r600g: eg+ support for FS_COLOR0_WRITES_ALL_CBUFS

Evergreen+ don't support multi-writes so we need to emulate
it in the shader. Fixes the following piglit tests:
fbo-drawbuffers-fragcolor
ati_draw_buffers-arbfp-no-option

Signed-off-by: Alex Deucher <alexdeucher@gmail.com>

diff --git a/src/gallium/drivers/r600/evergreen_state.c b/src/gallium/drivers/r600/evergreen_state.c
index dfe789649297f9a9bc660a5139850084d4449355..dc182611482c0d99d7e7cbb0a1bca27083e45d85 100644 (file)
--- a/src/gallium/drivers/r600/evergreen_state.c
+++ b/src/gallium/drivers/r600/evergreen_state.c
@@ -885,6 +885,7 @@ static void evergreen_set_framebuffer_state(struct pipe_context *ctx,
 
        /* build states */
        rctx->have_depth_fb = 0;
+       rctx->nr_cbufs = state->nr_cbufs;
        for (int i = 0; i < state->nr_cbufs; i++) {
                evergreen_cb(rctx, rstate, state, i);
        }
@@ -1631,7 +1632,10 @@ void evergreen_pipe_shader_ps(struct pipe_context *ctx, struct r600_pipe_shader
                    rshader->output[i].name == TGSI_SEMANTIC_STENCIL)
                        exports_ps |= 1;
                else if (rshader->output[i].name == TGSI_SEMANTIC_COLOR) {
-                       num_cout++;
+                       if (rshader->fs_write_all)
+                               num_cout = rshader->nr_cbufs;
+                       else
+                               num_cout++;
                }
        }
        exports_ps |= S_02884C_EXPORT_COLORS(num_cout);

diff --git a/src/gallium/drivers/r600/r600_asm.c b/src/gallium/drivers/r600/r600_asm.c
index c447a03106326b9b59b39e207f085f852eacbf43..065f955ebcbf26556c355e295858d8bf182ef92a 100644 (file)
--- a/src/gallium/drivers/r600/r600_asm.c
+++ b/src/gallium/drivers/r600/r600_asm.c
@@ -1863,6 +1863,8 @@ void r600_bc_dump(struct r600_bc *bc)
                        break;
                case V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_EXPORT:
                case V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_EXPORT_DONE:
+               case EG_V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_EXPORT:
+               case EG_V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_EXPORT_DONE:
                        fprintf(stderr, "%04d %08X EXPORT ", id, bc->bytecode[id]);
                        fprintf(stderr, "GPR:%X ", cf->output.gpr);
                        fprintf(stderr, "ELEM_SIZE:%X ", cf->output.elem_size);

diff --git a/src/gallium/drivers/r600/r600_pipe.h b/src/gallium/drivers/r600/r600_pipe.h
index dc9aad077825b20417db48ed040b86bd33314862..2667c80bcefbf47c51952469e1255db4de06bbaa 100644 (file)
--- a/src/gallium/drivers/r600/r600_pipe.h
+++ b/src/gallium/drivers/r600/r600_pipe.h
@@ -213,6 +213,7 @@ struct r600_pipe_context {
        boolean                         export_16bpc;
        unsigned                        alpha_ref;
        boolean                         alpha_ref_dirty;
+       unsigned                        nr_cbufs;
        struct r600_textures_info       ps_samplers;
 
        struct r600_pipe_fences         fences;

diff --git a/src/gallium/drivers/r600/r600_shader.c b/src/gallium/drivers/r600/r600_shader.c
index 2e5d4a6c49fb06aa412d5318d67b83fa33a90c14..f83d7079b298576bb7031a90b938d1794f18d912 100644 (file)
--- a/src/gallium/drivers/r600/r600_shader.c
+++ b/src/gallium/drivers/r600/r600_shader.c
@@ -606,7 +606,7 @@ static int r600_shader_from_tgsi(struct r600_pipe_context * rctx, struct r600_pi
        struct r600_bc_output output[32];
        unsigned output_done, noutput;
        unsigned opcode;
-       int i, r = 0, pos0;
+       int i, j, r = 0, pos0;
 
        ctx.bc = &shader->bc;
        ctx.shader = shader;
@@ -623,6 +623,8 @@ static int r600_shader_from_tgsi(struct r600_pipe_context * rctx, struct r600_pi
        shader->clamp_color = (((ctx.type == TGSI_PROCESSOR_FRAGMENT) && rctx->clamp_fragment_color) ||
                ((ctx.type == TGSI_PROCESSOR_VERTEX) && rctx->clamp_vertex_color));
 
+       shader->nr_cbufs = rctx->nr_cbufs;
+
        /* register allocations */
        /* Values [0,127] correspond to GPR[0..127].
         * Values [128,159] correspond to constant buffer bank 0
@@ -767,50 +769,68 @@ static int r600_shader_from_tgsi(struct r600_pipe_context * rctx, struct r600_pi
        }
 
        /* export output */
+       j = 0;
        for (i = 0, pos0 = 0; i < noutput; i++) {
                memset(&output[i], 0, sizeof(struct r600_bc_output));
-               output[i].gpr = shader->output[i].gpr;
-               output[i].elem_size = 3;
-               output[i].swizzle_x = 0;
-               output[i].swizzle_y = 1;
-               output[i].swizzle_z = 2;
-               output[i].swizzle_w = 3;
-               output[i].burst_count = 1;
-               output[i].barrier = 1;
-               output[i].type = V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PARAM;
-               output[i].array_base = i - pos0;
-               output[i].inst = BC_INST(ctx.bc, V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_EXPORT);
+               output[i + j].gpr = shader->output[i].gpr;
+               output[i + j].elem_size = 3;
+               output[i + j].swizzle_x = 0;
+               output[i + j].swizzle_y = 1;
+               output[i + j].swizzle_z = 2;
+               output[i + j].swizzle_w = 3;
+               output[i + j].burst_count = 1;
+               output[i + j].barrier = 1;
+               output[i + j].type = V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PARAM;
+               output[i + j].array_base = i - pos0;
+               output[i + j].inst = BC_INST(ctx.bc, V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_EXPORT);
                switch (ctx.type) {
                case TGSI_PROCESSOR_VERTEX:
                        if (shader->output[i].name == TGSI_SEMANTIC_POSITION) {
-                               output[i].array_base = 60;
-                               output[i].type = V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_POS;
+                               output[i + j].array_base = 60;
+                               output[i + j].type = V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_POS;
                                /* position doesn't count in array_base */
                                pos0++;
                        }
                        if (shader->output[i].name == TGSI_SEMANTIC_PSIZE) {
-                               output[i].array_base = 61;
-                               output[i].type = V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_POS;
+                               output[i + j].array_base = 61;
+                               output[i + j].type = V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_POS;
                                /* position doesn't count in array_base */
                                pos0++;
                        }
                        break;
                case TGSI_PROCESSOR_FRAGMENT:
                        if (shader->output[i].name == TGSI_SEMANTIC_COLOR) {
-                               output[i].array_base = shader->output[i].sid;
-                               output[i].type = V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PIXEL;
+                               output[i + j].array_base = shader->output[i].sid;
+                               output[i + j].type = V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PIXEL;
+                               if (shader->fs_write_all && (shader->family >= CHIP_CEDAR)) {
+                                       for (j = 1; j < shader->nr_cbufs; j++) {
+                                               memset(&output[i + j], 0, sizeof(struct r600_bc_output));
+                                               output[i + j].gpr = shader->output[i].gpr;
+                                               output[i + j].elem_size = 3;
+                                               output[i + j].swizzle_x = 0;
+                                               output[i + j].swizzle_y = 1;
+                                               output[i + j].swizzle_z = 2;
+                                               output[i + j].swizzle_w = 3;
+                                               output[i + j].burst_count = 1;
+                                               output[i + j].barrier = 1;
+                                               output[i + j].array_base = shader->output[i].sid + j;
+                                               output[i + j].inst = BC_INST(ctx.bc, V_SQ_CF_ALLOC_EXPORT_WORD1_SQ_CF_INST_EXPORT);
+                                               output[i + j].type = V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PIXEL;
+                                       }
+                                       j--;
+                               }
                        } else if (shader->output[i].name == TGSI_SEMANTIC_POSITION) {
-                               output[i].array_base = 61;
-                               output[i].swizzle_x = 2;
-                               output[i].swizzle_y = 7;
-                               output[i].swizzle_z = output[i].swizzle_w = 7;
-                               output[i].type = V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PIXEL;
+                               output[i + j].array_base = 61;
+                               output[i + j].swizzle_x = 2;
+                               output[i + j].swizzle_y = 7;
+                               output[i + j].swizzle_z = output[i + j].swizzle_w = 7;
+                               output[i + j].type = V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PIXEL;
                        } else if (shader->output[i].name == TGSI_SEMANTIC_STENCIL) {
-                               output[i].array_base = 61;
-                               output[i].swizzle_x = 7;
-                               output[i].swizzle_y = 1;
-                               output[i].swizzle_z = output[i].swizzle_w = 7;
-                               output[i].type = V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PIXEL;
+                               output[i + j].array_base = 61;
+                               output[i + j].swizzle_x = 7;
+                               output[i + j].swizzle_y = 1;
+                               output[i + j].swizzle_z = output[i + j].swizzle_w = 7;
+                               output[i + j].type = V_SQ_CF_ALLOC_EXPORT_WORD0_SQ_EXPORT_PIXEL;
                        } else {
                                R600_ERR("unsupported fragment output name %d\n", shader->output[i].name);
                                r = -EINVAL;
@@ -823,6 +843,7 @@ static int r600_shader_from_tgsi(struct r600_pipe_context * rctx, struct r600_pi
                        goto out_err;
                }
        }
+       noutput += j;
        /* add fake param output for vertex shader if no param is exported */
        if (ctx.type == TGSI_PROCESSOR_VERTEX) {
                for (i = 0, pos0 = 0; i < noutput; i++) {

diff --git a/src/gallium/drivers/r600/r600_shader.h b/src/gallium/drivers/r600/r600_shader.h
index c1e00f3b662845699d747ae955da078d38484fa7..76aebf2b1eab02b535ac951af2f6ed83846af8fb 100644 (file)
--- a/src/gallium/drivers/r600/r600_shader.h
+++ b/src/gallium/drivers/r600/r600_shader.h
@@ -47,6 +47,7 @@ struct r600_shader {
        boolean                 uses_kill;
        boolean                 fs_write_all;
        boolean                 clamp_color;
+       unsigned                nr_cbufs;
 };
 
 #endif

diff --git a/src/gallium/drivers/r600/r600_state_common.c b/src/gallium/drivers/r600/r600_state_common.c
index f1b0fc32b53fbdac81baa86c40a7008f111d3f70..bf4e90ecb2fc0c52576611f9cc14c880a608d0d1 100644 (file)
--- a/src/gallium/drivers/r600/r600_state_common.c
+++ b/src/gallium/drivers/r600/r600_state_common.c
@@ -599,7 +599,9 @@ void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *info)
        if (rctx->vs_shader->shader.clamp_color != rctx->clamp_vertex_color)
                r600_shader_rebuild(ctx, rctx->vs_shader);
 
-       if (rctx->ps_shader->shader.clamp_color != rctx->clamp_fragment_color)
+       if ((rctx->ps_shader->shader.clamp_color != rctx->clamp_fragment_color) ||
+           (rctx->ps_shader->shader.fs_write_all &&
+            (rctx->ps_shader->shader.nr_cbufs != rctx->nr_cbufs)))
                r600_shader_rebuild(ctx, rctx->ps_shader);
 
        if (rctx->spi_dirty)