freedreno/ir3: txf_ms support

Signed-off-by: Rob Clark <robdclark@gmail.com>

diff --git a/src/gallium/drivers/freedreno/ir3/ir3_compiler_nir.c b/src/gallium/drivers/freedreno/ir3/ir3_compiler_nir.c
index 0fe88e1f8264cdfe7a64c1ce1a907961cdda2ef7..323d278b431b97b0b52afde9250729cf633ae32a 100644 (file)
--- a/src/gallium/drivers/freedreno/ir3/ir3_compiler_nir.c
+++ b/src/gallium/drivers/freedreno/ir3/ir3_compiler_nir.c
@@ -118,6 +118,9 @@ struct ir3_context {
         */
        bool unminify_coords;
 
+       /* on a3xx do txf_ms w/ isaml and scaled coords: */
+       bool txf_ms_with_isaml;
+
        /* on a4xx, for array textures we need to add 0.5 to the array
         * index coordinate:
         */
@@ -126,6 +129,8 @@ struct ir3_context {
        /* on a4xx, bitmask of samplers which need astc+srgb workaround: */
        unsigned astc_srgb;
 
+       unsigned samples;             /* bitmask of x,y sample shifts */
+
        unsigned max_texture_index;
 
        /* set if we encounter something we can't handle yet, so we
@@ -155,23 +160,31 @@ compile_init(struct ir3_compiler *compiler,
                ctx->flat_bypass = true;
                ctx->levels_add_one = false;
                ctx->unminify_coords = false;
+               ctx->txf_ms_with_isaml = false;
                ctx->array_index_add_half = true;
 
-               if (so->type == SHADER_VERTEX)
+               if (so->type == SHADER_VERTEX) {
                        ctx->astc_srgb = so->key.vastc_srgb;
-               else if (so->type == SHADER_FRAGMENT)
+               } else if (so->type == SHADER_FRAGMENT) {
                        ctx->astc_srgb = so->key.fastc_srgb;
+               }
 
        } else {
                /* no special handling for "flat" */
                ctx->flat_bypass = false;
                ctx->levels_add_one = true;
                ctx->unminify_coords = true;
+               ctx->txf_ms_with_isaml = true;
                ctx->array_index_add_half = false;
+
+               if (so->type == SHADER_VERTEX) {
+                       ctx->samples = so->key.vsamples;
+               } else if (so->type == SHADER_FRAGMENT) {
+                       ctx->samples = so->key.fsamples;
+               }
        }
 
        ctx->compiler = compiler;
-       ctx->ir = so->ir;
        ctx->so = so;
        ctx->def_ht = _mesa_hash_table_create(ctx,
                        _mesa_hash_pointer, _mesa_key_pointer_equal);
@@ -2601,7 +2614,7 @@ emit_tex(struct ir3_context *ctx, nir_tex_instr *tex)
        struct ir3_block *b = ctx->block;
        struct ir3_instruction **dst, *sam, *src0[12], *src1[4];
        struct ir3_instruction * const *coord, * const *off, * const *ddx, * const *ddy;
-       struct ir3_instruction *lod, *compare, *proj;
+       struct ir3_instruction *lod, *compare, *proj, *sample_index;
        bool has_bias = false, has_lod = false, has_proj = false, has_off = false;
        unsigned i, coords, flags;
        unsigned nsrc0 = 0, nsrc1 = 0;
@@ -2609,7 +2622,7 @@ emit_tex(struct ir3_context *ctx, nir_tex_instr *tex)
        opc_t opc = 0;
 
        coord = off = ddx = ddy = NULL;
-       lod = proj = compare = NULL;
+       lod = proj = compare = sample_index = NULL;
 
        /* TODO: might just be one component for gathers? */
        dst = get_dst(ctx, &tex->dest, 4);
@@ -2644,6 +2657,9 @@ emit_tex(struct ir3_context *ctx, nir_tex_instr *tex)
                case nir_tex_src_ddy:
                        ddy = get_src(ctx, &tex->src[i].src);
                        break;
+               case nir_tex_src_ms_index:
+                       sample_index = get_src(ctx, &tex->src[i].src)[0];
+                       break;
                default:
                        compile_error(ctx, "Unhandled NIR tex src type: %d\n",
                                        tex->src[i].src_type);
@@ -2670,7 +2686,7 @@ emit_tex(struct ir3_context *ctx, nir_tex_instr *tex)
                case 3:              opc = OPC_GATHER4A; break;
                }
                break;
-       case nir_texop_txf_ms:
+       case nir_texop_txf_ms:   opc = OPC_ISAMM;    break;
        case nir_texop_txs:
        case nir_texop_query_levels:
        case nir_texop_texture_samples:
@@ -2699,6 +2715,27 @@ emit_tex(struct ir3_context *ctx, nir_tex_instr *tex)
 
        nsrc0 = i;
 
+       /* NOTE a3xx (and possibly a4xx?) might be different, using isaml
+        * with scaled x coord according to requested sample:
+        */
+       if (tex->op == nir_texop_txf_ms) {
+               if (ctx->txf_ms_with_isaml) {
+                       /* the samples are laid out in x dimension as
+                        *     0 1 2 3
+                        * x_ms = (x << ms) + sample_index;
+                        */
+                       struct ir3_instruction *ms;
+                       ms = create_immed(b, (ctx->samples >> (2 * tex->texture_index)) & 3);
+
+                       src0[0] = ir3_SHL_B(b, src0[0], 0, ms, 0);
+                       src0[0] = ir3_ADD_U(b, src0[0], 0, sample_index, 0);
+
+                       opc = OPC_ISAML;
+               } else {
+                       src0[nsrc0++] = sample_index;
+               }
+       }
+
        /* scale up integer coords for TXF based on the LOD */
        if (ctx->unminify_coords && (opc == OPC_ISAML)) {
                assert(has_lod);
@@ -3726,7 +3763,7 @@ ir3_compile_shader_nir(struct ir3_compiler *compiler,
                        so->varying_in++;
                        so->inputs[i].compmask = (1 << maxcomp) - 1;
                        inloc += maxcomp;
-               } else if (!so->inputs[i].sysval){
+               } else if (!so->inputs[i].sysval) {
                        so->inputs[i].compmask = compmask;
                }
                so->inputs[i].regid = regid;

diff --git a/src/gallium/drivers/freedreno/ir3/ir3_shader.c b/src/gallium/drivers/freedreno/ir3/ir3_shader.c
index 7205fb0d2745a99cd82feaec1f7e0be60e1fa1ed..59fb3ab5a5a822c330e3395b68aaf6e7b2dc8a89 100644 (file)
--- a/src/gallium/drivers/freedreno/ir3/ir3_shader.c
+++ b/src/gallium/drivers/freedreno/ir3/ir3_shader.c
@@ -248,6 +248,7 @@ ir3_shader_variant(struct ir3_shader *shader, struct ir3_shader_key key,
                        key.vsaturate_t = 0;
                        key.vsaturate_r = 0;
                        key.vastc_srgb = 0;
+                       key.vsamples = 0;
                }
                break;
        case SHADER_VERTEX:
@@ -259,6 +260,7 @@ ir3_shader_variant(struct ir3_shader *shader, struct ir3_shader_key key,
                        key.fsaturate_t = 0;
                        key.fsaturate_r = 0;
                        key.fastc_srgb = 0;
+                       key.fsamples = 0;
                }
                break;
        default:

diff --git a/src/gallium/drivers/freedreno/ir3/ir3_shader.h b/src/gallium/drivers/freedreno/ir3/ir3_shader.h
index 272368c69809604f96760dcb75d5deb7647c3d1a..acbeed567979ef20c376778fd454ef7395f04696 100644 (file)
--- a/src/gallium/drivers/freedreno/ir3/ir3_shader.h
+++ b/src/gallium/drivers/freedreno/ir3/ir3_shader.h
@@ -143,6 +143,9 @@ struct ir3_shader_key {
         */
        uint16_t fsaturate_s, fsaturate_t, fsaturate_r;
 
+       /* bitmask of ms shifts */
+       uint32_t vsamples, fsamples;
+
        /* bitmask of samplers which need astc srgb workaround: */
        uint16_t vastc_srgb, fastc_srgb;
 };
@@ -164,6 +167,7 @@ ir3_shader_key_changes_fs(struct ir3_shader_key *key, struct ir3_shader_key *las
                if ((last_key->fsaturate_s != key->fsaturate_s) ||
                                (last_key->fsaturate_t != key->fsaturate_t) ||
                                (last_key->fsaturate_r != key->fsaturate_r) ||
+                               (last_key->fsamples != key->fsamples) ||
                                (last_key->fastc_srgb != key->fastc_srgb))
                        return true;
        }
@@ -194,6 +198,7 @@ ir3_shader_key_changes_vs(struct ir3_shader_key *key, struct ir3_shader_key *las
                if ((last_key->vsaturate_s != key->vsaturate_s) ||
                                (last_key->vsaturate_t != key->vsaturate_t) ||
                                (last_key->vsaturate_r != key->vsaturate_r) ||
+                               (last_key->vsamples != key->vsamples) ||
                                (last_key->vastc_srgb != key->vastc_srgb))
                        return true;
        }