/**
* Read float[4] constant(s) from VS constant buffer.
* For relative addressing, two float[4] constants will be read into 'dest'.
* Otherwise, one float[4] constant will be read into the lower half of 'dest'.
*/
void brw_dp_READ_4_vs(struct brw_compile *p,
struct brw_reg dest,
GLuint oword,
GLboolean relAddr,
struct brw_reg addrReg,
GLuint location,
GLuint bind_table_index)
{
GLuint msg_reg_nr = 1;
assert(oword < 2);
/*
printf("vs const read msg, location %u, msg_reg_nr %d\n",
location, msg_reg_nr);
*/
/* Setup MRF[1] with location/offset into const buffer */
{
struct brw_reg b;
brw_push_insn_state(p);
brw_set_compression_control(p, BRW_COMPRESSION_NONE);
brw_set_mask_control(p, BRW_MASK_DISABLE);
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
/*brw_set_access_mode(p, BRW_ALIGN_16);*/
/* XXX I think we're setting all the dwords of MRF[1] to 'location'.
* when the docs say only dword[2] should be set. Hmmm. But it works.
*/
b = brw_message_reg(msg_reg_nr);
b = retype(b, BRW_REGISTER_TYPE_UD);
/*b = get_element_ud(b, 2);*/
if (relAddr) {
brw_ADD(p, b, addrReg, brw_imm_ud(location));
}
else {
brw_MOV(p, b, brw_imm_ud(location));
}
brw_pop_insn_state(p);
}
{
struct brw_instruction *insn = next_insn(p, BRW_OPCODE_SEND);
insn->header.predicate_control = BRW_PREDICATE_NONE;
insn->header.compression_control = BRW_COMPRESSION_NONE;
insn->header.destreg__conditionalmod = msg_reg_nr;
insn->header.mask_control = BRW_MASK_DISABLE;
/*insn->header.access_mode = BRW_ALIGN_16;*/
brw_set_dest(insn, dest);
brw_set_src0(insn, brw_null_reg());
brw_set_dp_read_message(p->brw,
insn,
bind_table_index,
oword, /* 0 = lower Oword, 1 = upper Oword */
BRW_DATAPORT_READ_MESSAGE_OWORD_BLOCK_READ, /* msg_type */
0, /* source cache = data cache */
1, /* msg_length */
1, /* response_length (1 Oword) */
0); /* eot */
}
}
/**
* Texture sample instruction.
* Note: the msg_type plus msg_length values determine exactly what kind
* of sampling operation is performed. See volume 4, page 161 of docs.
*/
void brw_SAMPLE(struct brw_compile *p,
struct brw_reg dest,
GLuint msg_reg_nr,
struct brw_reg src0,
GLuint binding_table_index,
GLuint sampler,
GLuint writemask,
GLuint msg_type,
GLuint response_length,
GLuint msg_length,
GLboolean eot,
GLuint header_present,
GLuint simd_mode)
{
GLboolean need_stall = 0;
if (writemask == 0) {
/*_mesa_printf("%s: zero writemask??\n", __FUNCTION__); */
return;
}
/* Hardware doesn't do destination dependency checking on send
* instructions properly. Add a workaround which generates the
* dependency by other means. In practice it seems like this bug
* only crops up for texture samples, and only where registers are
* written by the send and then written again later without being
* read in between. Luckily for us, we already track that
* information and use it to modify the writemask for the
* instruction, so that is a guide for whether a workaround is
* needed.
*/
if (writemask != WRITEMASK_XYZW) {
GLuint dst_offset = 0;
GLuint i, newmask = 0, len = 0;
for (i = 0; i < 4; i++) {
if (writemask & (1<<i))
break;
dst_offset += 2;
}
for (; i < 4; i++) {
if (!(writemask & (1<<i)))
break;
newmask |= 1<<i;
len++;
}
if (newmask != writemask) {
need_stall = 1;
/* _mesa_printf("need stall %x %x\n", newmask , writemask); */
}
else {
struct brw_reg m1 = brw_message_reg(msg_reg_nr);
newmask = ~newmask & WRITEMASK_XYZW;
brw_push_insn_state(p);
brw_set_compression_control(p, BRW_COMPRESSION_NONE);
brw_set_mask_control(p, BRW_MASK_DISABLE);
brw_MOV(p, m1, brw_vec8_grf(0,0));
brw_MOV(p, get_element_ud(m1, 2), brw_imm_ud(newmask << 12));
brw_pop_insn_state(p);
src0 = retype(brw_null_reg(), BRW_REGISTER_TYPE_UW);
dest = offset(dest, dst_offset);
response_length = len * 2;
}
}
{
struct brw_instruction *insn = next_insn(p, BRW_OPCODE_SEND);
insn->header.predicate_control = 0; /* XXX */
insn->header.compression_control = BRW_COMPRESSION_NONE;
insn->header.destreg__conditionalmod = msg_reg_nr;
brw_set_dest(insn, dest);
brw_set_src0(insn, src0);
brw_set_sampler_message(p->brw, insn,
binding_table_index,
sampler,
msg_type,
response_length,
msg_length,
eot,
header_present,
simd_mode);
}
if (need_stall) {
struct brw_reg reg = vec8(offset(dest, response_length-1));
/* mov (8) r9.0<1>:f r9.0<8;8,1>:f { Align1 }
*/
brw_push_insn_state(p);
brw_set_compression_control(p, BRW_COMPRESSION_NONE);
brw_MOV(p, reg, reg);
brw_pop_insn_state(p);
}
}
const GLuint *
brw_blorp_const_color_program::compile(struct brw_context *brw,
GLuint *program_size)
{
/* Set up prog_data */
memset(&prog_data, 0, sizeof(prog_data));
prog_data.persample_msaa_dispatch = false;
alloc_regs();
brw_set_compression_control(&func, BRW_COMPRESSION_NONE);
struct brw_reg mrf_rt_write =
retype(vec16(brw_message_reg(base_mrf)), BRW_REGISTER_TYPE_F);
uint32_t mlen, msg_type;
if (key->use_simd16_replicated_data) {
/* The message payload is a single register with the low 4 floats/ints
* filled with the constant clear color.
*/
brw_set_mask_control(&func, BRW_MASK_DISABLE);
brw_MOV(&func, vec4(brw_message_reg(base_mrf)), clear_rgba);
brw_set_mask_control(&func, BRW_MASK_ENABLE);
msg_type = BRW_DATAPORT_RENDER_TARGET_WRITE_SIMD16_SINGLE_SOURCE_REPLICATED;
mlen = 1;
} else {
for (int i = 0; i < 4; i++) {
/* The message payload is pairs of registers for 16 pixels each of r,
* g, b, and a.
*/
brw_set_compression_control(&func, BRW_COMPRESSION_COMPRESSED);
brw_MOV(&func,
brw_message_reg(base_mrf + i * 2),
brw_vec1_grf(clear_rgba.nr, i));
brw_set_compression_control(&func, BRW_COMPRESSION_NONE);
}
msg_type = BRW_DATAPORT_RENDER_TARGET_WRITE_SIMD16_SINGLE_SOURCE;
mlen = 8;
}
/* Now write to the render target and terminate the thread */
brw_fb_WRITE(&func,
16 /* dispatch_width */,
base_mrf /* msg_reg_nr */,
mrf_rt_write /* src0 */,
msg_type,
BRW_BLORP_RENDERBUFFER_BINDING_TABLE_INDEX,
mlen,
0 /* response_length */,
true /* eot */,
false /* header present */);
if (unlikely(INTEL_DEBUG & DEBUG_BLORP)) {
printf("Native code for BLORP clear:\n");
brw_dump_compile(&func, stdout, 0, func.next_insn_offset);
printf("\n");
}
return brw_get_program(&func, program_size);
}
static void compile_gs_prog( struct brw_context *brw,
struct brw_gs_prog_key *key )
{
struct intel_context *intel = &brw->intel;
struct brw_gs_compile c;
const GLuint *program;
void *mem_ctx;
GLuint program_size;
memset(&c, 0, sizeof(c));
c.key = *key;
/* The geometry shader needs to access the entire VUE. */
brw_compute_vue_map(&c.vue_map, intel, c.key.userclip_active, c.key.attrs);
c.nr_regs = (c.vue_map.num_slots + 1)/2;
mem_ctx = NULL;
/* Begin the compilation:
*/
brw_init_compile(brw, &c.func, mem_ctx);
c.func.single_program_flow = 1;
/* For some reason the thread is spawned with only 4 channels
* unmasked.
*/
brw_set_mask_control(&c.func, BRW_MASK_DISABLE);
if (intel->gen >= 6) {
unsigned num_verts;
bool check_edge_flag;
/* On Sandybridge, we use the GS for implementing transform feedback
* (called "Stream Out" in the PRM).
*/
switch (key->primitive) {
case _3DPRIM_POINTLIST:
num_verts = 1;
check_edge_flag = false;
break;
case _3DPRIM_LINELIST:
case _3DPRIM_LINESTRIP:
case _3DPRIM_LINELOOP:
num_verts = 2;
check_edge_flag = false;
break;
case _3DPRIM_TRILIST:
case _3DPRIM_TRIFAN:
case _3DPRIM_TRISTRIP:
case _3DPRIM_RECTLIST:
num_verts = 3;
check_edge_flag = false;
break;
case _3DPRIM_QUADLIST:
case _3DPRIM_QUADSTRIP:
case _3DPRIM_POLYGON:
num_verts = 3;
check_edge_flag = true;
break;
default:
assert(!"Unexpected primitive type in Gen6 SOL program.");
return;
}
gen6_sol_program(&c, key, num_verts, check_edge_flag);
} else {
/* On Gen4-5, we use the GS to decompose certain types of primitives.
* Note that primitives which don't require a GS program have already
* been weeded out by now.
*/
switch (key->primitive) {
case _3DPRIM_QUADLIST:
brw_gs_quads( &c, key );
break;
case _3DPRIM_QUADSTRIP:
brw_gs_quad_strip( &c, key );
break;
case _3DPRIM_LINELOOP:
brw_gs_lines( &c );
break;
default:
ralloc_free(mem_ctx);
return;
}
}
/* get the program
*/
program = brw_get_program(&c.func, &program_size);
if (unlikely(INTEL_DEBUG & DEBUG_GS)) {
int i;
printf("gs:\n");
for (i = 0; i < program_size / sizeof(struct brw_instruction); i++)
brw_disasm(stdout, &((struct brw_instruction *)program)[i],
intel->gen);
printf("\n");
}
brw_upload_cache(&brw->cache, BRW_GS_PROG,
&c.key, sizeof(c.key),
program, program_size,
&c.prog_data, sizeof(c.prog_data),
&brw->gs.prog_offset, &brw->gs.prog_data);
ralloc_free(mem_ctx);
}
void
vec4_generator::generate_gs_set_channel_masks(struct brw_reg dst,
struct brw_reg src)
{
/* From p21 of volume 4 part 2 of the Ivy Bridge PRM (2.4.3.1 Message
* Header: M0.5):
*
* 15 Vertex 1 DATA [3] / Vertex 0 DATA[7] Channel Mask
*
* When Swizzle Control = URB_INTERLEAVED this bit controls Vertex 1
* DATA[3], when Swizzle Control = URB_NOSWIZZLE this bit controls
* Vertex 0 DATA[7]. This bit is ANDed with the corresponding
* channel enable to determine the final channel enable. For the
* URB_READ_OWORD & URB_READ_HWORD messages, when final channel
* enable is 1 it indicates that Vertex 1 DATA [3] will be included
* in the writeback message. For the URB_WRITE_OWORD &
* URB_WRITE_HWORD messages, when final channel enable is 1 it
* indicates that Vertex 1 DATA [3] will be written to the surface.
*
* 0: Vertex 1 DATA [3] / Vertex 0 DATA[7] channel not included
* 1: Vertex DATA [3] / Vertex 0 DATA[7] channel included
*
* 14 Vertex 1 DATA [2] Channel Mask
* 13 Vertex 1 DATA [1] Channel Mask
* 12 Vertex 1 DATA [0] Channel Mask
* 11 Vertex 0 DATA [3] Channel Mask
* 10 Vertex 0 DATA [2] Channel Mask
* 9 Vertex 0 DATA [1] Channel Mask
* 8 Vertex 0 DATA [0] Channel Mask
*
* (This is from a section of the PRM that is agnostic to the particular
* type of shader being executed, so "Vertex 0" and "Vertex 1" refer to
* geometry shader invocations 0 and 1, respectively). Since we have the
* enable flags for geometry shader invocation 0 in bits 3:0 of DWORD 0,
* and the enable flags for geometry shader invocation 1 in bits 7:0 of
* DWORD 4, we just need to OR them together and store the result in bits
* 15:8 of DWORD 5.
*
* It's easier to get the EU to do this if we think of the src and dst
* registers as composed of 32 bytes each; then, we want to pick up the
* contents of bytes 0 and 16 from src, OR them together, and store them in
* byte 21.
*
* We can do that by the following EU instruction:
*
* or(1) dst.21<1>UB src<0,1,0>UB src.16<0,1,0>UB { align1 WE_all }
*
* Note: this relies on the source register having zeros in (a) bits 7:4 of
* DWORD 0 and (b) bits 3:0 of DWORD 4. We can rely on (b) because the
* source register was prepared by GS_OPCODE_PREPARE_CHANNEL_MASKS (which
* shifts DWORD 4 left by 4 bits), and we can rely on (a) because prior to
* the execution of GS_OPCODE_PREPARE_CHANNEL_MASKS, DWORDs 0 and 4 need to
* contain valid channel mask values (which are in the range 0x0-0xf).
*/
dst = retype(dst, BRW_REGISTER_TYPE_UB);
src = retype(src, BRW_REGISTER_TYPE_UB);
brw_push_insn_state(p);
brw_set_access_mode(p, BRW_ALIGN_1);
brw_set_mask_control(p, BRW_MASK_DISABLE);
brw_OR(p, suboffset(vec1(dst), 21), vec1(src), suboffset(vec1(src), 16));
brw_pop_insn_state(p);
}
void
vec4_generator::generate_tex(vec4_instruction *inst,
struct brw_reg dst,
struct brw_reg src)
{
int msg_type = -1;
if (brw->gen >= 5) {
switch (inst->opcode) {
case SHADER_OPCODE_TEX:
case SHADER_OPCODE_TXL:
if (inst->shadow_compare) {
msg_type = GEN5_SAMPLER_MESSAGE_SAMPLE_LOD_COMPARE;
} else {
msg_type = GEN5_SAMPLER_MESSAGE_SAMPLE_LOD;
}
break;
case SHADER_OPCODE_TXD:
if (inst->shadow_compare) {
/* Gen7.5+. Otherwise, lowered by brw_lower_texture_gradients(). */
assert(brw->is_haswell);
msg_type = HSW_SAMPLER_MESSAGE_SAMPLE_DERIV_COMPARE;
} else {
msg_type = GEN5_SAMPLER_MESSAGE_SAMPLE_DERIVS;
}
break;
case SHADER_OPCODE_TXF:
msg_type = GEN5_SAMPLER_MESSAGE_SAMPLE_LD;
break;
case SHADER_OPCODE_TXF_CMS:
if (brw->gen >= 7)
msg_type = GEN7_SAMPLER_MESSAGE_SAMPLE_LD2DMS;
else
msg_type = GEN5_SAMPLER_MESSAGE_SAMPLE_LD;
break;
case SHADER_OPCODE_TXF_MCS:
assert(brw->gen >= 7);
msg_type = GEN7_SAMPLER_MESSAGE_SAMPLE_LD_MCS;
break;
case SHADER_OPCODE_TXS:
msg_type = GEN5_SAMPLER_MESSAGE_SAMPLE_RESINFO;
break;
case SHADER_OPCODE_TG4:
if (inst->shadow_compare) {
msg_type = GEN7_SAMPLER_MESSAGE_SAMPLE_GATHER4_C;
} else {
msg_type = GEN7_SAMPLER_MESSAGE_SAMPLE_GATHER4;
}
break;
case SHADER_OPCODE_TG4_OFFSET:
if (inst->shadow_compare) {
msg_type = GEN7_SAMPLER_MESSAGE_SAMPLE_GATHER4_PO_C;
} else {
msg_type = GEN7_SAMPLER_MESSAGE_SAMPLE_GATHER4_PO;
}
break;
default:
assert(!"should not get here: invalid vec4 texture opcode");
break;
}
} else {
switch (inst->opcode) {
case SHADER_OPCODE_TEX:
case SHADER_OPCODE_TXL:
if (inst->shadow_compare) {
msg_type = BRW_SAMPLER_MESSAGE_SIMD4X2_SAMPLE_LOD_COMPARE;
assert(inst->mlen == 3);
} else {
msg_type = BRW_SAMPLER_MESSAGE_SIMD4X2_SAMPLE_LOD;
assert(inst->mlen == 2);
}
break;
case SHADER_OPCODE_TXD:
/* There is no sample_d_c message; comparisons are done manually. */
msg_type = BRW_SAMPLER_MESSAGE_SIMD4X2_SAMPLE_GRADIENTS;
assert(inst->mlen == 4);
break;
case SHADER_OPCODE_TXF:
msg_type = BRW_SAMPLER_MESSAGE_SIMD4X2_LD;
assert(inst->mlen == 2);
break;
case SHADER_OPCODE_TXS:
msg_type = BRW_SAMPLER_MESSAGE_SIMD4X2_RESINFO;
assert(inst->mlen == 2);
break;
default:
assert(!"should not get here: invalid vec4 texture opcode");
break;
}
}
assert(msg_type != -1);
/* Load the message header if present. If there's a texture offset, we need
* to set it up explicitly and load the offset bitfield. Otherwise, we can
* use an implied move from g0 to the first message register.
*/
if (inst->header_present) {
if (brw->gen < 6 && !inst->texture_offset) {
/* Set up an implied move from g0 to the MRF. */
src = brw_vec8_grf(0, 0);
} else {
struct brw_reg header =
retype(brw_message_reg(inst->base_mrf), BRW_REGISTER_TYPE_UD);
/* Explicitly set up the message header by copying g0 to the MRF. */
brw_push_insn_state(p);
brw_set_mask_control(p, BRW_MASK_DISABLE);
brw_MOV(p, header, retype(brw_vec8_grf(0, 0), BRW_REGISTER_TYPE_UD));
brw_set_access_mode(p, BRW_ALIGN_1);
if (inst->texture_offset) {
/* Set the texel offset bits in DWord 2. */
brw_MOV(p, get_element_ud(header, 2),
brw_imm_ud(inst->texture_offset));
}
if (inst->sampler >= 16) {
/* The "Sampler Index" field can only store values between 0 and 15.
* However, we can add an offset to the "Sampler State Pointer"
* field, effectively selecting a different set of 16 samplers.
*
* The "Sampler State Pointer" needs to be aligned to a 32-byte
* offset, and each sampler state is only 16-bytes, so we can't
* exclusively use the offset - we have to use both.
*/
assert(brw->is_haswell); /* field only exists on Haswell */
brw_ADD(p,
get_element_ud(header, 3),
get_element_ud(brw_vec8_grf(0, 0), 3),
brw_imm_ud(16 * (inst->sampler / 16) *
sizeof(gen7_sampler_state)));
}
brw_pop_insn_state(p);
}
}
uint32_t return_format;
switch (dst.type) {
case BRW_REGISTER_TYPE_D:
return_format = BRW_SAMPLER_RETURN_FORMAT_SINT32;
break;
case BRW_REGISTER_TYPE_UD:
return_format = BRW_SAMPLER_RETURN_FORMAT_UINT32;
break;
default:
return_format = BRW_SAMPLER_RETURN_FORMAT_FLOAT32;
break;
}
uint32_t surface_index = ((inst->opcode == SHADER_OPCODE_TG4 ||
inst->opcode == SHADER_OPCODE_TG4_OFFSET)
? prog_data->base.binding_table.gather_texture_start
: prog_data->base.binding_table.texture_start) + inst->sampler;
brw_SAMPLE(p,
dst,
inst->base_mrf,
src,
surface_index,
inst->sampler % 16,
msg_type,
1, /* response length */
inst->mlen,
inst->header_present,
BRW_SAMPLER_SIMD_MODE_SIMD4X2,
return_format);
mark_surface_used(surface_index);
}
void
vec4_generator::generate_code(exec_list *instructions)
{
int last_native_insn_offset = 0;
const char *last_annotation_string = NULL;
const void *last_annotation_ir = NULL;
if (unlikely(debug_flag)) {
if (shader_prog) {
printf("Native code for vertex shader %d:\n", shader_prog->Name);
} else {
printf("Native code for vertex program %d:\n", prog->Id);
}
}
foreach_list(node, instructions) {
vec4_instruction *inst = (vec4_instruction *)node;
struct brw_reg src[3], dst;
if (unlikely(debug_flag)) {
if (last_annotation_ir != inst->ir) {
last_annotation_ir = inst->ir;
if (last_annotation_ir) {
printf(" ");
if (shader_prog) {
((ir_instruction *) last_annotation_ir)->print();
} else {
const prog_instruction *vpi;
vpi = (const prog_instruction *) inst->ir;
printf("%d: ", (int)(vpi - prog->Instructions));
_mesa_fprint_instruction_opt(stdout, vpi, 0,
PROG_PRINT_DEBUG, NULL);
}
printf("\n");
}
}
if (last_annotation_string != inst->annotation) {
last_annotation_string = inst->annotation;
if (last_annotation_string)
printf(" %s\n", last_annotation_string);
}
}
for (unsigned int i = 0; i < 3; i++) {
src[i] = inst->get_src(this->prog_data, i);
}
dst = inst->get_dst();
brw_set_conditionalmod(p, inst->conditional_mod);
brw_set_predicate_control(p, inst->predicate);
brw_set_predicate_inverse(p, inst->predicate_inverse);
brw_set_saturate(p, inst->saturate);
brw_set_mask_control(p, inst->force_writemask_all);
unsigned pre_emit_nr_insn = p->nr_insn;
generate_vec4_instruction(inst, dst, src);
if (inst->no_dd_clear || inst->no_dd_check) {
assert(p->nr_insn == pre_emit_nr_insn + 1 ||
!"no_dd_check or no_dd_clear set for IR emitting more "
"than 1 instruction");
struct brw_instruction *last = &p->store[pre_emit_nr_insn];
if (inst->no_dd_clear)
last->header.dependency_control |= BRW_DEPENDENCY_NOTCLEARED;
if (inst->no_dd_check)
last->header.dependency_control |= BRW_DEPENDENCY_NOTCHECKED;
}
if (unlikely(debug_flag)) {
brw_dump_compile(p, stdout,
last_native_insn_offset, p->next_insn_offset);
}
last_native_insn_offset = p->next_insn_offset;
}
static void compile_clip_prog( struct brw_context *brw,
struct brw_clip_prog_key *key )
{
struct intel_context *intel = &brw->intel;
struct brw_clip_compile c;
const GLuint *program;
GLuint program_size;
GLuint delta;
GLuint i;
GLuint header_regs;
memset(&c, 0, sizeof(c));
/* Begin the compilation:
*/
brw_init_compile(brw, &c.func);
c.func.single_program_flow = 1;
c.key = *key;
/* Need to locate the two positions present in vertex + header.
* These are currently hardcoded:
*/
c.header_position_offset = ATTR_SIZE;
if (intel->gen == 5)
header_regs = 3;
else
header_regs = 1;
delta = header_regs * REG_SIZE;
for (i = 0; i < VERT_RESULT_MAX; i++) {
if (c.key.attrs & BITFIELD64_BIT(i)) {
c.offset[i] = delta;
delta += ATTR_SIZE;
c.idx_to_attr[c.nr_attrs] = i;
c.nr_attrs++;
}
}
/* The vertex attributes start at a URB row-aligned offset after
* the 8-20 dword vertex header, and continue for a URB row-aligned
* length. nr_regs determines the urb_read_length from the start
* of the header to the end of the vertex data.
*/
c.nr_regs = header_regs + (c.nr_attrs + 1) / 2;
c.nr_bytes = c.nr_regs * REG_SIZE;
c.prog_data.clip_mode = c.key.clip_mode; /* XXX */
/* For some reason the thread is spawned with only 4 channels
* unmasked.
*/
brw_set_mask_control(&c.func, BRW_MASK_DISABLE);
/* Would ideally have the option of producing a program which could
* do all three:
*/
switch (key->primitive) {
case GL_TRIANGLES:
if (key->do_unfilled)
brw_emit_unfilled_clip( &c );
else
brw_emit_tri_clip( &c );
break;
case GL_LINES:
brw_emit_line_clip( &c );
break;
case GL_POINTS:
brw_emit_point_clip( &c );
break;
default:
assert(0);
return;
}
/* get the program
*/
program = brw_get_program(&c.func, &program_size);
if (INTEL_DEBUG & DEBUG_CLIP) {
printf("clip:\n");
for (i = 0; i < program_size / sizeof(struct brw_instruction); i++)
brw_disasm(stdout, &((struct brw_instruction *)program)[i],
intel->gen);
printf("\n");
}
/* Upload
*/
drm_intel_bo_unreference(brw->clip.prog_bo);
brw->clip.prog_bo = brw_upload_cache_with_auxdata(&brw->cache,
BRW_CLIP_PROG,
&c.key, sizeof(c.key),
NULL, 0,
program, program_size,
&c.prog_data,
sizeof(c.prog_data),
&brw->clip.prog_data);
}
static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c)
{
#define MAX_IFSN 32
#define MAX_LOOP_DEPTH 32
struct brw_instruction *if_inst[MAX_IFSN], *loop_inst[MAX_LOOP_DEPTH];
struct brw_instruction *inst0, *inst1;
int i, if_insn = 0, loop_insn = 0;
struct brw_compile *p = &c->func;
struct brw_indirect stack_index = brw_indirect(0, 0);
c->reg_index = 0;
prealloc_reg(c);
brw_set_compression_control(p, BRW_COMPRESSION_NONE);
brw_MOV(p, get_addr_reg(stack_index), brw_address(c->stack));
for (i = 0; i < c->nr_fp_insns; i++) {
struct prog_instruction *inst = &c->prog_instructions[i];
struct prog_instruction *orig_inst;
if ((orig_inst = inst->Data) != 0)
orig_inst->Data = current_insn(p);
if (inst->CondUpdate)
brw_set_conditionalmod(p, BRW_CONDITIONAL_NZ);
else
brw_set_conditionalmod(p, BRW_CONDITIONAL_NONE);
switch (inst->Opcode) {
case WM_PIXELXY:
emit_pixel_xy(c, inst);
break;
case WM_DELTAXY:
emit_delta_xy(c, inst);
break;
case WM_PIXELW:
emit_pixel_w(c, inst);
break;
case WM_LINTERP:
emit_linterp(c, inst);
break;
case WM_PINTERP:
emit_pinterp(c, inst);
break;
case WM_CINTERP:
emit_cinterp(c, inst);
break;
case WM_WPOSXY:
emit_wpos_xy(c, inst);
break;
case WM_FB_WRITE:
emit_fb_write(c, inst);
break;
case OPCODE_ABS:
emit_abs(c, inst);
break;
case OPCODE_ADD:
emit_add(c, inst);
break;
case OPCODE_SUB:
emit_sub(c, inst);
break;
case OPCODE_FRC:
emit_frc(c, inst);
break;
case OPCODE_FLR:
emit_flr(c, inst);
break;
case OPCODE_LRP:
emit_lrp(c, inst);
break;
case OPCODE_INT:
emit_int(c, inst);
break;
case OPCODE_MOV:
emit_mov(c, inst);
break;
case OPCODE_DP3:
emit_dp3(c, inst);
break;
case OPCODE_DP4:
emit_dp4(c, inst);
break;
case OPCODE_XPD:
emit_xpd(c, inst);
break;
case OPCODE_DPH:
emit_dph(c, inst);
break;
case OPCODE_RCP:
emit_rcp(c, inst);
break;
case OPCODE_RSQ:
emit_rsq(c, inst);
break;
case OPCODE_SIN:
emit_sin(c, inst);
break;
case OPCODE_COS:
emit_cos(c, inst);
break;
case OPCODE_EX2:
emit_ex2(c, inst);
break;
case OPCODE_LG2:
emit_lg2(c, inst);
break;
case OPCODE_MAX:
emit_max(c, inst);
break;
case OPCODE_MIN:
emit_min(c, inst);
break;
case OPCODE_DDX:
emit_ddx(c, inst);
break;
case OPCODE_DDY:
emit_ddy(c, inst);
break;
case OPCODE_SLT:
emit_slt(c, inst);
break;
case OPCODE_SLE:
emit_sle(c, inst);
break;
case OPCODE_SGT:
emit_sgt(c, inst);
break;
case OPCODE_SGE:
emit_sge(c, inst);
break;
case OPCODE_SEQ:
emit_seq(c, inst);
break;
case OPCODE_SNE:
emit_sne(c, inst);
break;
case OPCODE_MUL:
emit_mul(c, inst);
break;
case OPCODE_POW:
emit_pow(c, inst);
break;
case OPCODE_MAD:
emit_mad(c, inst);
break;
case OPCODE_TEX:
emit_tex(c, inst);
break;
case OPCODE_TXB:
emit_txb(c, inst);
break;
case OPCODE_KIL_NV:
emit_kil(c);
break;
case OPCODE_IF:
assert(if_insn < MAX_IFSN);
if_inst[if_insn++] = brw_IF(p, BRW_EXECUTE_8);
break;
case OPCODE_ELSE:
if_inst[if_insn-1] = brw_ELSE(p, if_inst[if_insn-1]);
break;
case OPCODE_ENDIF:
assert(if_insn > 0);
brw_ENDIF(p, if_inst[--if_insn]);
break;
case OPCODE_BGNSUB:
case OPCODE_ENDSUB:
break;
case OPCODE_CAL:
brw_push_insn_state(p);
brw_set_mask_control(p, BRW_MASK_DISABLE);
brw_set_access_mode(p, BRW_ALIGN_1);
brw_ADD(p, deref_1ud(stack_index, 0), brw_ip_reg(), brw_imm_d(3*16));
brw_set_access_mode(p, BRW_ALIGN_16);
brw_ADD(p, get_addr_reg(stack_index),
get_addr_reg(stack_index), brw_imm_d(4));
orig_inst = inst->Data;
orig_inst->Data = &p->store[p->nr_insn];
brw_ADD(p, brw_ip_reg(), brw_ip_reg(), brw_imm_d(1*16));
brw_pop_insn_state(p);
break;
case OPCODE_RET:
brw_push_insn_state(p);
brw_set_mask_control(p, BRW_MASK_DISABLE);
brw_ADD(p, get_addr_reg(stack_index),
get_addr_reg(stack_index), brw_imm_d(-4));
brw_set_access_mode(p, BRW_ALIGN_1);
brw_MOV(p, brw_ip_reg(), deref_1ud(stack_index, 0));
brw_set_access_mode(p, BRW_ALIGN_16);
brw_pop_insn_state(p);
break;
case OPCODE_BGNLOOP:
loop_inst[loop_insn++] = brw_DO(p, BRW_EXECUTE_8);
break;
case OPCODE_BRK:
brw_BREAK(p);
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
break;
case OPCODE_CONT:
brw_CONT(p);
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
break;
case OPCODE_ENDLOOP:
loop_insn--;
inst0 = inst1 = brw_WHILE(p, loop_inst[loop_insn]);
/* patch all the BREAK instructions from
last BEGINLOOP */
while (inst0 > loop_inst[loop_insn]) {
inst0--;
if (inst0->header.opcode == BRW_OPCODE_BREAK) {
inst0->bits3.if_else.jump_count = inst1 - inst0 + 1;
inst0->bits3.if_else.pop_count = 0;
} else if (inst0->header.opcode == BRW_OPCODE_CONTINUE) {
inst0->bits3.if_else.jump_count = inst1 - inst0;
inst0->bits3.if_else.pop_count = 0;
}
}
break;
default:
_mesa_printf("unsupported IR in fragment shader %d\n",
inst->Opcode);
}
if (inst->CondUpdate)
brw_set_predicate_control(p, BRW_PREDICATE_NORMAL);
else
brw_set_predicate_control(p, BRW_PREDICATE_NONE);
}
post_wm_emit(c);
for (i = 0; i < c->fp->program.Base.NumInstructions; i++)
c->fp->program.Base.Instructions[i].Data = NULL;
}