |
-                       A3XX_TEX_SAMP_0_WRAP_R(tex_clamp(cso->wrap_r));
+                       A3XX_TEX_SAMP_0_XY_MAG(tex_filter(cso->mag_img_filter, aniso)) |
+                       A3XX_TEX_SAMP_0_XY_MIN(tex_filter(cso->min_img_filter, aniso)) |
+                       A3XX_TEX_SAMP_0_ANISO(aniso) |
+                       A3XX_TEX_SAMP_0_WRAP_S(tex_clamp(cso->wrap_s, clamp_to_edge, &so->needs_border)) |
+                       A3XX_TEX_SAMP_0_WRAP_T(tex_clamp(cso->wrap_t, clamp_to_edge, &so->needs_border)) |
+                       A3XX_TEX_SAMP_0_WRAP_R(tex_clamp(cso->wrap_r, clamp_to_edge, &so->needs_border));
 
        if (cso->compare_mode)
                so->texsamp0 |= A3XX_TEX_SAMP_0_COMPARE_FUNC(cso->compare_func); /* maps 1:1 */
 
+       so->texsamp1 = A3XX_TEX_SAMP_1_LOD_BIAS(cso->lod_bias);
+
        if (cso->min_mip_filter != PIPE_TEX_MIPFILTER_NONE) {
-               so->texsamp1 =
-                               A3XX_TEX_SAMP_1_MIN_LOD(cso->min_lod) |
-                               A3XX_TEX_SAMP_1_MAX_LOD(cso->max_lod);
+               so->texsamp1 |=
+                       A3XX_TEX_SAMP_1_MIN_LOD(cso->min_lod) |
+                       A3XX_TEX_SAMP_1_MAX_LOD(cso->max_lod);
        } else {
-               so->texsamp1 = 0x00000000;
+               /* If we're not doing mipmap filtering, we still need a slightly > 0
+                * LOD clamp so the HW can decide between min and mag filtering of
+                * level 0.
+                */
+               so->texsamp1 |=
+                       A3XX_TEX_SAMP_1_MIN_LOD(MIN2(cso->min_lod, 0.125)) |
+                       A3XX_TEX_SAMP_1_MAX_LOD(MIN2(cso->max_lod, 0.125));
        }
 
        return so;
 }
 
+static void
+fd3_sampler_states_bind(struct pipe_context *pctx,
+               enum pipe_shader_type shader, unsigned start,
+               unsigned nr, void **hwcso)
+{
+       struct fd_context *ctx = fd_context(pctx);
+       struct fd3_context *fd3_ctx = fd3_context(ctx);
+       uint16_t saturate_s = 0, saturate_t = 0, saturate_r = 0;
+       unsigned i;
+
+       if (!hwcso)
+               nr = 0;
+
+       for (i = 0; i < nr; i++) {
+               if (hwcso[i]) {
+                       struct fd3_sampler_stateobj *sampler =
+                                       fd3_sampler_stateobj(hwcso[i]);
+                       if (sampler->saturate_s)
+                               saturate_s |= (1 << i);
+                       if (sampler->saturate_t)
+                               saturate_t |= (1 << i);
+                       if (sampler->saturate_r)
+                               saturate_r |= (1 << i);
+               }
+       }
+
+       fd_sampler_states_bind(pctx, shader, start, nr, hwcso);
+
+       if (shader == PIPE_SHADER_FRAGMENT) {
+               fd3_ctx->fsaturate =
+                       (saturate_s != 0) ||
+                       (saturate_t != 0) ||
+                       (saturate_r != 0);
+               fd3_ctx->fsaturate_s = saturate_s;
+               fd3_ctx->fsaturate_t = saturate_t;
+               fd3_ctx->fsaturate_r = saturate_r;
+       } else if (shader == PIPE_SHADER_VERTEX) {
+               fd3_ctx->vsaturate =
+                       (saturate_s != 0) ||
+                       (saturate_t != 0) ||
+                       (saturate_r != 0);
+               fd3_ctx->vsaturate_s = saturate_s;
+               fd3_ctx->vsaturate_t = saturate_t;
+               fd3_ctx->vsaturate_r = saturate_r;
+       }
+}
+
 static enum a3xx_tex_type
 tex_type(unsigned target)
 {
@@ -144,7 +221,8 @@ fd3_sampler_view_create(struct pipe_context *pctx, struct pipe_resource *prsc,
 {
        struct fd3_pipe_sampler_view *so = CALLOC_STRUCT(fd3_pipe_sampler_view);
        struct fd_resource *rsc = fd_resource(prsc);
-       unsigned miplevels = cso->u.tex.last_level - cso->u.tex.first_level;
+       unsigned lvl;
+       uint32_t sz2 = 0;
 
        if (!so)
                return NULL;
@@ -155,27 +233,58 @@ fd3_sampler_view_create(struct pipe_context *pctx, struct pipe_resource *prsc,
        so->base.reference.count = 1;
        so->base.context = pctx;
 
-       so->tex_resource =  rsc;
-       so->mipaddrs = 1 + miplevels;
-
        so->texconst0 =
+                       A3XX_TEX_CONST_0_TILE_MODE(rsc->tile_mode) |
                        A3XX_TEX_CONST_0_TYPE(tex_type(prsc->target)) |
                        A3XX_TEX_CONST_0_FMT(fd3_pipe2tex(cso->format)) |
-                       A3XX_TEX_CONST_0_MIPLVLS(miplevels) |
                        fd3_tex_swiz(cso->format, cso->swizzle_r, cso->swizzle_g,
                                                cso->swizzle_b, cso->swizzle_a);
 
+       if (prsc->target == PIPE_BUFFER || util_format_is_pure_integer(cso->format))
+               so->texconst0 |= A3XX_TEX_CONST_0_NOCONVERT;
        if (util_format_is_srgb(cso->format))
                so->texconst0 |= A3XX_TEX_CONST_0_SRGB;
 
-       so->texconst1 =
+       if (prsc->target == PIPE_BUFFER) {
+               lvl = 0;
+               so->texconst1 =
                        A3XX_TEX_CONST_1_FETCHSIZE(fd3_pipe2fetchsize(cso->format)) |
-                       A3XX_TEX_CONST_1_WIDTH(prsc->width0) |
-                       A3XX_TEX_CONST_1_HEIGHT(prsc->height0);
+                       A3XX_TEX_CONST_1_WIDTH(cso->u.buf.size / util_format_get_blocksize(cso->format)) |
+                       A3XX_TEX_CONST_1_HEIGHT(1);
+       } else {
+               unsigned miplevels;
+
+               lvl = fd_sampler_first_level(cso);
+               miplevels = fd_sampler_last_level(cso) - lvl;
+
+               so->texconst0 |= A3XX_TEX_CONST_0_MIPLVLS(miplevels);
+               so->texconst1 =
+                       A3XX_TEX_CONST_1_FETCHSIZE(fd3_pipe2fetchsize(cso->format)) |
+                       A3XX_TEX_CONST_1_WIDTH(u_minify(prsc->width0, lvl)) |
+                       A3XX_TEX_CONST_1_HEIGHT(u_minify(prsc->height0, lvl));
+       }
        /* when emitted, A3XX_TEX_CONST_2_INDX() must be OR'd in: */
        so->texconst2 =
-                       A3XX_TEX_CONST_2_PITCH(rsc->slices[0].pitch * rsc->cpp);
-       so->texconst3 = 0x00000000;  /* ??? */
+                       A3XX_TEX_CONST_2_PITCH(fd3_pipe2nblocksx(cso->format, rsc->slices[lvl].pitch) * rsc->cpp);
+       switch (prsc->target) {
+       case PIPE_TEXTURE_1D_ARRAY:
+       case PIPE_TEXTURE_2D_ARRAY:
+               so->texconst3 =
+                               A3XX_TEX_CONST_3_DEPTH(prsc->array_size - 1) |
+                               A3XX_TEX_CONST_3_LAYERSZ1(rsc->slices[0].size0);
+               break;
+       case PIPE_TEXTURE_3D:
+               so->texconst3 =
+                               A3XX_TEX_CONST_3_DEPTH(u_minify(prsc->depth0, lvl)) |
+                               A3XX_TEX_CONST_3_LAYERSZ1(rsc->slices[lvl].size0);
+               while (lvl < cso->u.tex.last_level && sz2 != rsc->slices[lvl+1].size0)
+                       sz2 = rsc->slices[++lvl].size0;
+               so->texconst3 |= A3XX_TEX_CONST_3_LAYERSZ2(sz2);
+               break;
+       default:
+               so->texconst3 = 0x00000000;
+               break;
+       }
 
        return &so->base;
 }
@@ -184,5 +293,7 @@ void
 fd3_texture_init(struct pipe_context *pctx)
 {
        pctx->create_sampler_state = fd3_sampler_state_create;
+       pctx->bind_sampler_states = fd3_sampler_states_bind;
        pctx->create_sampler_view = fd3_sampler_view_create;
+       pctx->set_sampler_views = fd_set_sampler_views;
 }