summaryrefslogtreecommitdiff
path: root/src/mesa/shader
diff options
context:
space:
mode:
Diffstat (limited to 'src/mesa/shader')
-rw-r--r--src/mesa/shader/prog_optimize.c116
-rw-r--r--src/mesa/shader/prog_print.c6
-rw-r--r--src/mesa/shader/program_parse.tab.c2
-rw-r--r--src/mesa/shader/program_parse.y2
-rw-r--r--src/mesa/shader/program_parser.h2
-rw-r--r--src/mesa/shader/programopt.c8
-rw-r--r--src/mesa/shader/slang/slang_codegen.c2
-rw-r--r--src/mesa/shader/slang/slang_link.c25
8 files changed, 106 insertions, 57 deletions
diff --git a/src/mesa/shader/prog_optimize.c b/src/mesa/shader/prog_optimize.c
index 8638754e24..3d28d885a4 100644
--- a/src/mesa/shader/prog_optimize.c
+++ b/src/mesa/shader/prog_optimize.c
@@ -65,6 +65,7 @@ get_src_arg_mask(const struct prog_instruction *inst, int arg)
case OPCODE_DP2:
return WRITEMASK_XY;
case OPCODE_DP3:
+ case OPCODE_XPD:
return WRITEMASK_XYZ;
default:
return WRITEMASK_XYZW;
@@ -396,6 +397,26 @@ find_next_temp_use(const struct gl_program *prog, GLuint start, GLuint index)
return END;
}
+static GLboolean _mesa_is_flow_control_opcode(enum prog_opcode opcode)
+{
+ switch (opcode) {
+ case OPCODE_BGNLOOP:
+ case OPCODE_BGNSUB:
+ case OPCODE_BRA:
+ case OPCODE_CAL:
+ case OPCODE_CONT:
+ case OPCODE_IF:
+ case OPCODE_ELSE:
+ case OPCODE_END:
+ case OPCODE_ENDIF:
+ case OPCODE_ENDLOOP:
+ case OPCODE_ENDSUB:
+ case OPCODE_RET:
+ return GL_TRUE;
+ default:
+ return GL_FALSE;
+ }
+}
/**
* Try to remove use of extraneous MOV instructions, to free them up for dead
@@ -404,7 +425,7 @@ find_next_temp_use(const struct gl_program *prog, GLuint start, GLuint index)
static void
_mesa_remove_extra_move_use(struct gl_program *prog)
{
- GLuint i;
+ GLuint i, j;
if (dbg) {
_mesa_printf("Optimize: Begin remove extra move use\n");
@@ -414,63 +435,86 @@ _mesa_remove_extra_move_use(struct gl_program *prog)
/*
* Look for sequences such as this:
* MOV tmpX, arg0;
+ * ...
* FOO tmpY, tmpX, arg1;
* and convert into:
* MOV tmpX, arg0;
+ * ...
* FOO tmpY, arg0, arg1;
*/
for (i = 0; i < prog->NumInstructions - 1; i++) {
const struct prog_instruction *mov = prog->Instructions + i;
- struct prog_instruction *inst2 = prog->Instructions + i + 1;
- int arg;
if (mov->Opcode != OPCODE_MOV ||
mov->DstReg.File != PROGRAM_TEMPORARY ||
mov->DstReg.RelAddr ||
mov->DstReg.CondMask != COND_TR ||
- mov->SaturateMode != SATURATE_OFF)
+ mov->SaturateMode != SATURATE_OFF ||
+ mov->SrcReg[0].RelAddr)
continue;
- for (arg = 0; arg < _mesa_num_inst_src_regs(inst2->Opcode); arg++) {
- int comp;
- int read_mask = get_src_arg_mask(inst2, arg);
+ /* Walk through remaining instructions until the or src reg gets
+ * rewritten or we get into some flow-control, eliminating the use of
+ * this MOV.
+ */
+ for (j = i + 1; j < prog->NumInstructions; j++) {
+ struct prog_instruction *inst2 = prog->Instructions + j;
+ int arg;
- if (inst2->SrcReg[arg].File != mov->DstReg.File ||
- inst2->SrcReg[arg].Index != mov->DstReg.Index ||
- inst2->SrcReg[arg].RelAddr ||
- inst2->SrcReg[arg].Abs)
- continue;
+ if (_mesa_is_flow_control_opcode(inst2->Opcode))
+ break;
- /* Check that all the sources for this arg of inst2 come from inst1
- * or constants.
- */
- for (comp = 0; comp < 4; comp++) {
- int src_swz = GET_SWZ(inst2->SrcReg[arg].Swizzle, comp);
+ /* First rewrite this instruction's args if appropriate. */
+ for (arg = 0; arg < _mesa_num_inst_src_regs(inst2->Opcode); arg++) {
+ int comp;
+ int read_mask = get_src_arg_mask(inst2, arg);
- /* If the MOV didn't write that channel, can't use it. */
- if ((read_mask & (1 << comp)) &&
- src_swz <= SWIZZLE_W &&
- (mov->DstReg.WriteMask & (1 << src_swz)) == 0)
- break;
- }
- if (comp != 4)
- continue;
+ if (inst2->SrcReg[arg].File != mov->DstReg.File ||
+ inst2->SrcReg[arg].Index != mov->DstReg.Index ||
+ inst2->SrcReg[arg].RelAddr ||
+ inst2->SrcReg[arg].Abs)
+ continue;
- /* Adjust the swizzles of inst2 to point at MOV's source */
- for (comp = 0; comp < 4; comp++) {
- int inst2_swz = GET_SWZ(inst2->SrcReg[arg].Swizzle, comp);
+ /* Check that all the sources for this arg of inst2 come from inst1
+ * or constants.
+ */
+ for (comp = 0; comp < 4; comp++) {
+ int src_swz = GET_SWZ(inst2->SrcReg[arg].Swizzle, comp);
+
+ /* If the MOV didn't write that channel, can't use it. */
+ if ((read_mask & (1 << comp)) &&
+ src_swz <= SWIZZLE_W &&
+ (mov->DstReg.WriteMask & (1 << src_swz)) == 0)
+ break;
+ }
+ if (comp != 4)
+ continue;
- if (inst2_swz <= SWIZZLE_W) {
- GLuint s = GET_SWZ(mov->SrcReg[0].Swizzle, inst2_swz);
- inst2->SrcReg[arg].Swizzle &= ~(7 << (3 * comp));
- inst2->SrcReg[arg].Swizzle |= s << (3 * comp);
- inst2->SrcReg[arg].Negate ^= (((mov->SrcReg[0].Negate >>
- inst2_swz) & 0x1) << comp);
+ /* Adjust the swizzles of inst2 to point at MOV's source */
+ for (comp = 0; comp < 4; comp++) {
+ int inst2_swz = GET_SWZ(inst2->SrcReg[arg].Swizzle, comp);
+
+ if (inst2_swz <= SWIZZLE_W) {
+ GLuint s = GET_SWZ(mov->SrcReg[0].Swizzle, inst2_swz);
+ inst2->SrcReg[arg].Swizzle &= ~(7 << (3 * comp));
+ inst2->SrcReg[arg].Swizzle |= s << (3 * comp);
+ inst2->SrcReg[arg].Negate ^= (((mov->SrcReg[0].Negate >>
+ inst2_swz) & 0x1) << comp);
+ }
}
+ inst2->SrcReg[arg].File = mov->SrcReg[0].File;
+ inst2->SrcReg[arg].Index = mov->SrcReg[0].Index;
}
- inst2->SrcReg[arg].File = mov->SrcReg[0].File;
- inst2->SrcReg[arg].Index = mov->SrcReg[0].Index;
+
+ /* If this instruction overwrote part of the move, our time is up. */
+ if ((inst2->DstReg.File == mov->DstReg.File &&
+ (inst2->DstReg.RelAddr ||
+ inst2->DstReg.Index == mov->DstReg.Index)) ||
+ (inst2->DstReg.File == mov->SrcReg[0].File &&
+ (inst2->DstReg.RelAddr ||
+ inst2->DstReg.Index == mov->SrcReg[0].Index)))
+ break;
}
}
diff --git a/src/mesa/shader/prog_print.c b/src/mesa/shader/prog_print.c
index ba4d39452f..52c102cbaa 100644
--- a/src/mesa/shader/prog_print.c
+++ b/src/mesa/shader/prog_print.c
@@ -826,11 +826,11 @@ _mesa_print_program(const struct gl_program *prog)
* XXX move to imports.[ch] if useful elsewhere.
*/
static const char *
-binary(GLbitfield val)
+binary(GLbitfield64 val)
{
- static char buf[50];
+ static char buf[80];
GLint i, len = 0;
- for (i = 31; i >= 0; --i) {
+ for (i = 63; i >= 0; --i) {
if (val & (1 << i))
buf[len++] = '1';
else if (len > 0 || i == 0)
diff --git a/src/mesa/shader/program_parse.tab.c b/src/mesa/shader/program_parse.tab.c
index b9ef88b64b..d4f8429488 100644
--- a/src/mesa/shader/program_parse.tab.c
+++ b/src/mesa/shader/program_parse.tab.c
@@ -2622,7 +2622,7 @@ yyreduce:
YYERROR;
}
- state->prog->OutputsWritten |= (1U << (yyval.dst_reg).Index);
+ state->prog->OutputsWritten |= BITFIELD64_BIT((yyval.dst_reg).Index);
}
;}
break;
diff --git a/src/mesa/shader/program_parse.y b/src/mesa/shader/program_parse.y
index d07bf85b36..8ca6f9805b 100644
--- a/src/mesa/shader/program_parse.y
+++ b/src/mesa/shader/program_parse.y
@@ -643,7 +643,7 @@ maskedDstReg: dstReg optionalMask optionalCcMask
YYERROR;
}
- state->prog->OutputsWritten |= (1U << $$.Index);
+ state->prog->OutputsWritten |= BITFIELD64_BIT($$.Index);
}
}
;
diff --git a/src/mesa/shader/program_parser.h b/src/mesa/shader/program_parser.h
index bce6041381..c170948f73 100644
--- a/src/mesa/shader/program_parser.h
+++ b/src/mesa/shader/program_parser.h
@@ -35,7 +35,7 @@ enum asm_type {
at_attrib,
at_param,
at_temp,
- at_output,
+ at_output
};
struct asm_symbol {
diff --git a/src/mesa/shader/programopt.c b/src/mesa/shader/programopt.c
index 3b8529592d..a0daac1b80 100644
--- a/src/mesa/shader/programopt.c
+++ b/src/mesa/shader/programopt.c
@@ -109,7 +109,7 @@ _mesa_insert_mvp_dp4_code(GLcontext *ctx, struct gl_vertex_program *vprog)
vprog->Base.Instructions = newInst;
vprog->Base.NumInstructions = newLen;
vprog->Base.InputsRead |= VERT_BIT_POS;
- vprog->Base.OutputsWritten |= (1 << VERT_RESULT_HPOS);
+ vprog->Base.OutputsWritten |= BITFIELD64_BIT(VERT_RESULT_HPOS);
}
@@ -211,7 +211,7 @@ _mesa_insert_mvp_mad_code(GLcontext *ctx, struct gl_vertex_program *vprog)
vprog->Base.Instructions = newInst;
vprog->Base.NumInstructions = newLen;
vprog->Base.InputsRead |= VERT_BIT_POS;
- vprog->Base.OutputsWritten |= (1 << VERT_RESULT_HPOS);
+ vprog->Base.OutputsWritten |= BITFIELD64_BIT(VERT_RESULT_HPOS);
}
@@ -613,7 +613,7 @@ _mesa_nop_fragment_program(GLcontext *ctx, struct gl_fragment_program *prog)
prog->Base.Instructions = inst;
prog->Base.NumInstructions = 2;
prog->Base.InputsRead = 1 << inputAttr;
- prog->Base.OutputsWritten = 1 << FRAG_RESULT_COLOR;
+ prog->Base.OutputsWritten = BITFIELD64_BIT(FRAG_RESULT_COLOR);
}
@@ -657,7 +657,7 @@ _mesa_nop_vertex_program(GLcontext *ctx, struct gl_vertex_program *prog)
prog->Base.Instructions = inst;
prog->Base.NumInstructions = 2;
prog->Base.InputsRead = 1 << inputAttr;
- prog->Base.OutputsWritten = 1 << VERT_RESULT_COL0;
+ prog->Base.OutputsWritten = BITFIELD64_BIT(VERT_RESULT_COL0);
/*
* Now insert code to do standard modelview/projection transformation.
diff --git a/src/mesa/shader/slang/slang_codegen.c b/src/mesa/shader/slang/slang_codegen.c
index 344dfdc680..ee5a50ca82 100644
--- a/src/mesa/shader/slang/slang_codegen.c
+++ b/src/mesa/shader/slang/slang_codegen.c
@@ -925,7 +925,7 @@ gen_return_with_expression(slang_assemble_ctx *A, slang_operation *oper)
slang_operation_copy(rhs, &oper->children[0]);
}
- ///blockOper->locals->outer_scope = oper->locals->outer_scope;
+ /*blockOper->locals->outer_scope = oper->locals->outer_scope;*/
/*slang_print_tree(blockOper, 0);*/
diff --git a/src/mesa/shader/slang/slang_link.c b/src/mesa/shader/slang/slang_link.c
index 144c126525..0a2bc49780 100644
--- a/src/mesa/shader/slang/slang_link.c
+++ b/src/mesa/shader/slang/slang_link.c
@@ -515,7 +515,7 @@ _slang_update_inputs_outputs(struct gl_program *prog)
}
if (inst->DstReg.File == PROGRAM_OUTPUT) {
- prog->OutputsWritten |= 1 << inst->DstReg.Index;
+ prog->OutputsWritten |= BITFIELD64_BIT(inst->DstReg.Index);
if (inst->DstReg.RelAddr) {
/* If the output attribute is indexed with relative addressing
* we know that it must be a varying or texcoord such as
@@ -528,14 +528,17 @@ _slang_update_inputs_outputs(struct gl_program *prog)
if (prog->Target == GL_VERTEX_PROGRAM_ARB) {
if (inst->DstReg.Index == VERT_RESULT_TEX0) {
/* mark all texcoord outputs as written */
- const GLbitfield mask =
- ((1 << MAX_TEXTURE_COORD_UNITS) - 1) << VERT_RESULT_TEX0;
+ const GLbitfield64 mask =
+ BITFIELD64_RANGE(VERT_RESULT_TEX0,
+ (VERT_RESULT_TEX0
+ + MAX_TEXTURE_COORD_UNITS - 1));
prog->OutputsWritten |= mask;
}
else if (inst->DstReg.Index == VERT_RESULT_VAR0) {
/* mark all generic varying outputs as written */
- const GLbitfield mask =
- ((1 << MAX_VARYING) - 1) << VERT_RESULT_VAR0;
+ const GLbitfield64 mask =
+ BITFIELD64_RANGE(VERT_RESULT_VAR0,
+ (VERT_RESULT_VAR0 + MAX_VARYING - 1));
prog->OutputsWritten |= mask;
}
}
@@ -807,7 +810,8 @@ _slang_link(GLcontext *ctx,
if (shProg->VertexProgram) {
_slang_update_inputs_outputs(&shProg->VertexProgram->Base);
_slang_count_temporaries(&shProg->VertexProgram->Base);
- if (!(shProg->VertexProgram->Base.OutputsWritten & (1 << VERT_RESULT_HPOS))) {
+ if (!(shProg->VertexProgram->Base.OutputsWritten
+ & BITFIELD64_BIT(VERT_RESULT_HPOS))) {
/* the vertex program did not compute a vertex position */
link_error(shProg,
"gl_Position was not written by vertex shader\n");
@@ -825,7 +829,7 @@ _slang_link(GLcontext *ctx,
if (shProg->FragmentProgram) {
const GLbitfield varyingRead
= shProg->FragmentProgram->Base.InputsRead >> FRAG_ATTRIB_VAR0;
- const GLbitfield varyingWritten = shProg->VertexProgram ?
+ const GLbitfield64 varyingWritten = shProg->VertexProgram ?
shProg->VertexProgram->Base.OutputsWritten >> VERT_RESULT_VAR0 : 0x0;
if ((varyingRead & varyingWritten) != varyingRead) {
link_error(shProg,
@@ -836,9 +840,10 @@ _slang_link(GLcontext *ctx,
/* check that gl_FragColor and gl_FragData are not both written to */
if (shProg->FragmentProgram) {
- GLbitfield outputsWritten = shProg->FragmentProgram->Base.OutputsWritten;
- if ((outputsWritten & ((1 << FRAG_RESULT_COLOR))) &&
- (outputsWritten >= (1 << FRAG_RESULT_DATA0))) {
+ const GLbitfield64 outputsWritten =
+ shProg->FragmentProgram->Base.OutputsWritten;
+ if ((outputsWritten & BITFIELD64_BIT(FRAG_RESULT_COLOR)) &&
+ (outputsWritten >= BITFIELD64_BIT(FRAG_RESULT_DATA0))) {
link_error(shProg, "Fragment program cannot write both gl_FragColor"
" and gl_FragData[].\n");
return;
generated by cgit v1.2.3 (git 2.51.0) at 2025-12-28 01:45:10 +0000