/**
* Make simple fragment color pass-through shader that replicates OUT[0]
* to all bound colorbuffers.
*/
void *
util_make_fragment_passthrough_shader(struct pipe_context *pipe,
int input_semantic,
int input_interpolate,
boolean write_all_cbufs)
{
static const char shader_templ[] =
"FRAG\n"
"%s"
"DCL IN[0], %s[0], %s\n"
"DCL OUT[0], COLOR[0]\n"
"MOV OUT[0], IN[0]\n"
"END\n";
char text[sizeof(shader_templ)+100];
struct tgsi_token tokens[1000];
struct pipe_shader_state state;
sprintf(text, shader_templ,
write_all_cbufs ? "PROPERTY FS_COLOR0_WRITES_ALL_CBUFS 1\n" : "",
tgsi_semantic_names[input_semantic],
tgsi_interpolate_names[input_interpolate]);
if (!tgsi_text_translate(text, tokens, ARRAY_SIZE(tokens))) {
assert(0);
return NULL;
}
pipe_shader_state_from_tgsi(&state, tokens);
#if 0
tgsi_dump(state.tokens, 0);
#endif
return pipe->create_fs_state(pipe, &state);
}
/**
* Takes position and color, and outputs position, color, and instance id.
*/
void *util_make_layered_clear_helper_vertex_shader(struct pipe_context *pipe)
{
static const char text[] =
"VERT\n"
"DCL IN[0]\n"
"DCL IN[1]\n"
"DCL SV[0], INSTANCEID\n"
"DCL OUT[0], POSITION\n"
"DCL OUT[1], GENERIC[0]\n"
"DCL OUT[2], GENERIC[1]\n"
"MOV OUT[0], IN[0]\n"
"MOV OUT[1], IN[1]\n"
"MOV OUT[2].x, SV[0].xxxx\n"
"END\n";
struct tgsi_token tokens[1000];
struct pipe_shader_state state;
if (!tgsi_text_translate(text, tokens, ARRAY_SIZE(tokens))) {
assert(0);
return NULL;
}
pipe_shader_state_from_tgsi(&state, tokens);
return pipe->create_vs_state(pipe, &state);
}
void
util_test_constant_buffer(struct pipe_context *ctx,
struct pipe_resource *constbuf)
{
struct cso_context *cso;
struct pipe_resource *cb;
void *fs, *vs;
bool pass = true;
static const float zero[] = {0, 0, 0, 0};
cso = cso_create_context(ctx, 0);
cb = util_create_texture2d(ctx->screen, 256, 256,
PIPE_FORMAT_R8G8B8A8_UNORM, 0);
util_set_common_states_and_clear(cso, ctx, cb);
pipe_set_constant_buffer(ctx, PIPE_SHADER_FRAGMENT, 0, constbuf);
/* Fragment shader. */
{
static const char *text = /* I don't like ureg... */
"FRAG\n"
"DCL CONST[0][0]\n"
"DCL OUT[0], COLOR\n"
"MOV OUT[0], CONST[0][0]\n"
"END\n";
struct tgsi_token tokens[1000];
struct pipe_shader_state state;
if (!tgsi_text_translate(text, tokens, ARRAY_SIZE(tokens))) {
puts("Can't compile a fragment shader.");
util_report_result(FAIL);
return;
}
pipe_shader_state_from_tgsi(&state, tokens);
fs = ctx->create_fs_state(ctx, &state);
cso_set_fragment_shader_handle(cso, fs);
}
/* Vertex shader. */
vs = util_set_passthrough_vertex_shader(cso, ctx, false);
util_draw_fullscreen_quad(cso);
/* Probe pixels. */
pass = pass && util_probe_rect_rgba(ctx, cb, 0, 0, cb->width0,
cb->height0, zero);
/* Cleanup. */
cso_destroy_context(cso);
ctx->delete_vs_state(ctx, vs);
ctx->delete_fs_state(ctx, fs);
pipe_resource_reference(&cb, NULL);
util_report_result(pass);
}
int main(int argc, char **argv)
{
if(argc < 2)
{
fprintf(stderr, "Must pass shader source name as argument\n");
exit(1);
}
const char *text = load_text_file(argv[1]);
if(!text)
{
fprintf(stderr, "Unable to open %s\n", argv[1]);
exit(1);
}
struct tgsi_token tokens[1024];
if(tgsi_text_translate(text, tokens, Elements(tokens)))
{
tgsi_dump(tokens, 0);
union {
struct tgsi_header h;
struct tgsi_token t;
} hdr;
hdr.t = tokens[0];
printf("Header size: %i\n", hdr.h.HeaderSize);
printf("Body size: %i\n", hdr.h.BodySize);
int totalSize = hdr.h.HeaderSize + hdr.h.BodySize;
for(int i=0; i<totalSize; ++i)
{
printf("%08x ", *((uint32_t*)&tokens[i]));
}
printf("\n");
/* Parsing test */
struct tgsi_parse_context ctx = {};
unsigned status = TGSI_PARSE_OK;
status = tgsi_parse_init(&ctx, tokens);
assert(status == TGSI_PARSE_OK);
while(!tgsi_parse_end_of_tokens(&ctx))
{
tgsi_parse_token(&ctx);
/* declaration / immediate / instruction / property */
printf("Parsed token! %i\n", ctx.FullToken.Token.Type);
}
tgsi_parse_free(&ctx);
} else {
fprintf(stderr, "Unable to parse %s\n", argv[1]);
}
return 0;
}
static const struct tgsi_token *substitute_vs(
unsigned shader_id,
const struct tgsi_token *old_tokens )
{
#if 0
if (shader_id == 12) {
static struct tgsi_token tokens[300];
const char *text =
"VERT1.1\n"
"DCL IN[0]\n"
"DCL IN[1]\n"
"DCL IN[2]\n"
"DCL OUT[0], POSITION\n"
"DCL TEMP[0..4]\n"
"IMM FLT32 { 1.0000, 1.0000, 1.0000, 1.0000 }\n"
"IMM FLT32 { 0.45, 1.0000, 1.0000, 1.0000 }\n"
"IMM FLT32 { 1.297863, 0.039245, 0.035993, 0.035976}\n"
"IMM FLT32 { -0.019398, 1.696131, -0.202151, -0.202050 }\n"
"IMM FLT32 { 0.051711, -0.348713, -0.979204, -0.978714 }\n"
"IMM FLT32 { 0.000000, 0.000003, 139.491577, 141.421356 }\n"
"DCL CONST[0..7]\n"
"DCL CONST[9..16]\n"
" MOV TEMP[2], IMM[0]\n"
" MOV TEMP[2].xyz, IN[2]\n"
" MOV TEMP[2].xyz, IN[0]\n"
" MOV TEMP[2].xyz, IN[1]\n"
" MUL TEMP[1], IMM[3], TEMP[2].yyyy\n"
" MAD TEMP[3], IMM[2], TEMP[2].xxxx, TEMP[1]\n"
" MAD TEMP[1], IMM[4], TEMP[2].zzzz, TEMP[3]\n"
" MAD TEMP[4], IMM[5], TEMP[2].wwww, TEMP[1]\n"
" MOV OUT[0], TEMP[4]\n"
" END\n";
if (!tgsi_text_translate( text,
tokens,
Elements(tokens) ))
{
assert(0);
return NULL;
}
return tokens;
}
#endif
return old_tokens;
}
static INLINE struct tgsi_token *tokens_from_assembly(const char *txt, int num_tokens)
{
struct tgsi_token *tokens;
tokens = (struct tgsi_token *) MALLOC(num_tokens * sizeof(tokens[0]));
tgsi_text_translate(txt, tokens, num_tokens);
#if DEBUG_SHADERS
tgsi_dump(tokens, 0);
#endif
return tokens;
}
static void * assemble_tgsi(struct pipe_context *pctx,
const char *src, bool frag)
{
struct tgsi_token toks[32];
struct pipe_shader_state cso = {
.tokens = toks,
};
tgsi_text_translate(src, toks, ARRAY_SIZE(toks));
if (frag)
return pctx->create_fs_state(pctx, &cso);
else
return pctx->create_vs_state(pctx, &cso);
}
static void *
util_make_fs_blit_msaa_gen(struct pipe_context *pipe,
enum tgsi_texture_type tgsi_tex,
const char *samp_type,
const char *output_semantic,
const char *output_mask,
const char *conversion_decl,
const char *conversion)
{
static const char shader_templ[] =
"FRAG\n"
"DCL IN[0], GENERIC[0], LINEAR\n"
"DCL SAMP[0]\n"
"DCL SVIEW[0], %s, %s\n"
"DCL OUT[0], %s\n"
"DCL TEMP[0]\n"
"%s"
"F2U TEMP[0], IN[0]\n"
"TXF TEMP[0], TEMP[0], SAMP[0], %s\n"
"%s"
"MOV OUT[0]%s, TEMP[0]\n"
"END\n";
const char *type = tgsi_texture_names[tgsi_tex];
char text[sizeof(shader_templ)+100];
struct tgsi_token tokens[1000];
struct pipe_shader_state state;
assert(tgsi_tex == TGSI_TEXTURE_2D_MSAA ||
tgsi_tex == TGSI_TEXTURE_2D_ARRAY_MSAA);
util_snprintf(text, sizeof(text), shader_templ, type, samp_type,
output_semantic, conversion_decl, type, conversion, output_mask);
if (!tgsi_text_translate(text, tokens, ARRAY_SIZE(tokens))) {
puts(text);
assert(0);
return NULL;
}
pipe_shader_state_from_tgsi(&state, tokens);
#if 0
tgsi_dump(state.tokens, 0);
#endif
return pipe->create_fs_state(pipe, &state);
}
/**
* Convert the TGSI assembly to a runnable shader.
*
* We need not care about geometry shaders. All we have is screen quads.
*/
void *
pp_tgsi_to_state(struct pipe_context *pipe, const char *text, bool isvs,
const char *name)
{
struct pipe_shader_state state;
struct tgsi_token tokens[PP_MAX_TOKENS];
if (tgsi_text_translate(text, tokens, Elements(tokens)) == FALSE) {
pp_debug("Failed to translate %s\n", name);
return NULL;
}
state.tokens = tokens;
memset(&state.stream_output, 0, sizeof(state.stream_output));
if (isvs)
return pipe->create_vs_state(pipe, &state);
else
return pipe->create_fs_state(pipe, &state);
}
/**
* Make a fragment shader that sets the output depth and stencil to depth
* and stencil values fetched from two multisample textures / samplers.
* The sizes of both textures should match (it should be one depth-stencil
* texture).
* \param tex_target one of PIPE_TEXTURE_x
*/
void *
util_make_fs_blit_msaa_depthstencil(struct pipe_context *pipe,
enum tgsi_texture_type tgsi_tex)
{
static const char shader_templ[] =
"FRAG\n"
"DCL IN[0], GENERIC[0], LINEAR\n"
"DCL SAMP[0..1]\n"
"DCL SVIEW[0..1], %s, FLOAT\n"
"DCL OUT[0], POSITION\n"
"DCL OUT[1], STENCIL\n"
"DCL TEMP[0]\n"
"F2U TEMP[0], IN[0]\n"
"TXF OUT[0].z, TEMP[0], SAMP[0], %s\n"
"TXF OUT[1].y, TEMP[0], SAMP[1], %s\n"
"END\n";
const char *type = tgsi_texture_names[tgsi_tex];
char text[sizeof(shader_templ)+100];
struct tgsi_token tokens[1000];
struct pipe_shader_state state;
assert(tgsi_tex == TGSI_TEXTURE_2D_MSAA ||
tgsi_tex == TGSI_TEXTURE_2D_ARRAY_MSAA);
sprintf(text, shader_templ, type, type, type);
if (!tgsi_text_translate(text, tokens, ARRAY_SIZE(tokens))) {
assert(0);
return NULL;
}
pipe_shader_state_from_tgsi(&state, tokens);
#if 0
tgsi_dump(state.tokens, 0);
#endif
return pipe->create_fs_state(pipe, &state);
}
/**
* Takes position, color, and target layer, and emits vertices on that target
* layer, with the specified color.
*/
void *util_make_layered_clear_geometry_shader(struct pipe_context *pipe)
{
static const char text[] =
"GEOM\n"
"PROPERTY GS_INPUT_PRIMITIVE TRIANGLES\n"
"PROPERTY GS_OUTPUT_PRIMITIVE TRIANGLE_STRIP\n"
"PROPERTY GS_MAX_OUTPUT_VERTICES 3\n"
"PROPERTY GS_INVOCATIONS 1\n"
"DCL IN[][0], POSITION\n" /* position */
"DCL IN[][1], GENERIC[0]\n" /* color */
"DCL IN[][2], GENERIC[1]\n" /* vs invocation */
"DCL OUT[0], POSITION\n"
"DCL OUT[1], GENERIC[0]\n"
"DCL OUT[2], LAYER\n"
"IMM[0] INT32 {0, 0, 0, 0}\n"
"MOV OUT[0], IN[0][0]\n"
"MOV OUT[1], IN[0][1]\n"
"MOV OUT[2].x, IN[0][2].xxxx\n"
"EMIT IMM[0].xxxx\n"
"MOV OUT[0], IN[1][0]\n"
"MOV OUT[1], IN[1][1]\n"
"MOV OUT[2].x, IN[1][2].xxxx\n"
"EMIT IMM[0].xxxx\n"
"MOV OUT[0], IN[2][0]\n"
"MOV OUT[1], IN[2][1]\n"
"MOV OUT[2].x, IN[2][2].xxxx\n"
"EMIT IMM[0].xxxx\n"
"END\n";
struct tgsi_token tokens[1000];
struct pipe_shader_state state;
if (!tgsi_text_translate(text, tokens, ARRAY_SIZE(tokens))) {
assert(0);
return NULL;
}
pipe_shader_state_from_tgsi(&state, tokens);
return pipe->create_gs_state(pipe, &state);
}
static void *
util_make_fs_blit_msaa_gen(struct pipe_context *pipe,
unsigned tgsi_tex,
const char *output_semantic,
const char *output_mask)
{
static const char shader_templ[] =
"FRAG\n"
"DCL IN[0], GENERIC[0], LINEAR\n"
"DCL SAMP[0]\n"
"DCL OUT[0], %s\n"
"DCL TEMP[0]\n"
"F2U TEMP[0], IN[0]\n"
"TXF OUT[0]%s, TEMP[0], SAMP[0], %s\n"
"END\n";
const char *type = tgsi_texture_names[tgsi_tex];
char text[sizeof(shader_templ)+100];
struct tgsi_token tokens[1000];
struct pipe_shader_state state = {tokens};
assert(tgsi_tex == TGSI_TEXTURE_2D_MSAA ||
tgsi_tex == TGSI_TEXTURE_2D_ARRAY_MSAA);
sprintf(text, shader_templ, output_semantic, output_mask, type);
if (!tgsi_text_translate(text, tokens, Elements(tokens))) {
puts(text);
assert(0);
return NULL;
}
#if 0
tgsi_dump(state.tokens, 0);
#endif
return pipe->create_fs_state(pipe, &state);
}
/**
* Convert the TGSI assembly to a runnable shader.
*
* We need not care about geometry shaders. All we have is screen quads.
*/
void *
pp_tgsi_to_state(struct pipe_context *pipe, const char *text, bool isvs,
const char *name)
{
struct pipe_shader_state state;
struct tgsi_token *tokens = NULL;
void *ret_state = NULL;
/*
* Allocate temporary token storage. State creation will duplicate
* tokens so we must free them on exit.
*/
tokens = tgsi_alloc_tokens(PP_MAX_TOKENS);
if (!tokens) {
pp_debug("Failed to allocate temporary token storage.\n");
return NULL;
}
if (tgsi_text_translate(text, tokens, PP_MAX_TOKENS) == FALSE) {
_debug_printf("pp: Failed to translate a shader for %s\n", name);
return NULL;
}
state.tokens = tokens;
memset(&state.stream_output, 0, sizeof(state.stream_output));
if (isvs) {
ret_state = pipe->create_vs_state(pipe, &state);
FREE(tokens);
} else {
ret_state = pipe->create_fs_state(pipe, &state);
FREE(tokens);
}
return ret_state;
}
/* Create the compute shader that is used to collect the results.
*
* One compute grid with a single thread is launched for every query result
* buffer. The thread (optionally) reads a previous summary buffer, then
* accumulates data from the query result buffer, and writes the result either
* to a summary buffer to be consumed by the next grid invocation or to the
* user-supplied buffer.
*
* Data layout:
*
* CONST
* 0.x = end_offset
* 0.y = result_stride
* 0.z = result_count
* 0.w = bit field:
* 1: read previously accumulated values
* 2: write accumulated values for chaining
* 4: write result available
* 8: convert result to boolean (0/1)
* 16: only read one dword and use that as result
* 32: apply timestamp conversion
* 64: store full 64 bits result
* 128: store signed 32 bits result
* 256: SO_OVERFLOW mode: take the difference of two successive half-pairs
* 1.x = fence_offset
* 1.y = pair_stride
* 1.z = pair_count
*
* BUFFER[0] = query result buffer
* BUFFER[1] = previous summary buffer
* BUFFER[2] = next summary buffer or user-supplied buffer
*/
void *si_create_query_result_cs(struct si_context *sctx)
{
/* TEMP[0].xy = accumulated result so far
* TEMP[0].z = result not available
*
* TEMP[1].x = current result index
* TEMP[1].y = current pair index
*/
static const char text_tmpl[] =
"COMP\n"
"PROPERTY CS_FIXED_BLOCK_WIDTH 1\n"
"PROPERTY CS_FIXED_BLOCK_HEIGHT 1\n"
"PROPERTY CS_FIXED_BLOCK_DEPTH 1\n"
"DCL BUFFER[0]\n"
"DCL BUFFER[1]\n"
"DCL BUFFER[2]\n"
"DCL CONST[0][0..1]\n"
"DCL TEMP[0..5]\n"
"IMM[0] UINT32 {0, 31, 2147483647, 4294967295}\n"
"IMM[1] UINT32 {1, 2, 4, 8}\n"
"IMM[2] UINT32 {16, 32, 64, 128}\n"
"IMM[3] UINT32 {1000000, 0, %u, 0}\n" /* for timestamp conversion */
"IMM[4] UINT32 {256, 0, 0, 0}\n"
"AND TEMP[5], CONST[0][0].wwww, IMM[2].xxxx\n"
"UIF TEMP[5]\n"
/* Check result availability. */
"LOAD TEMP[1].x, BUFFER[0], CONST[0][1].xxxx\n"
"ISHR TEMP[0].z, TEMP[1].xxxx, IMM[0].yyyy\n"
"MOV TEMP[1], TEMP[0].zzzz\n"
"NOT TEMP[0].z, TEMP[0].zzzz\n"
/* Load result if available. */
"UIF TEMP[1]\n"
"LOAD TEMP[0].xy, BUFFER[0], IMM[0].xxxx\n"
"ENDIF\n"
"ELSE\n"
/* Load previously accumulated result if requested. */
"MOV TEMP[0], IMM[0].xxxx\n"
"AND TEMP[4], CONST[0][0].wwww, IMM[1].xxxx\n"
"UIF TEMP[4]\n"
"LOAD TEMP[0].xyz, BUFFER[1], IMM[0].xxxx\n"
"ENDIF\n"
"MOV TEMP[1].x, IMM[0].xxxx\n"
"BGNLOOP\n"
/* Break if accumulated result so far is not available. */
"UIF TEMP[0].zzzz\n"
"BRK\n"
"ENDIF\n"
/* Break if result_index >= result_count. */
"USGE TEMP[5], TEMP[1].xxxx, CONST[0][0].zzzz\n"
"UIF TEMP[5]\n"
"BRK\n"
"ENDIF\n"
/* Load fence and check result availability */
"UMAD TEMP[5].x, TEMP[1].xxxx, CONST[0][0].yyyy, CONST[0][1].xxxx\n"
"LOAD TEMP[5].x, BUFFER[0], TEMP[5].xxxx\n"
"ISHR TEMP[0].z, TEMP[5].xxxx, IMM[0].yyyy\n"
"NOT TEMP[0].z, TEMP[0].zzzz\n"
"UIF TEMP[0].zzzz\n"
"BRK\n"
"ENDIF\n"
"MOV TEMP[1].y, IMM[0].xxxx\n"
"BGNLOOP\n"
/* Load start and end. */
"UMUL TEMP[5].x, TEMP[1].xxxx, CONST[0][0].yyyy\n"
"UMAD TEMP[5].x, TEMP[1].yyyy, CONST[0][1].yyyy, TEMP[5].xxxx\n"
"LOAD TEMP[2].xy, BUFFER[0], TEMP[5].xxxx\n"
"UADD TEMP[5].y, TEMP[5].xxxx, CONST[0][0].xxxx\n"
"LOAD TEMP[3].xy, BUFFER[0], TEMP[5].yyyy\n"
"U64ADD TEMP[4].xy, TEMP[3], -TEMP[2]\n"
"AND TEMP[5].z, CONST[0][0].wwww, IMM[4].xxxx\n"
"UIF TEMP[5].zzzz\n"
/* Load second start/end half-pair and
* take the difference
*/
"UADD TEMP[5].xy, TEMP[5], IMM[1].wwww\n"
"LOAD TEMP[2].xy, BUFFER[0], TEMP[5].xxxx\n"
"LOAD TEMP[3].xy, BUFFER[0], TEMP[5].yyyy\n"
"U64ADD TEMP[3].xy, TEMP[3], -TEMP[2]\n"
"U64ADD TEMP[4].xy, TEMP[4], -TEMP[3]\n"
"ENDIF\n"
"U64ADD TEMP[0].xy, TEMP[0], TEMP[4]\n"
/* Increment pair index */
"UADD TEMP[1].y, TEMP[1].yyyy, IMM[1].xxxx\n"
"USGE TEMP[5], TEMP[1].yyyy, CONST[0][1].zzzz\n"
"UIF TEMP[5]\n"
"BRK\n"
"ENDIF\n"
"ENDLOOP\n"
/* Increment result index */
"UADD TEMP[1].x, TEMP[1].xxxx, IMM[1].xxxx\n"
"ENDLOOP\n"
"ENDIF\n"
"AND TEMP[4], CONST[0][0].wwww, IMM[1].yyyy\n"
"UIF TEMP[4]\n"
/* Store accumulated data for chaining. */
"STORE BUFFER[2].xyz, IMM[0].xxxx, TEMP[0]\n"
"ELSE\n"
"AND TEMP[4], CONST[0][0].wwww, IMM[1].zzzz\n"
"UIF TEMP[4]\n"
/* Store result availability. */
"NOT TEMP[0].z, TEMP[0]\n"
"AND TEMP[0].z, TEMP[0].zzzz, IMM[1].xxxx\n"
"STORE BUFFER[2].x, IMM[0].xxxx, TEMP[0].zzzz\n"
"AND TEMP[4], CONST[0][0].wwww, IMM[2].zzzz\n"
"UIF TEMP[4]\n"
"STORE BUFFER[2].y, IMM[0].xxxx, IMM[0].xxxx\n"
"ENDIF\n"
"ELSE\n"
/* Store result if it is available. */
"NOT TEMP[4], TEMP[0].zzzz\n"
"UIF TEMP[4]\n"
/* Apply timestamp conversion */
"AND TEMP[4], CONST[0][0].wwww, IMM[2].yyyy\n"
"UIF TEMP[4]\n"
"U64MUL TEMP[0].xy, TEMP[0], IMM[3].xyxy\n"
"U64DIV TEMP[0].xy, TEMP[0], IMM[3].zwzw\n"
"ENDIF\n"
/* Convert to boolean */
"AND TEMP[4], CONST[0][0].wwww, IMM[1].wwww\n"
"UIF TEMP[4]\n"
"U64SNE TEMP[0].x, TEMP[0].xyxy, IMM[4].zwzw\n"
"AND TEMP[0].x, TEMP[0].xxxx, IMM[1].xxxx\n"
"MOV TEMP[0].y, IMM[0].xxxx\n"
"ENDIF\n"
"AND TEMP[4], CONST[0][0].wwww, IMM[2].zzzz\n"
"UIF TEMP[4]\n"
"STORE BUFFER[2].xy, IMM[0].xxxx, TEMP[0].xyxy\n"
"ELSE\n"
/* Clamping */
"UIF TEMP[0].yyyy\n"
"MOV TEMP[0].x, IMM[0].wwww\n"
"ENDIF\n"
"AND TEMP[4], CONST[0][0].wwww, IMM[2].wwww\n"
"UIF TEMP[4]\n"
"UMIN TEMP[0].x, TEMP[0].xxxx, IMM[0].zzzz\n"
"ENDIF\n"
"STORE BUFFER[2].x, IMM[0].xxxx, TEMP[0].xxxx\n"
"ENDIF\n"
"ENDIF\n"
"ENDIF\n"
"ENDIF\n"
"END\n";
char text[sizeof(text_tmpl) + 32];
struct tgsi_token tokens[1024];
struct pipe_compute_state state = {};
/* Hard code the frequency into the shader so that the backend can
* use the full range of optimizations for divide-by-constant.
*/
snprintf(text, sizeof(text), text_tmpl,
sctx->screen->info.clock_crystal_freq);
if (!tgsi_text_translate(text, tokens, ARRAY_SIZE(tokens))) {
assert(false);
return NULL;
}
state.ir_type = PIPE_SHADER_IR_TGSI;
state.prog = tokens;
return sctx->b.create_compute_state(&sctx->b, &state);
}
int
main(int argc, char *argv[])
{
struct tgsi_token tokens[4096];
int i, chipset = 0, type = -1;
const char *filename = NULL;
FILE *f;
char text[65536] = {0};
unsigned size, *code;
for (i = 1; i < argc; i++) {
if (!strcmp(argv[i], "-a"))
chipset = strtol(argv[++i], NULL, 16);
else
filename = argv[i];
}
if (!chipset) {
_debug_printf("Must specify a chipset (-a)\n");
return 1;
}
if (!filename) {
_debug_printf("Must specify a filename\n");
return 1;
}
if (!strcmp(filename, "-"))
f = stdin;
else
f = fopen(filename, "r");
if (f == NULL) {
_debug_printf("Error opening file '%s': %s\n", filename, strerror(errno));
return 1;
}
if (!fread(text, 1, sizeof(text), f) || ferror(f)) {
_debug_printf("Error reading file '%s'\n", filename);
fclose(f);
return 1;
}
fclose(f);
_debug_printf("Compiling for NV%X\n", chipset);
if (!strncmp(text, "FRAG", 4))
type = PIPE_SHADER_FRAGMENT;
else if (!strncmp(text, "VERT", 4))
type = PIPE_SHADER_VERTEX;
else if (!strncmp(text, "GEOM", 4))
type = PIPE_SHADER_GEOMETRY;
else if (!strncmp(text, "COMP", 4))
type = PIPE_SHADER_COMPUTE;
else {
_debug_printf("Unrecognized TGSI header\n");
return 1;
}
if (!tgsi_text_translate(text, tokens, Elements(tokens))) {
_debug_printf("Failed to parse TGSI shader\n");
return 1;
}
if (chipset >= 0x50) {
i = nouveau_codegen(chipset, type, tokens, &size, &code);
} else if (chipset >= 0x30) {
i = nv30_codegen(chipset, type, tokens, &size, &code);
} else {
_debug_printf("chipset NV%02X not supported\n", chipset);
i = 1;
}
if (i)
return i;
_debug_printf("program binary (%d bytes)\n", size);
for (i = 0; i < size; i += 4) {
printf("%08x ", code[i / 4]);
if (i % (8 * 4) == (7 * 4))
printf("\n");
}
if (i % (8 * 4) != 0)
printf("\n");
return 0;
}
int main(int argc, char **argv)
{
int ret = 0, n = 1;
const char *filename;
struct tgsi_token toks[65536];
struct tgsi_parse_context parse;
struct ir3_compiler *compiler;
struct ir3_shader_variant v;
struct ir3_shader_key key = {};
const char *info;
void *ptr;
size_t size;
fd_mesa_debug |= FD_DBG_DISASM;
/* cmdline args which impact shader variant get spit out in a
* comment on the first line.. a quick/dirty way to preserve
* that info so when ir3test recompiles the shader with a new
* compiler version, we use the same shader-key settings:
*/
debug_printf("; options:");
while (n < argc) {
if (!strcmp(argv[n], "--verbose")) {
fd_mesa_debug |= FD_DBG_MSGS | FD_DBG_OPTMSGS;
n++;
continue;
}
if (!strcmp(argv[n], "--binning-pass")) {
debug_printf(" %s", argv[n]);
key.binning_pass = true;
n++;
continue;
}
if (!strcmp(argv[n], "--color-two-side")) {
debug_printf(" %s", argv[n]);
key.color_two_side = true;
n++;
continue;
}
if (!strcmp(argv[n], "--half-precision")) {
debug_printf(" %s", argv[n]);
key.half_precision = true;
n++;
continue;
}
if (!strcmp(argv[n], "--saturate-s")) {
debug_printf(" %s %s", argv[n], argv[n+1]);
key.vsaturate_s = key.fsaturate_s = strtol(argv[n+1], NULL, 0);
n += 2;
continue;
}
if (!strcmp(argv[n], "--saturate-t")) {
debug_printf(" %s %s", argv[n], argv[n+1]);
key.vsaturate_t = key.fsaturate_t = strtol(argv[n+1], NULL, 0);
n += 2;
continue;
}
if (!strcmp(argv[n], "--saturate-r")) {
debug_printf(" %s %s", argv[n], argv[n+1]);
key.vsaturate_r = key.fsaturate_r = strtol(argv[n+1], NULL, 0);
n += 2;
continue;
}
if (!strcmp(argv[n], "--help")) {
print_usage();
return 0;
}
break;
}
debug_printf("\n");
filename = argv[n];
memset(&v, 0, sizeof(v));
v.key = key;
ret = read_file(filename, &ptr, &size);
if (ret) {
print_usage();
return ret;
}
if (fd_mesa_debug & FD_DBG_OPTMSGS)
debug_printf("%s\n", (char *)ptr);
if (!tgsi_text_translate(ptr, toks, Elements(toks)))
errx(1, "could not parse `%s'", filename);
tgsi_parse_init(&parse, toks);
switch (parse.FullHeader.Processor.Processor) {
case TGSI_PROCESSOR_FRAGMENT:
v.type = SHADER_FRAGMENT;
break;
case TGSI_PROCESSOR_VERTEX:
v.type = SHADER_VERTEX;
break;
case TGSI_PROCESSOR_COMPUTE:
v.type = SHADER_COMPUTE;
break;
}
/* TODO cmdline option to target different gpus: */
compiler = ir3_compiler_create(320);
info = "NIR compiler";
ret = ir3_compile_shader_nir(compiler, &v, toks, key);
if (ret) {
fprintf(stderr, "compiler failed!\n");
return ret;
}
dump_info(&v, info);
}
int main(int argc, char **argv)
{
int ret = 0, n = 1;
char *filenames[2];
int num_files = 0;
unsigned stage = 0;
struct ir3_shader_variant v;
struct ir3_shader s;
struct ir3_shader_key key = {};
/* TODO cmdline option to target different gpus: */
unsigned gpu_id = 320;
const char *info;
void *ptr;
size_t size;
memset(&s, 0, sizeof(s));
memset(&v, 0, sizeof(v));
/* cmdline args which impact shader variant get spit out in a
* comment on the first line.. a quick/dirty way to preserve
* that info so when ir3test recompiles the shader with a new
* compiler version, we use the same shader-key settings:
*/
debug_printf("; options:");
while (n < argc) {
if (!strcmp(argv[n], "--verbose")) {
fd_mesa_debug |= FD_DBG_MSGS | FD_DBG_OPTMSGS | FD_DBG_DISASM;
n++;
continue;
}
if (!strcmp(argv[n], "--binning-pass")) {
debug_printf(" %s", argv[n]);
key.binning_pass = true;
n++;
continue;
}
if (!strcmp(argv[n], "--color-two-side")) {
debug_printf(" %s", argv[n]);
key.color_two_side = true;
n++;
continue;
}
if (!strcmp(argv[n], "--half-precision")) {
debug_printf(" %s", argv[n]);
key.half_precision = true;
n++;
continue;
}
if (!strcmp(argv[n], "--saturate-s")) {
debug_printf(" %s %s", argv[n], argv[n+1]);
key.vsaturate_s = key.fsaturate_s = strtol(argv[n+1], NULL, 0);
n += 2;
continue;
}
if (!strcmp(argv[n], "--saturate-t")) {
debug_printf(" %s %s", argv[n], argv[n+1]);
key.vsaturate_t = key.fsaturate_t = strtol(argv[n+1], NULL, 0);
n += 2;
continue;
}
if (!strcmp(argv[n], "--saturate-r")) {
debug_printf(" %s %s", argv[n], argv[n+1]);
key.vsaturate_r = key.fsaturate_r = strtol(argv[n+1], NULL, 0);
n += 2;
continue;
}
if (!strcmp(argv[n], "--astc-srgb")) {
debug_printf(" %s %s", argv[n], argv[n+1]);
key.vastc_srgb = key.fastc_srgb = strtol(argv[n+1], NULL, 0);
n += 2;
continue;
}
if (!strcmp(argv[n], "--stream-out")) {
struct pipe_stream_output_info *so = &s.stream_output;
debug_printf(" %s", argv[n]);
/* TODO more dynamic config based on number of outputs, etc
* rather than just hard-code for first output:
*/
so->num_outputs = 1;
so->stride[0] = 4;
so->output[0].register_index = 0;
so->output[0].start_component = 0;
so->output[0].num_components = 4;
so->output[0].output_buffer = 0;
so->output[0].dst_offset = 2;
so->output[0].stream = 0;
n++;
continue;
}
if (!strcmp(argv[n], "--ucp")) {
debug_printf(" %s %s", argv[n], argv[n+1]);
key.ucp_enables = strtol(argv[n+1], NULL, 0);
n += 2;
continue;
}
if (!strcmp(argv[n], "--gpu")) {
debug_printf(" %s %s", argv[n], argv[n+1]);
gpu_id = strtol(argv[n+1], NULL, 0);
n += 2;
continue;
}
if (!strcmp(argv[n], "--help")) {
print_usage();
return 0;
}
break;
}
debug_printf("\n");
while (n < argc) {
char *filename = argv[n];
char *ext = rindex(filename, '.');
if (strcmp(ext, ".tgsi") == 0) {
if (num_files != 0)
errx(1, "in TGSI mode, only a single file may be specified");
s.from_tgsi = true;
} else if (strcmp(ext, ".frag") == 0) {
if (s.from_tgsi)
errx(1, "cannot mix GLSL and TGSI");
if (num_files >= ARRAY_SIZE(filenames))
errx(1, "too many GLSL files");
stage = MESA_SHADER_FRAGMENT;
} else if (strcmp(ext, ".vert") == 0) {
if (s.from_tgsi)
errx(1, "cannot mix GLSL and TGSI");
if (num_files >= ARRAY_SIZE(filenames))
errx(1, "too many GLSL files");
stage = MESA_SHADER_VERTEX;
} else {
print_usage();
return -1;
}
filenames[num_files++] = filename;
n++;
}
nir_shader *nir;
if (s.from_tgsi) {
struct tgsi_token toks[65536];
ret = read_file(filenames[0], &ptr, &size);
if (ret) {
print_usage();
return ret;
}
if (fd_mesa_debug & FD_DBG_OPTMSGS)
debug_printf("%s\n", (char *)ptr);
if (!tgsi_text_translate(ptr, toks, ARRAY_SIZE(toks)))
errx(1, "could not parse `%s'", filenames[0]);
if (fd_mesa_debug & FD_DBG_OPTMSGS)
tgsi_dump(toks, 0);
nir = ir3_tgsi_to_nir(toks);
} else if (num_files > 0) {
nir = load_glsl(num_files, filenames, stage);
} else {
print_usage();
return -1;
}
s.compiler = ir3_compiler_create(NULL, gpu_id);
s.nir = ir3_optimize_nir(&s, nir, NULL);
v.key = key;
v.shader = &s;
switch (nir->stage) {
case MESA_SHADER_FRAGMENT:
s.type = v.type = SHADER_FRAGMENT;
break;
case MESA_SHADER_VERTEX:
s.type = v.type = SHADER_VERTEX;
break;
case MESA_SHADER_COMPUTE:
s.type = v.type = SHADER_COMPUTE;
break;
default:
errx(1, "unhandled shader stage: %d", nir->stage);
}
info = "NIR compiler";
ret = ir3_compile_shader_nir(s.compiler, &v);
if (ret) {
fprintf(stderr, "compiler failed!\n");
return ret;
}
dump_info(&v, info);
}
static void
test_texture_barrier(struct pipe_context *ctx, bool use_fbfetch,
unsigned num_samples)
{
struct cso_context *cso;
struct pipe_resource *cb;
struct pipe_sampler_view *view = NULL;
char name[256];
const char *text;
assert(num_samples >= 1 && num_samples <= 8);
util_snprintf(name, sizeof(name), "%s: %s, %u samples", __func__,
use_fbfetch ? "FBFETCH" : "sampler", MAX2(num_samples, 1));
if (!ctx->screen->get_param(ctx->screen, PIPE_CAP_TEXTURE_BARRIER)) {
util_report_result_helper(SKIP, name);
return;
}
if (use_fbfetch &&
!ctx->screen->get_param(ctx->screen, PIPE_CAP_TGSI_FS_FBFETCH)) {
util_report_result_helper(SKIP, name);
return;
}
cso = cso_create_context(ctx, 0);
cb = util_create_texture2d(ctx->screen, 256, 256,
PIPE_FORMAT_R8G8B8A8_UNORM, num_samples);
util_set_common_states_and_clear(cso, ctx, cb);
/* Clear each sample to a different value. */
if (num_samples > 1) {
void *fs =
util_make_fragment_passthrough_shader(ctx, TGSI_SEMANTIC_GENERIC,
TGSI_INTERPOLATE_LINEAR, TRUE);
cso_set_fragment_shader_handle(cso, fs);
/* Vertex shader. */
void *vs = util_set_passthrough_vertex_shader(cso, ctx, false);
for (unsigned i = 0; i < num_samples / 2; i++) {
float value;
/* 2 consecutive samples should have the same color to test MSAA
* compression properly.
*/
if (num_samples == 2) {
value = 0.1;
} else {
/* The average value must be 0.1 */
static const float values[] = {
0.0, 0.2, 0.05, 0.15
};
value = values[i];
}
ctx->set_sample_mask(ctx, 0x3 << (i * 2));
util_draw_fullscreen_quad_fill(cso, value, value, value, value);
}
ctx->set_sample_mask(ctx, ~0);
cso_set_vertex_shader_handle(cso, NULL);
cso_set_fragment_shader_handle(cso, NULL);
ctx->delete_vs_state(ctx, vs);
ctx->delete_fs_state(ctx, fs);
}
if (use_fbfetch) {
/* Fragment shader. */
text = "FRAG\n"
"DCL OUT[0], COLOR[0]\n"
"DCL TEMP[0]\n"
"IMM[0] FLT32 { 0.1, 0.2, 0.3, 0.4}\n"
"FBFETCH TEMP[0], OUT[0]\n"
"ADD OUT[0], TEMP[0], IMM[0]\n"
"END\n";
} else {
struct pipe_sampler_view templ = {{0}};
templ.format = cb->format;
templ.target = cb->target;
templ.swizzle_r = PIPE_SWIZZLE_X;
templ.swizzle_g = PIPE_SWIZZLE_Y;
templ.swizzle_b = PIPE_SWIZZLE_Z;
templ.swizzle_a = PIPE_SWIZZLE_W;
view = ctx->create_sampler_view(ctx, cb, &templ);
ctx->set_sampler_views(ctx, PIPE_SHADER_FRAGMENT, 0, 1, &view);
/* Fragment shader. */
if (num_samples > 1) {
text = "FRAG\n"
"DCL SV[0], POSITION\n"
"DCL SV[1], SAMPLEID\n"
"DCL SAMP[0]\n"
"DCL SVIEW[0], 2D_MSAA, FLOAT\n"
"DCL OUT[0], COLOR[0]\n"
"DCL TEMP[0]\n"
"IMM[0] FLT32 { 0.1, 0.2, 0.3, 0.4}\n"
"F2I TEMP[0].xy, SV[0].xyyy\n"
"MOV TEMP[0].w, SV[1].xxxx\n"
"TXF TEMP[0], TEMP[0], SAMP[0], 2D_MSAA\n"
"ADD OUT[0], TEMP[0], IMM[0]\n"
"END\n";
} else {
text = "FRAG\n"
"DCL SV[0], POSITION\n"
"DCL SAMP[0]\n"
"DCL SVIEW[0], 2D, FLOAT\n"
"DCL OUT[0], COLOR[0]\n"
"DCL TEMP[0]\n"
"IMM[0] FLT32 { 0.1, 0.2, 0.3, 0.4}\n"
"IMM[1] INT32 { 0, 0, 0, 0}\n"
"F2I TEMP[0].xy, SV[0].xyyy\n"
"MOV TEMP[0].zw, IMM[1]\n"
"TXF TEMP[0], TEMP[0], SAMP[0], 2D\n"
"ADD OUT[0], TEMP[0], IMM[0]\n"
"END\n";
}
}
struct tgsi_token tokens[1000];
struct pipe_shader_state state;
if (!tgsi_text_translate(text, tokens, ARRAY_SIZE(tokens))) {
assert(0);
util_report_result_helper(FAIL, name);
return;
}
pipe_shader_state_from_tgsi(&state, tokens);
void *fs = ctx->create_fs_state(ctx, &state);
cso_set_fragment_shader_handle(cso, fs);
/* Vertex shader. */
void *vs = util_set_passthrough_vertex_shader(cso, ctx, false);
if (num_samples > 1 && !use_fbfetch)
ctx->set_min_samples(ctx, num_samples);
for (int i = 0; i < 2; i++) {
ctx->texture_barrier(ctx,
use_fbfetch ? PIPE_TEXTURE_BARRIER_FRAMEBUFFER :
PIPE_TEXTURE_BARRIER_SAMPLER);
util_draw_fullscreen_quad(cso);
}
if (num_samples > 1 && !use_fbfetch)
ctx->set_min_samples(ctx, 1);
/* Probe pixels.
*
* For single sample:
* result = 0.1 (clear) + (0.1, 0.2, 0.3, 0.4) * 2 = (0.3, 0.5, 0.7, 0.9)
*
* For MSAA 4x:
* sample0 = 0.0 (clear) + (0.1, 0.2, 0.3, 0.4) * 2 = (0.2, 0.4, 0.6, 0.8)
* sample1 = sample0
* sample2 = 0.2 (clear) + (0.1, 0.2, 0.3, 0.4) * 2 = (0.4, 0.6, 0.8, 1.0)
* sample3 = sample2
* resolved = sum(sample[0:3]) / 4 = (0.3, 0.5, 0.7, 0.9)
*/
static const float expected[] = {0.3, 0.5, 0.7, 0.9};
bool pass = util_probe_rect_rgba(ctx, cb, 0, 0,
cb->width0, cb->height0, expected);
/* Cleanup. */
cso_destroy_context(cso);
ctx->delete_vs_state(ctx, vs);
ctx->delete_fs_state(ctx, fs);
pipe_sampler_view_reference(&view, NULL);
pipe_resource_reference(&cb, NULL);
util_report_result_helper(pass, name);
}
