X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fgallium%2Fdrivers%2Fsoftpipe%2Fsp_fs_sse.c;h=c873af125bd94c1d9a32d97df7c87d1f3cf240a7;hb=6b3bbf52b884ef4b5f0049623ec7154dd3c1dc31;hp=f4fa0905d7410a48e54f23a5f193e04181a7cc1f;hpb=7549a8397b310acf672f97a08c8e7d866cdf492c;p=mesa.git diff --git a/src/gallium/drivers/softpipe/sp_fs_sse.c b/src/gallium/drivers/softpipe/sp_fs_sse.c index f4fa0905d74..c873af125bd 100644 --- a/src/gallium/drivers/softpipe/sp_fs_sse.c +++ b/src/gallium/drivers/softpipe/sp_fs_sse.c @@ -48,11 +48,11 @@ /** - * Subclass of sp_fragment_shader + * Subclass of sp_fragment_shader_variant */ struct sp_sse_fragment_shader { - struct sp_fragment_shader base; + struct sp_fragment_shader_variant base; struct x86_function sse2_program; tgsi_sse2_fs_function func; float immediates[TGSI_EXEC_NUM_IMMEDIATES][4]; @@ -61,14 +61,14 @@ struct sp_sse_fragment_shader /** cast wrapper */ static INLINE struct sp_sse_fragment_shader * -sp_sse_fragment_shader(const struct sp_fragment_shader *base) +sp_sse_fragment_shader(const struct sp_fragment_shader_variant *base) { return (struct sp_sse_fragment_shader *) base; } static void -fs_sse_prepare( const struct sp_fragment_shader *base, +fs_sse_prepare( const struct sp_fragment_shader_variant *base, struct tgsi_exec_machine *machine, struct tgsi_sampler **samplers ) { @@ -76,40 +76,120 @@ fs_sse_prepare( const struct sp_fragment_shader *base, } + +/** + * Compute quad X,Y,Z,W for the four fragments in a quad. + * + * This should really be part of the compiled shader. + */ +static void +setup_pos_vector(const struct tgsi_interp_coef *coef, + float x, float y, + struct tgsi_exec_vector *quadpos) +{ + uint chan; + /* do X */ + quadpos->xyzw[0].f[0] = x; + quadpos->xyzw[0].f[1] = x + 1; + quadpos->xyzw[0].f[2] = x; + quadpos->xyzw[0].f[3] = x + 1; + + /* do Y */ + quadpos->xyzw[1].f[0] = y; + quadpos->xyzw[1].f[1] = y; + quadpos->xyzw[1].f[2] = y + 1; + quadpos->xyzw[1].f[3] = y + 1; + + /* do Z and W for all fragments in the quad */ + for (chan = 2; chan < 4; chan++) { + const float dadx = coef->dadx[chan]; + const float dady = coef->dady[chan]; + const float a0 = coef->a0[chan] + dadx * x + dady * y; + quadpos->xyzw[chan].f[0] = a0; + quadpos->xyzw[chan].f[1] = a0 + dadx; + quadpos->xyzw[chan].f[2] = a0 + dady; + quadpos->xyzw[chan].f[3] = a0 + dadx + dady; + } +} + + /* TODO: codegenerate the whole run function, skip this wrapper. * TODO: break dependency on tgsi_exec_machine struct * TODO: push Position calculation into the generated shader * TODO: process >1 quad at a time */ static unsigned -fs_sse_run( const struct sp_fragment_shader *base, +fs_sse_run( const struct sp_fragment_shader_variant *base, struct tgsi_exec_machine *machine, struct quad_header *quad ) { struct sp_sse_fragment_shader *shader = sp_sse_fragment_shader(base); /* Compute X, Y, Z, W vals for this quad -- place in temp[0] for now */ - sp_setup_pos_vector(quad->posCoef, - (float)quad->input.x0, (float)quad->input.y0, - machine->Temps); + setup_pos_vector(quad->posCoef, + (float)quad->input.x0, (float)quad->input.y0, + machine->Temps); /* init kill mask */ tgsi_set_kill_mask(machine, 0x0); tgsi_set_exec_mask(machine, 1, 1, 1, 1); shader->func( machine, - machine->Consts, + (const float (*)[4])machine->Consts[0], (const float (*)[4])shader->immediates, machine->InterpCoefs - // , &machine->QuadPos + /*, &machine->QuadPos*/ ); - return ~(machine->Temps[TGSI_EXEC_TEMP_KILMASK_I].xyzw[TGSI_EXEC_TEMP_KILMASK_C].u[0]); + quad->inout.mask &= ~(machine->Temps[TGSI_EXEC_TEMP_KILMASK_I].xyzw[TGSI_EXEC_TEMP_KILMASK_C].u[0]); + if (quad->inout.mask == 0) + return FALSE; + + /* store outputs */ + { + const ubyte *sem_name = base->info.output_semantic_name; + const ubyte *sem_index = base->info.output_semantic_index; + const uint n = base->info.num_outputs; + uint i; + for (i = 0; i < n; i++) { + switch (sem_name[i]) { + case TGSI_SEMANTIC_COLOR: + { + uint cbuf = sem_index[i]; + + assert(sizeof(quad->output.color[cbuf]) == + sizeof(machine->Outputs[i])); + + /* copy float[4][4] result */ + memcpy(quad->output.color[cbuf], + &machine->Outputs[i], + sizeof(quad->output.color[0]) ); + } + break; + case TGSI_SEMANTIC_POSITION: + { + uint j; + for (j = 0; j < 4; j++) + quad->output.depth[j] = machine->Outputs[i].xyzw[2].f[j]; + } + break; + case TGSI_SEMANTIC_STENCIL: + { + uint j; + for (j = 0; j < 4; j++) + quad->output.stencil[j] = machine->Outputs[i].xyzw[1].f[j]; + } + break; + } + } + } + + return TRUE; } static void -fs_sse_delete( struct sp_fragment_shader *base ) +fs_sse_delete( struct sp_fragment_shader_variant *base ) { struct sp_sse_fragment_shader *shader = sp_sse_fragment_shader(base); @@ -118,9 +198,9 @@ fs_sse_delete( struct sp_fragment_shader *base ) } -struct sp_fragment_shader * -softpipe_create_fs_sse(struct softpipe_context *softpipe, - const struct pipe_shader_state *templ) +struct sp_fragment_shader_variant * +softpipe_create_fs_variant_sse(struct softpipe_context *softpipe, + const struct pipe_shader_state *templ) { struct sp_sse_fragment_shader *shader; @@ -146,7 +226,6 @@ softpipe_create_fs_sse(struct softpipe_context *softpipe, return NULL; } - shader->base.shader.tokens = NULL; /* don't hold reference to templ->tokens */ shader->base.prepare = fs_sse_prepare; shader->base.run = fs_sse_run; shader->base.delete = fs_sse_delete; @@ -159,9 +238,9 @@ softpipe_create_fs_sse(struct softpipe_context *softpipe, /* Maybe put this variant in the header file. */ -struct sp_fragment_shader * -softpipe_create_fs_sse(struct softpipe_context *softpipe, - const struct pipe_shader_state *templ) +struct sp_fragment_shader_variant * +softpipe_create_fs_variant_sse(struct softpipe_context *softpipe, + const struct pipe_shader_state *templ) { return NULL; }