<div class="gmail_extra"><br><br><div class="gmail_quote">On 20 November 2012 21:40, Kenneth Graunke <span dir="ltr"><<a href="mailto:kenneth@whitecape.org" target="_blank">kenneth@whitecape.org</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">Also, rather than having brw_wm_fs_emit poke at it directly, make it a<br>
parameter to the fs_visitor constructor.<br>
<br>
All other changes generated by search and replace (with occasional<br>
whitespace fixup).<br>
---<br>
src/mesa/drivers/dri/i965/brw_fs.cpp | 39 ++++++++++------------<br>
src/mesa/drivers/dri/i965/brw_fs.h | 4 ++-<br>
src/mesa/drivers/dri/i965/brw_fs_emit.cpp | 26 +++++++--------<br>
src/mesa/drivers/dri/i965/brw_fs_fp.cpp | 2 +-<br>
.../drivers/dri/i965/brw_fs_live_variables.cpp | 8 ++---<br>
src/mesa/drivers/dri/i965/brw_fs_reg_allocate.cpp | 10 +++---<br>
.../dri/i965/brw_fs_schedule_instructions.cpp | 2 +-<br>
src/mesa/drivers/dri/i965/brw_fs_visitor.cpp | 35 +++++++++----------<br>
src/mesa/drivers/dri/i965/brw_wm.h | 2 --<br>
9 files changed, 63 insertions(+), 65 deletions(-)<br>
<br>
diff --git a/src/mesa/drivers/dri/i965/brw_fs.cpp b/src/mesa/drivers/dri/i965/brw_fs.cpp<br>
index 8971272..bcb6727 100644<br>
--- a/src/mesa/drivers/dri/i965/brw_fs.cpp<br>
+++ b/src/mesa/drivers/dri/i965/brw_fs.cpp<br>
@@ -483,11 +483,11 @@ fs_visitor::implied_mrf_writes(fs_inst *inst)<br>
case SHADER_OPCODE_LOG2:<br>
case SHADER_OPCODE_SIN:<br>
case SHADER_OPCODE_COS:<br>
- return 1 * c->dispatch_width / 8;<br>
+ return 1 * dispatch_width / 8;<br>
case SHADER_OPCODE_POW:<br>
case SHADER_OPCODE_INT_QUOTIENT:<br>
case SHADER_OPCODE_INT_REMAINDER:<br>
- return 2 * c->dispatch_width / 8;<br>
+ return 2 * dispatch_width / 8;<br>
case SHADER_OPCODE_TEX:<br>
case FS_OPCODE_TXB:<br>
case SHADER_OPCODE_TXD:<br>
@@ -901,7 +901,7 @@ fs_visitor::emit_math(enum opcode opcode, fs_reg dst, fs_reg src)<br>
<br>
if (intel->gen < 6) {<br>
inst->base_mrf = 2;<br>
- inst->mlen = c->dispatch_width / 8;<br>
+ inst->mlen = dispatch_width / 8;<br>
}<br>
<br>
return inst;<br>
@@ -964,7 +964,7 @@ fs_visitor::emit_math(enum opcode opcode, fs_reg dst, fs_reg src0, fs_reg src1)<br>
inst = emit(opcode, dst, op0, reg_null_f);<br>
<br>
inst->base_mrf = base_mrf;<br>
- inst->mlen = 2 * c->dispatch_width / 8;<br>
+ inst->mlen = 2 * dispatch_width / 8;<br>
}<br>
return inst;<br>
}<br>
@@ -977,7 +977,7 @@ fs_visitor::emit_math(enum opcode opcode, fs_reg dst, fs_reg src0, fs_reg src1)<br>
void<br>
fs_visitor::setup_paramvalues_refs()<br>
{<br>
- if (c->dispatch_width != 8)<br>
+ if (dispatch_width != 8)<br>
return;<br>
<br>
/* Set up the pointers to ParamValues now that that array is finalized. */<br>
@@ -992,7 +992,7 @@ void<br>
fs_visitor::assign_curb_setup()<br>
{<br>
c->prog_data.curb_read_length = ALIGN(c->prog_data.nr_params, 8) / 8;<br>
- if (c->dispatch_width == 8) {<br>
+ if (dispatch_width == 8) {<br>
c->prog_data.first_curbe_grf = c->nr_payload_regs;<br>
} else {<br>
c->prog_data.first_curbe_grf_16 = c->nr_payload_regs;<br>
@@ -1273,7 +1273,7 @@ fs_visitor::compact_virtual_grfs()<br>
bool<br>
fs_visitor::remove_dead_constants()<br>
{<br>
- if (c->dispatch_width == 8) {<br>
+ if (dispatch_width == 8) {<br>
this->params_remap = ralloc_array(mem_ctx, int, c->prog_data.nr_params);<br>
<br>
for (unsigned int i = 0; i < c->prog_data.nr_params; i++)<br>
@@ -1367,7 +1367,7 @@ fs_visitor::setup_pull_constants()<br>
if (c->prog_data.nr_params <= max_uniform_components)<br>
return;<br>
<br>
- if (c->dispatch_width == 16) {<br>
+ if (dispatch_width == 16) {<br>
fail("Pull constants not supported in 16-wide\n");<br>
return;<br>
}<br>
@@ -1711,7 +1711,7 @@ fs_visitor::compute_to_mrf()<br>
int mrf_high;<br>
if (inst->dst.reg & BRW_MRF_COMPR4) {<br>
mrf_high = mrf_low + 4;<br>
- } else if (c->dispatch_width == 16 &&<br>
+ } else if (dispatch_width == 16 &&<br>
(!inst->force_uncompressed && !inst->force_sechalf)) {<br>
mrf_high = mrf_low + 1;<br>
} else {<br>
@@ -1816,7 +1816,7 @@ fs_visitor::compute_to_mrf()<br>
<br>
if (scan_inst->dst.reg & BRW_MRF_COMPR4) {<br>
scan_mrf_high = scan_mrf_low + 4;<br>
- } else if (c->dispatch_width == 16 &&<br>
+ } else if (dispatch_width == 16 &&<br>
(!scan_inst->force_uncompressed &&<br>
!scan_inst->force_sechalf)) {<br>
scan_mrf_high = scan_mrf_low + 1;<br>
@@ -1867,7 +1867,7 @@ fs_visitor::remove_duplicate_mrf_writes()<br>
bool progress = false;<br>
<br>
/* Need to update the MRF tracking for compressed instructions. */<br>
- if (c->dispatch_width == 16)<br>
+ if (dispatch_width == 16)<br>
return false;<br>
<br>
memset(last_mrf_move, 0, sizeof(last_mrf_move));<br>
@@ -1989,7 +1989,7 @@ fs_visitor::setup_payload_gen6()<br>
if (barycentric_interp_modes & (1 << i)) {<br>
c->barycentric_coord_reg[i] = c->nr_payload_regs;<br>
c->nr_payload_regs += 2;<br>
- if (c->dispatch_width == 16) {<br>
+ if (dispatch_width == 16) {<br>
c->nr_payload_regs += 2;<br>
}<br>
}<br>
@@ -1999,7 +1999,7 @@ fs_visitor::setup_payload_gen6()<br>
if (uses_depth) {<br>
c->source_depth_reg = c->nr_payload_regs;<br>
c->nr_payload_regs++;<br>
- if (c->dispatch_width == 16) {<br>
+ if (dispatch_width == 16) {<br>
/* R28: interpolated depth if not 8-wide. */<br>
c->nr_payload_regs++;<br>
}<br>
@@ -2008,7 +2008,7 @@ fs_visitor::setup_payload_gen6()<br>
if (uses_depth) {<br>
c->source_w_reg = c->nr_payload_regs;<br>
c->nr_payload_regs++;<br>
- if (c->dispatch_width == 16) {<br>
+ if (dispatch_width == 16) {<br>
/* R30: interpolated W if not 8-wide. */<br>
c->nr_payload_regs++;<br>
}<br>
@@ -2033,7 +2033,7 @@ fs_visitor::run()<br>
else<br>
setup_payload_gen4();<br>
<br>
- if (c->dispatch_width == 16) {<br>
+ if (dispatch_width == 16) {<br>
/* We have to do a compaction pass now, or the one at the end of<br>
* execution will squash down where our prog_offset start needs<br>
* to be.<br>
@@ -2131,7 +2131,7 @@ fs_visitor::run()<br>
<br>
generate_code();<br>
<br>
- if (c->dispatch_width == 8) {<br>
+ if (dispatch_width == 8) {<br>
c->prog_data.reg_blocks = brw_register_blocks(grf_used);<br>
} else {<br>
c->prog_data.reg_blocks_16 = brw_register_blocks(grf_used);<br>
@@ -2177,9 +2177,7 @@ brw_wm_fs_emit(struct brw_context *brw, struct brw_wm_compile *c,<br>
<br>
/* Now the main event: Visit the shader IR and generate our FS IR for it.<br>
*/<br>
- c->dispatch_width = 8;<br>
-<br>
- fs_visitor v(c, prog, shader);<br>
+ fs_visitor v(c, prog, shader, 8);<br>
if (!v.run()) {<br>
prog->LinkStatus = false;<br>
ralloc_strcat(&prog->InfoLog, v.fail_msg);<br>
@@ -2191,8 +2189,7 @@ brw_wm_fs_emit(struct brw_context *brw, struct brw_wm_compile *c,<br>
}<br>
<br>
if (intel->gen >= 5 && c->prog_data.nr_pull_params == 0) {<br>
- c->dispatch_width = 16;<br>
- fs_visitor v2(c, prog, shader);<br>
+ fs_visitor v2(c, prog, shader, 16);<br>
v2.import_uniforms(&v);<br>
if (!v2.run()) {<br>
perf_debug("16-wide shader failed to compile, falling back to "<br>
diff --git a/src/mesa/drivers/dri/i965/brw_fs.h b/src/mesa/drivers/dri/i965/brw_fs.h<br>
index 7ec489e..7490eb6 100644<br>
--- a/src/mesa/drivers/dri/i965/brw_fs.h<br>
+++ b/src/mesa/drivers/dri/i965/brw_fs.h<br>
@@ -185,7 +185,7 @@ class fs_visitor : public backend_visitor<br>
public:<br>
<br>
fs_visitor(struct brw_wm_compile *c, struct gl_shader_program *prog,<br>
- struct brw_shader *shader);<br>
+ struct brw_shader *shader, unsigned dispatch_width);<br>
~fs_visitor();<br>
<br>
fs_reg *variable_storage(ir_variable *var);<br>
@@ -447,6 +447,8 @@ public:<br>
<br>
int grf_used;<br>
<br>
+ unsigned dispatch_width; /** 8 or 16 */<br>
+<br>
int force_uncompressed_stack;<br>
int force_sechalf_stack;<br>
};<br>
diff --git a/src/mesa/drivers/dri/i965/brw_fs_emit.cpp b/src/mesa/drivers/dri/i965/brw_fs_emit.cpp<br>
index 29c73cf..7fdf526 100644<br>
--- a/src/mesa/drivers/dri/i965/brw_fs_emit.cpp<br>
+++ b/src/mesa/drivers/dri/i965/brw_fs_emit.cpp<br>
@@ -93,7 +93,7 @@ fs_visitor::generate_fb_write(fs_inst *inst)<br>
<br>
if (this->dual_src_output.file != BAD_FILE)<br>
msg_control = BRW_DATAPORT_RENDER_TARGET_WRITE_SIMD8_DUAL_SOURCE_SUBSPAN01;<br>
- else if (c->dispatch_width == 16)<br>
+ else if (dispatch_width == 16)<br>
msg_control = BRW_DATAPORT_RENDER_TARGET_WRITE_SIMD16_SINGLE_SOURCE;<br>
else<br>
msg_control = BRW_DATAPORT_RENDER_TARGET_WRITE_SIMD8_SINGLE_SOURCE_SUBSPAN01;<br>
@@ -101,7 +101,7 @@ fs_visitor::generate_fb_write(fs_inst *inst)<br>
brw_pop_insn_state(p);<br>
<br>
brw_fb_WRITE(p,<br>
- c->dispatch_width,<br>
+ dispatch_width,<br>
inst->base_mrf,<br>
implied_header,<br>
msg_control,<br>
@@ -133,7 +133,7 @@ fs_visitor::generate_pixel_xy(struct brw_reg dst, bool is_x)<br>
deltas = brw_imm_v(0x11001100);<br>
}<br>
<br>
- if (c->dispatch_width == 16) {<br>
+ if (dispatch_width == 16) {<br>
dst = vec16(dst);<br>
}<br>
<br>
@@ -203,7 +203,7 @@ fs_visitor::generate_math1_gen6(fs_inst *inst,<br>
BRW_MATH_DATA_VECTOR,<br>
BRW_MATH_PRECISION_FULL);<br>
<br>
- if (c->dispatch_width == 16) {<br>
+ if (dispatch_width == 16) {<br>
brw_set_compression_control(p, BRW_COMPRESSION_2NDHALF);<br>
brw_math(p, sechalf(dst),<br>
op,<br>
@@ -227,7 +227,7 @@ fs_visitor::generate_math2_gen6(fs_inst *inst,<br>
brw_set_compression_control(p, BRW_COMPRESSION_NONE);<br>
brw_math2(p, dst, op, src0, src1);<br>
<br>
- if (c->dispatch_width == 16) {<br>
+ if (dispatch_width == 16) {<br>
brw_set_compression_control(p, BRW_COMPRESSION_2NDHALF);<br>
brw_math2(p, sechalf(dst), op, sechalf(src0), sechalf(src1));<br>
brw_set_compression_control(p, BRW_COMPRESSION_COMPRESSED);<br>
@@ -250,7 +250,7 @@ fs_visitor::generate_math_gen4(fs_inst *inst,<br>
BRW_MATH_DATA_VECTOR,<br>
BRW_MATH_PRECISION_FULL);<br>
<br>
- if (c->dispatch_width == 16) {<br>
+ if (dispatch_width == 16) {<br>
brw_set_compression_control(p, BRW_COMPRESSION_2NDHALF);<br>
brw_math(p, sechalf(dst),<br>
op,<br>
@@ -282,7 +282,7 @@ fs_visitor::generate_tex(fs_inst *inst, struct brw_reg dst, struct brw_reg src)<br>
break;<br>
}<br>
<br>
- if (c->dispatch_width == 16)<br>
+ if (dispatch_width == 16)<br>
simd_mode = BRW_SAMPLER_SIMD_MODE_SIMD16;<br>
<br>
if (intel->gen >= 5) {<br>
@@ -328,7 +328,7 @@ fs_visitor::generate_tex(fs_inst *inst, struct brw_reg dst, struct brw_reg src)<br>
/* Note that G45 and older determines shadow compare and dispatch width<br>
* from message length for most messages.<br>
*/<br>
- assert(c->dispatch_width == 8);<br>
+ assert(dispatch_width == 8);<br>
msg_type = BRW_SAMPLER_MESSAGE_SIMD8_SAMPLE;<br>
if (inst->shadow_compare) {<br>
assert(inst->mlen == 6);<br>
@@ -731,10 +731,10 @@ fs_visitor::generate_code()<br>
if (unlikely(INTEL_DEBUG & DEBUG_WM)) {<br>
if (shader) {<br>
printf("Native code for fragment shader %d (%d-wide dispatch):\n",<br>
- prog->Name, c->dispatch_width);<br>
+ prog->Name, dispatch_width);<br>
} else {<br>
printf("Native code for fragment program %d (%d-wide dispatch):\n",<br>
- c->fp-><a href="http://program.Base.Id" target="_blank">program.Base.Id</a>, c->dispatch_width);<br>
+ c->fp-><a href="http://program.Base.Id" target="_blank">program.Base.Id</a>, dispatch_width);<br>
}<br>
}<br>
<br>
@@ -807,7 +807,7 @@ fs_visitor::generate_code()<br>
brw_set_predicate_inverse(p, inst->predicate_inverse);<br>
brw_set_saturate(p, inst->saturate);<br>
<br>
- if (inst->force_uncompressed || c->dispatch_width == 8) {<br>
+ if (inst->force_uncompressed || dispatch_width == 8) {<br>
brw_set_compression_control(p, BRW_COMPRESSION_NONE);<br>
} else if (inst->force_sechalf) {<br>
brw_set_compression_control(p, BRW_COMPRESSION_2NDHALF);<br>
@@ -833,7 +833,7 @@ fs_visitor::generate_code()<br>
<br>
case BRW_OPCODE_MAD:<br>
brw_set_access_mode(p, BRW_ALIGN_16);<br>
- if (c->dispatch_width == 16) {<br>
+ if (dispatch_width == 16) {<br>
brw_set_compression_control(p, BRW_COMPRESSION_NONE);<br>
brw_MAD(p, dst, src[0], src[1], src[2]);<br>
brw_set_compression_control(p, BRW_COMPRESSION_2NDHALF);<br>
@@ -893,7 +893,7 @@ fs_visitor::generate_code()<br>
assert(intel->gen == 6);<br>
gen6_IF(p, inst->conditional_mod, src[0], src[1]);<br>
} else {<br>
- brw_IF(p, c->dispatch_width == 16 ? BRW_EXECUTE_16 : BRW_EXECUTE_8);<br>
+ brw_IF(p, dispatch_width == 16 ? BRW_EXECUTE_16 : BRW_EXECUTE_8);<br>
}<br>
break;<br>
<br>
diff --git a/src/mesa/drivers/dri/i965/brw_fs_fp.cpp b/src/mesa/drivers/dri/i965/brw_fs_fp.cpp<br>
index 88b0976..dc5a386 100644<br>
--- a/src/mesa/drivers/dri/i965/brw_fs_fp.cpp<br>
+++ b/src/mesa/drivers/dri/i965/brw_fs_fp.cpp<br>
@@ -587,7 +587,7 @@ fs_visitor::setup_fp_regs()<br>
fp_temp_regs[i] = fs_reg(this, glsl_type::vec4_type);<br>
<br>
/* PROGRAM_STATE_VAR etc. */<br>
- if (c->dispatch_width == 8) {<br>
+ if (dispatch_width == 8) {<br>
for (unsigned p = 0;<br>
p < c->fp->program.Base.Parameters->NumParameters; p++) {<br>
for (unsigned int i = 0; i < 4; i++) {<br>
diff --git a/src/mesa/drivers/dri/i965/brw_fs_live_variables.cpp b/src/mesa/drivers/dri/i965/brw_fs_live_variables.cpp<br>
index d7bb721..db8f397 100644<br>
--- a/src/mesa/drivers/dri/i965/brw_fs_live_variables.cpp<br>
+++ b/src/mesa/drivers/dri/i965/brw_fs_live_variables.cpp<br>
@@ -280,10 +280,10 @@ fs_visitor::virtual_grf_interferes(int a, int b)<br>
* so our second half values in g6 got overwritten in the first<br>
* half.<br>
*/<br>
- if (c->dispatch_width == 16 && (this->pixel_x.reg == a ||<br>
- this->pixel_x.reg == b ||<br>
- this->pixel_y.reg == a ||<br>
- this->pixel_y.reg == b)) {<br>
+ if (dispatch_width == 16 && (this->pixel_x.reg == a ||<br>
+ this->pixel_x.reg == b ||<br>
+ this->pixel_y.reg == a ||<br>
+ this->pixel_y.reg == b)) {<br>
return start <= end;<br>
}<br>
<br>
diff --git a/src/mesa/drivers/dri/i965/brw_fs_reg_allocate.cpp b/src/mesa/drivers/dri/i965/brw_fs_reg_allocate.cpp<br>
index f87cbbc..c5fd6dc 100644<br>
--- a/src/mesa/drivers/dri/i965/brw_fs_reg_allocate.cpp<br>
+++ b/src/mesa/drivers/dri/i965/brw_fs_reg_allocate.cpp<br>
@@ -45,7 +45,7 @@ fs_visitor::assign_regs_trivial()<br>
{<br>
int hw_reg_mapping[this->virtual_grf_count + 1];<br>
int i;<br>
- int reg_width = c->dispatch_width / 8;<br>
+ int reg_width = dispatch_width / 8;<br>
<br>
/* Note that compressed instructions require alignment to 2 registers. */<br>
hw_reg_mapping[0] = ALIGN(this->first_non_payload_grf, reg_width);<br>
@@ -215,7 +215,7 @@ fs_visitor::setup_payload_interference(struct ra_graph *g,<br>
int payload_node_count,<br>
int first_payload_node)<br>
{<br>
- int reg_width = c->dispatch_width / 8;<br>
+ int reg_width = dispatch_width / 8;<br>
int loop_depth = 0;<br>
int loop_end_ip = 0;<br>
<br>
@@ -337,7 +337,7 @@ void<br>
fs_visitor::setup_mrf_hack_interference(struct ra_graph *g, int first_mrf_node)<br>
{<br>
int mrf_count = BRW_MAX_GRF - GEN7_MRF_HACK_START;<br>
- int reg_width = c->dispatch_width / 8;<br>
+ int reg_width = dispatch_width / 8;<br>
<br>
/* Identify all the MRFs used in the program. */<br>
bool mrf_used[mrf_count];<br>
@@ -393,7 +393,7 @@ fs_visitor::assign_regs()<br>
* registers it's allocating be contiguous physical pairs of regs<br>
* for reg_width == 2.<br>
*/<br>
- int reg_width = c->dispatch_width / 8;<br>
+ int reg_width = dispatch_width / 8;<br>
int hw_reg_mapping[this->virtual_grf_count];<br>
int payload_node_count = (ALIGN(this->first_non_payload_grf, reg_width) /<br>
reg_width);<br>
@@ -450,7 +450,7 @@ fs_visitor::assign_regs()<br>
<br>
if (reg == -1) {<br>
fail("no register to spill\n");<br>
- } else if (c->dispatch_width == 16) {<br>
+ } else if (dispatch_width == 16) {<br>
fail("Failure to register allocate. Reduce number of live scalar "<br>
"values to avoid this.");<br>
} else {<br>
diff --git a/src/mesa/drivers/dri/i965/brw_fs_schedule_instructions.cpp b/src/mesa/drivers/dri/i965/brw_fs_schedule_instructions.cpp<br>
index 11e9858..6b7c412 100644<br>
--- a/src/mesa/drivers/dri/i965/brw_fs_schedule_instructions.cpp<br>
+++ b/src/mesa/drivers/dri/i965/brw_fs_schedule_instructions.cpp<br>
@@ -239,7 +239,7 @@ instruction_scheduler::add_barrier_deps(schedule_node *n)<br>
bool<br>
instruction_scheduler::is_compressed(fs_inst *inst)<br>
{<br>
- return (v->c->dispatch_width == 16 &&<br>
+ return (v->dispatch_width == 16 &&<br>
!inst->force_uncompressed &&<br>
!inst->force_sechalf);<br>
}<br>
diff --git a/src/mesa/drivers/dri/i965/brw_fs_visitor.cpp b/src/mesa/drivers/dri/i965/brw_fs_visitor.cpp<br>
index 5d94181..28c7c5a 100644<br>
--- a/src/mesa/drivers/dri/i965/brw_fs_visitor.cpp<br>
+++ b/src/mesa/drivers/dri/i965/brw_fs_visitor.cpp<br>
@@ -110,7 +110,7 @@ fs_visitor::visit(ir_variable *ir)<br>
if (ir->uniform_block != -1)<br>
return;<br>
<br>
- if (c->dispatch_width == 16) {<br>
+ if (dispatch_width == 16) {<br>
if (!variable_storage(ir)) {<br>
fail("Failed to find uniform '%s' in 16-wide\n", ir->name);<br>
}<br>
@@ -381,7 +381,7 @@ fs_visitor::visit(ir_expression *ir)<br>
* FINISHME: Emit just the MUL if we know an operand is small<br>
* enough.<br>
*/<br>
- if (intel->gen >= 7 && c->dispatch_width == 16)<br>
+ if (intel->gen >= 7 && dispatch_width == 16)<br>
fail("16-wide explicit accumulator operands unsupported\n");<br>
<br>
struct brw_reg acc = retype(brw_acc_reg(), BRW_REGISTER_TYPE_D);<br>
@@ -394,7 +394,7 @@ fs_visitor::visit(ir_expression *ir)<br>
}<br>
break;<br>
case ir_binop_div:<br>
- if (intel->gen >= 7 && c->dispatch_width == 16)<br>
+ if (intel->gen >= 7 && dispatch_width == 16)<br>
fail("16-wide INTDIV unsupported\n");<br>
<br>
/* Floating point should be lowered by DIV_TO_MUL_RCP in the compiler. */<br>
@@ -402,7 +402,7 @@ fs_visitor::visit(ir_expression *ir)<br>
emit_math(SHADER_OPCODE_INT_QUOTIENT, this->result, op[0], op[1]);<br>
break;<br>
case ir_binop_mod:<br>
- if (intel->gen >= 7 && c->dispatch_width == 16)<br>
+ if (intel->gen >= 7 && dispatch_width == 16)<br>
fail("16-wide INTDIV unsupported\n");<br>
<br>
/* Floating point should be lowered by MOD_TO_FRACT in the compiler. */<br>
@@ -888,7 +888,7 @@ fs_visitor::emit_texture_gen5(ir_texture *ir, fs_reg dst, fs_reg coordinate,<br>
{<br>
int mlen = 0;<br>
int base_mrf = 2;<br>
- int reg_width = c->dispatch_width / 8;<br>
+ int reg_width = dispatch_width / 8;<br>
bool header_present = false;<br>
const int vector_elements =<br>
ir->coordinate ? ir->coordinate->type->vector_elements : 0;<br>
@@ -1005,7 +1005,7 @@ fs_visitor::emit_texture_gen7(ir_texture *ir, fs_reg dst, fs_reg coordinate,<br>
{<br>
int mlen = 0;<br>
int base_mrf = 2;<br>
- int reg_width = c->dispatch_width / 8;<br>
+ int reg_width = dispatch_width / 8;<br>
bool header_present = false;<br>
int offsets[3];<br>
<br>
@@ -1036,7 +1036,7 @@ fs_visitor::emit_texture_gen7(ir_texture *ir, fs_reg dst, fs_reg coordinate,<br>
mlen += reg_width;<br>
break;<br>
case ir_txd: {<br>
- if (c->dispatch_width == 16)<br>
+ if (dispatch_width == 16)<br>
fail("Gen7 does not support sample_d/sample_d_c in SIMD16 mode.");<br>
<br>
/* Load dPdx and the coordinate together:<br>
@@ -1149,7 +1149,7 @@ fs_visitor::rescale_texcoord(ir_texture *ir, fs_reg coordinate,<br>
0<br>
};<br>
<br>
- if (c->dispatch_width == 16) {<br>
+ if (dispatch_width == 16) {<br>
fail("rectangle scale uniform setup not supported on 16-wide\n");<br>
return coordinate;<br>
}<br>
@@ -1615,7 +1615,7 @@ fs_visitor::emit_if_gen6(ir_if *ir)<br>
void<br>
fs_visitor::visit(ir_if *ir)<br>
{<br>
- if (intel->gen < 6 && c->dispatch_width == 16) {<br>
+ if (intel->gen < 6 && dispatch_width == 16) {<br>
fail("Can't support (non-uniform) control flow on 16-wide\n");<br>
}<br>
<br>
@@ -1658,7 +1658,7 @@ fs_visitor::visit(ir_loop *ir)<br>
{<br>
fs_reg counter = reg_undef;<br>
<br>
- if (intel->gen < 6 && c->dispatch_width == 16) {<br>
+ if (intel->gen < 6 && dispatch_width == 16) {<br>
fail("Can't support (non-uniform) control flow on 16-wide\n");<br>
}<br>
<br>
@@ -1790,7 +1790,7 @@ fs_visitor::emit(fs_inst *inst)<br>
void<br>
fs_visitor::emit_dummy_fs()<br>
{<br>
- int reg_width = c->dispatch_width / 8;<br>
+ int reg_width = dispatch_width / 8;<br>
<br>
/* Everyone's favorite color. */<br>
emit(MOV(fs_reg(MRF, 2 + 0 * reg_width), fs_reg(1.0f)));<br>
@@ -1911,7 +1911,7 @@ fs_visitor::emit_interpolation_setup_gen6()<br>
void<br>
fs_visitor::emit_color_write(int target, int index, int first_color_mrf)<br>
{<br>
- int reg_width = c->dispatch_width / 8;<br>
+ int reg_width = dispatch_width / 8;<br>
fs_inst *inst;<br>
fs_reg color = outputs[target];<br>
fs_reg mrf;<br>
@@ -1922,7 +1922,7 @@ fs_visitor::emit_color_write(int target, int index, int first_color_mrf)<br>
<br>
color.reg_offset += index;<br>
<br>
- if (c->dispatch_width == 8 || intel->gen >= 6) {<br>
+ if (dispatch_width == 8 || intel->gen >= 6) {<br>
/* SIMD8 write looks like:<br>
* m + 0: r0<br>
* m + 1: r1<br>
@@ -1992,11 +1992,11 @@ fs_visitor::emit_fb_writes()<br>
*/<br>
int base_mrf = 1;<br>
int nr = base_mrf;<br>
- int reg_width = c->dispatch_width / 8;<br>
+ int reg_width = dispatch_width / 8;<br>
bool do_dual_src = this->dual_src_output.file != BAD_FILE;<br>
bool src0_alpha_to_render_target = false;<br>
<br>
- if (c->dispatch_width == 16 && do_dual_src) {<br>
+ if (dispatch_width == 16 && do_dual_src) {<br>
fail("GL_ARB_blend_func_extended not yet supported in 16-wide.");<br>
do_dual_src = false;<br>
}<br>
@@ -2040,7 +2040,7 @@ fs_visitor::emit_fb_writes()<br>
nr += reg_width;<br>
<br>
if (c->source_depth_to_render_target) {<br>
- if (intel->gen == 6 && c->dispatch_width == 16) {<br>
+ if (intel->gen == 6 && dispatch_width == 16) {<br>
/* For outputting oDepth on gen6, SIMD8 writes have to be<br>
* used. This would require 8-wide moves of each half to<br>
* message regs, kind of like pre-gen5 SIMD16 FB writes.<br>
@@ -2175,7 +2175,7 @@ fs_visitor::resolve_bool_comparison(ir_rvalue *rvalue, fs_reg *reg)<br>
}<br>
<br>
fs_visitor::fs_visitor(struct brw_wm_compile *c, struct gl_shader_program *prog,<br>
- struct brw_shader *shader)<br>
+ struct brw_shader *shader, unsigned dispatch_width)<br>
{<br>
this->c = c;<br>
this->p = &c->func;<br>
@@ -2186,6 +2186,7 @@ fs_visitor::fs_visitor(struct brw_wm_compile *c, struct gl_shader_program *prog,<br>
this->ctx = &intel->ctx;<br>
this->mem_ctx = ralloc_context(NULL);<br>
this->shader = shader;<br>
+ this->dispatch_width = dispatch_width;<br>
this->failed = false;<br>
this->variable_ht = hash_table_ctor(0,<br>
hash_table_pointer_hash,<br>
diff --git a/src/mesa/drivers/dri/i965/brw_wm.h b/src/mesa/drivers/dri/i965/brw_wm.h<br>
index b8d8df3..a5ebddd 100644<br>
--- a/src/mesa/drivers/dri/i965/brw_wm.h<br>
+++ b/src/mesa/drivers/dri/i965/brw_wm.h<br>
@@ -93,8 +93,6 @@ struct brw_wm_compile {<br>
GLuint runtime_check_aads_emit:1;<br>
<br>
GLuint last_scratch;<br>
-<br>
- GLuint dispatch_width;<br>
};<br>
<br>
bool brw_wm_fs_emit(struct brw_context *brw, struct brw_wm_compile *c,<br>
<span class=""><font color="#888888">--<br>
1.8.0<br>
<br>
_______________________________________________<br>
mesa-dev mailing list<br>
<a href="mailto:mesa-dev@lists.freedesktop.org">mesa-dev@lists.freedesktop.org</a><br>
<a href="http://lists.freedesktop.org/mailman/listinfo/mesa-dev" target="_blank">http://lists.freedesktop.org/mailman/listinfo/mesa-dev</a><br>
</font></span></blockquote></div><br>FYI, commit 97747ac (i965: Fix hangs with FP KIL instructions pre-gen6.) introduces a new reference to brw_wm_compile::dispatch_width, so when you get around to rebasing the series onto master, you'll need to fix that up.<br>
<br>I'm just getting started reviewing this patch series, so hopefully I'll have more interesting things to say soon :)<br></div>