freedreno/a4xx: better workaround for astc+srgb
[mesa.git] / src / gallium / drivers / freedreno / ir3 / ir3_shader.c
index 7b5653322560c7f42589baeb6bcabd555974f96d..435a565e61ec1b8e9019084cc07dfcbb8bb94cad 100644 (file)
@@ -39,7 +39,7 @@
 
 #include "ir3_shader.h"
 #include "ir3_compiler.h"
-
+#include "ir3_nir.h"
 
 static void
 delete_variant(struct ir3_shader_variant *v)
@@ -127,14 +127,14 @@ void * ir3_shader_assemble(struct ir3_shader_variant *v, uint32_t gpu_id)
 static void
 assemble_variant(struct ir3_shader_variant *v)
 {
-       struct fd_context *ctx = fd_context(v->shader->pctx);
-       uint32_t gpu_id = v->shader->compiler->gpu_id;
+       struct ir3_compiler *compiler = v->shader->compiler;
+       uint32_t gpu_id = compiler->gpu_id;
        uint32_t sz, *bin;
 
        bin = ir3_shader_assemble(v, gpu_id);
        sz = v->info.sizedwords * 4;
 
-       v->bo = fd_bo_new(ctx->dev, sz,
+       v->bo = fd_bo_new(compiler->dev, sz,
                        DRM_FREEDRENO_GEM_CACHE_WCOMBINE |
                        DRM_FREEDRENO_GEM_TYPE_KMEM);
 
@@ -187,12 +187,6 @@ create_variant(struct ir3_shader *shader, struct ir3_shader_key key)
        v->key = key;
        v->type = shader->type;
 
-       if (fd_mesa_debug & FD_DBG_DISASM) {
-               DBG("dump tgsi: type=%d, k={bp=%u,cts=%u,hp=%u}", shader->type,
-                       key.binning_pass, key.color_two_side, key.half_precision);
-               tgsi_dump(shader->tokens, 0);
-       }
-
        ret = ir3_compile_shader_nir(shader->compiler, v);
        if (ret) {
                debug_error("compile failed!");
@@ -229,6 +223,7 @@ ir3_shader_variant(struct ir3_shader *shader, struct ir3_shader_key key)
                        key.vsaturate_s = 0;
                        key.vsaturate_t = 0;
                        key.vsaturate_r = 0;
+                       key.vastc_srgb = 0;
                }
                break;
        case SHADER_VERTEX:
@@ -239,6 +234,7 @@ ir3_shader_variant(struct ir3_shader *shader, struct ir3_shader_key key)
                        key.fsaturate_s = 0;
                        key.fsaturate_t = 0;
                        key.fsaturate_r = 0;
+                       key.fastc_srgb = 0;
                }
                break;
        }
@@ -267,28 +263,37 @@ ir3_shader_destroy(struct ir3_shader *shader)
                v = v->next;
                delete_variant(t);
        }
-       free((void *)shader->tokens);
+       ralloc_free(shader->nir);
        free(shader);
 }
 
 struct ir3_shader *
-ir3_shader_create(struct pipe_context *pctx,
+ir3_shader_create(struct ir3_compiler *compiler,
                const struct pipe_shader_state *cso,
                enum shader_t type)
 {
        struct ir3_shader *shader = CALLOC_STRUCT(ir3_shader);
-       shader->compiler = fd_context(pctx)->screen->compiler;
+       shader->compiler = compiler;
        shader->id = ++shader->compiler->shader_count;
-       shader->pctx = pctx;
        shader->type = type;
-       shader->tokens = tgsi_dup_tokens(cso->tokens);
+       if (fd_mesa_debug & FD_DBG_DISASM) {
+               DBG("dump tgsi: type=%d", shader->type);
+               tgsi_dump(cso->tokens, 0);
+       }
+       nir_shader *nir = ir3_tgsi_to_nir(cso->tokens);
+       /* do first pass optimization, ignoring the key: */
+       shader->nir = ir3_optimize_nir(shader, nir, NULL);
+       if (fd_mesa_debug & FD_DBG_DISASM) {
+               DBG("dump nir%d: type=%d", shader->id, shader->type);
+               nir_print_shader(shader->nir, stdout);
+       }
        shader->stream_output = cso->stream_output;
        if (fd_mesa_debug & FD_DBG_SHADERDB) {
                /* if shader-db run, create a standard variant immediately
                 * (as otherwise nothing will trigger the shader to be
                 * actually compiled)
                 */
-               static struct ir3_shader_key key = {};
+               static struct ir3_shader_key key = {0};
                ir3_shader_variant(shader, key);
        }
        return shader;
@@ -459,10 +464,9 @@ ir3_shader_disasm(struct ir3_shader_variant *so, uint32_t *bin)
 #include "freedreno_resource.h"
 
 static void
-emit_user_consts(struct ir3_shader_variant *v, struct fd_ringbuffer *ring,
-               struct fd_constbuf_stateobj *constbuf)
+emit_user_consts(struct fd_context *ctx, const struct ir3_shader_variant *v,
+               struct fd_ringbuffer *ring, struct fd_constbuf_stateobj *constbuf)
 {
-       struct fd_context *ctx = fd_context(v->shader->pctx);
        const unsigned index = 0;     /* user consts are index 0 */
        /* TODO save/restore dirty_mask for binning pass instead: */
        uint32_t dirty_mask = constbuf->enabled_mask;
@@ -498,12 +502,11 @@ emit_user_consts(struct ir3_shader_variant *v, struct fd_ringbuffer *ring,
 }
 
 static void
-emit_ubos(struct ir3_shader_variant *v, struct fd_ringbuffer *ring,
-               struct fd_constbuf_stateobj *constbuf)
+emit_ubos(struct fd_context *ctx, const struct ir3_shader_variant *v,
+               struct fd_ringbuffer *ring, struct fd_constbuf_stateobj *constbuf)
 {
        uint32_t offset = v->first_driver_param + IR3_UBOS_OFF;
        if (v->constlen > offset) {
-               struct fd_context *ctx = fd_context(v->shader->pctx);
                uint32_t params = MIN2(4, v->constlen - offset) * 4;
                uint32_t offsets[params];
                struct fd_bo *bos[params];
@@ -528,9 +531,9 @@ emit_ubos(struct ir3_shader_variant *v, struct fd_ringbuffer *ring,
 }
 
 static void
-emit_immediates(struct ir3_shader_variant *v, struct fd_ringbuffer *ring)
+emit_immediates(struct fd_context *ctx, const struct ir3_shader_variant *v,
+               struct fd_ringbuffer *ring)
 {
-       struct fd_context *ctx = fd_context(v->shader->pctx);
        int size = v->immediates_count;
        uint32_t base = v->first_immediate;
 
@@ -552,12 +555,12 @@ emit_immediates(struct ir3_shader_variant *v, struct fd_ringbuffer *ring)
 
 /* emit stream-out buffers: */
 static void
-emit_tfbos(struct ir3_shader_variant *v, struct fd_ringbuffer *ring)
+emit_tfbos(struct fd_context *ctx, const struct ir3_shader_variant *v,
+               struct fd_ringbuffer *ring)
 {
        /* streamout addresses after driver-params: */
        uint32_t offset = v->first_driver_param + IR3_TFBOS_OFF;
        if (v->constlen > offset) {
-               struct fd_context *ctx = fd_context(v->shader->pctx);
                struct fd_streamout_stateobj *so = &ctx->streamout;
                struct pipe_stream_output_info *info = &v->shader->stream_output;
                uint32_t params = 4;
@@ -583,9 +586,8 @@ emit_tfbos(struct ir3_shader_variant *v, struct fd_ringbuffer *ring)
 }
 
 static uint32_t
-max_tf_vtx(struct ir3_shader_variant *v)
+max_tf_vtx(struct fd_context *ctx, const struct ir3_shader_variant *v)
 {
-       struct fd_context *ctx = fd_context(v->shader->pctx);
        struct fd_streamout_stateobj *so = &ctx->streamout;
        struct pipe_stream_output_info *info = &v->shader->stream_output;
        uint32_t maxvtxcnt = 0x7fffffff;
@@ -629,11 +631,9 @@ max_tf_vtx(struct ir3_shader_variant *v)
 }
 
 void
-ir3_emit_consts(struct ir3_shader_variant *v, struct fd_ringbuffer *ring,
-               const struct pipe_draw_info *info, uint32_t dirty)
+ir3_emit_consts(const struct ir3_shader_variant *v, struct fd_ringbuffer *ring,
+               struct fd_context *ctx, const struct pipe_draw_info *info, uint32_t dirty)
 {
-       struct fd_context *ctx = fd_context(v->shader->pctx);
-
        if (dirty & (FD_DIRTY_PROG | FD_DIRTY_CONSTBUF)) {
                struct fd_constbuf_stateobj *constbuf;
                bool shader_dirty;
@@ -649,10 +649,10 @@ ir3_emit_consts(struct ir3_shader_variant *v, struct fd_ringbuffer *ring,
                        return;
                }
 
-               emit_user_consts(v, ring, constbuf);
-               emit_ubos(v, ring, constbuf);
+               emit_user_consts(ctx, v, ring, constbuf);
+               emit_ubos(ctx, v, ring, constbuf);
                if (shader_dirty)
-                       emit_immediates(v, ring);
+                       emit_immediates(ctx, v, ring);
        }
 
        /* emit driver params every time: */
@@ -663,7 +663,7 @@ ir3_emit_consts(struct ir3_shader_variant *v, struct fd_ringbuffer *ring,
                        uint32_t vertex_params[IR3_DP_COUNT] = {
                                [IR3_DP_VTXID_BASE] = info->indexed ?
                                                info->index_bias : info->start,
-                               [IR3_DP_VTXCNT_MAX] = max_tf_vtx(v),
+                               [IR3_DP_VTXCNT_MAX] = max_tf_vtx(ctx, v),
                        };
                        /* if no user-clip-planes, we don't need to emit the
                         * entire thing:
@@ -688,7 +688,7 @@ ir3_emit_consts(struct ir3_shader_variant *v, struct fd_ringbuffer *ring,
 
                        /* if needed, emit stream-out buffer addresses: */
                        if (vertex_params[IR3_DP_VTXCNT_MAX] > 0) {
-                               emit_tfbos(v, ring);
+                               emit_tfbos(ctx, v, ring);
                        }
                }
        }