summaryrefslogtreecommitdiff
path: root/src/mesa/shader
diff options
context:
space:
mode:
Diffstat (limited to 'src/mesa/shader')
-rw-r--r--src/mesa/shader/arbprogparse.c113
-rw-r--r--src/mesa/shader/arbprogram.c13
-rw-r--r--src/mesa/shader/nvprogram.c2
-rw-r--r--src/mesa/shader/prog_instruction.c50
-rw-r--r--src/mesa/shader/prog_instruction.h3
-rw-r--r--src/mesa/shader/prog_optimize.c66
-rw-r--r--src/mesa/shader/prog_optimize.h12
-rw-r--r--src/mesa/shader/prog_print.c4
-rw-r--r--src/mesa/shader/programopt.c119
-rw-r--r--src/mesa/shader/shader_api.c10
-rw-r--r--src/mesa/shader/slang/slang_compile.c6
-rw-r--r--src/mesa/shader/slang/slang_link.c17
12 files changed, 347 insertions, 68 deletions
diff --git a/src/mesa/shader/arbprogparse.c b/src/mesa/shader/arbprogparse.c
index b47bf360cf..36267c336f 100644
--- a/src/mesa/shader/arbprogparse.c
+++ b/src/mesa/shader/arbprogparse.c
@@ -563,6 +563,7 @@ struct var_cache
* we take up with our state tokens or constants. Note that
* this is _not_ the same as the number of param registers
* we eventually use */
+ GLuint swizzle; /**< swizzle to access this variable */
struct var_cache *next;
};
@@ -581,6 +582,7 @@ var_cache_create (struct var_cache **va)
(**va).param_binding_begin = ~0;
(**va).param_binding_length = ~0;
(**va).alias_binding = NULL;
+ (**va).swizzle = SWIZZLE_XYZW;
(**va).next = NULL;
}
}
@@ -872,15 +874,16 @@ parse_signed_float (const GLubyte ** inst, struct arb_program *Program)
* This picks out a constant value from the parsed array. The constant vector is r
* returned in the *values array, which should be of length 4.
*
- * \param values - The 4 component vector with the constant value in it
+ * \param values - return the vector constant values.
+ * \param size - returns the number elements in valuesOut [1..4]
*/
static GLvoid
-parse_constant (const GLubyte ** inst, GLfloat *values, struct arb_program *Program,
- GLboolean use)
+parse_constant(const GLubyte ** inst, GLfloat *values, GLint *size,
+ struct arb_program *Program,
+ GLboolean use)
{
GLuint components, i;
-
switch (*(*inst)++) {
case CONSTANT_SCALAR:
if (use == GL_TRUE) {
@@ -893,7 +896,7 @@ parse_constant (const GLubyte ** inst, GLfloat *values, struct arb_program *Prog
values[1] =
values[2] = values[3] = parse_signed_float (inst, Program);
}
-
+ *size = 1;
break;
case CONSTANT_VECTOR:
values[0] = values[1] = values[2] = 0;
@@ -902,7 +905,12 @@ parse_constant (const GLubyte ** inst, GLfloat *values, struct arb_program *Prog
for (i = 0; i < components; i++) {
values[i] = parse_signed_float (inst, Program);
}
+ *size = 4;
break;
+ default:
+ _mesa_problem(NULL, "unexpected case in parse_constant()");
+ values[0] = 0.0F;
+ *size = 0;
}
}
@@ -1513,10 +1521,16 @@ generic_attrib_check(struct var_cache *vc_head)
curr = vc_head;
while (curr) {
if (curr->type == vt_attrib) {
- if (curr->attrib_is_generic)
- genericAttrib[ curr->attrib_binding ] = GL_TRUE;
- else
+ if (curr->attrib_is_generic) {
+ GLuint attr = (curr->attrib_binding == 0)
+ ? 0 : (curr->attrib_binding - VERT_ATTRIB_GENERIC0);
+ assert(attr < MAX_VERTEX_PROGRAM_ATTRIBS);
+ genericAttrib[attr] = GL_TRUE;
+ }
+ else {
+ assert(curr->attrib_binding < MAX_VERTEX_PROGRAM_ATTRIBS);
explicitAttrib[ curr->attrib_binding ] = GL_TRUE;
+ }
}
curr = curr->next;
@@ -1810,7 +1824,6 @@ parse_param_elements (GLcontext * ctx, const GLubyte ** inst,
GLint idx;
GLuint err = 0;
gl_state_index state_tokens[STATE_LENGTH] = {0, 0, 0, 0, 0};
- GLfloat const_values[4];
GLubyte token = *(*inst)++;
@@ -1902,18 +1915,31 @@ parse_param_elements (GLcontext * ctx, const GLubyte ** inst,
case PARAM_CONSTANT:
/* parsing something like {1.0, 2.0, 3.0, 4.0} */
- parse_constant (inst, const_values, Program, use);
- idx = _mesa_add_named_constant(Program->Base.Parameters,
- (char *) param_var->name,
- const_values, 4);
- if (param_var->param_binding_begin == ~0U)
- param_var->param_binding_begin = idx;
- param_var->param_binding_type = PROGRAM_STATE_VAR;
- /* Note: when we reference this parameter in an instruction later,
- * we'll check if it's really a constant/immediate and set the
- * instruction register type appropriately.
- */
- param_var->param_binding_length++;
+ {
+ GLfloat const_values[4];
+ GLint size;
+ parse_constant(inst, const_values, &size, Program, use);
+ if (param_var->name[0] == ' ') {
+ /* this is an unnamed constant */
+ idx = _mesa_add_unnamed_constant(Program->Base.Parameters,
+ const_values, size,
+ &param_var->swizzle);
+ }
+ else {
+ /* named parameter/constant */
+ idx = _mesa_add_named_constant(Program->Base.Parameters,
+ (char *) param_var->name,
+ const_values, size);
+ }
+ if (param_var->param_binding_begin == ~0U)
+ param_var->param_binding_begin = idx;
+ param_var->param_binding_type = PROGRAM_STATE_VAR;
+ /* Note: when we reference this parameter in an instruction later,
+ * we'll check if it's really a constant/immediate and set the
+ * instruction register type appropriately.
+ */
+ param_var->param_binding_length++;
+ }
break;
default:
@@ -2422,6 +2448,9 @@ parse_swizzle_mask(const GLubyte ** inst, GLubyte *swizzle, GLint len)
return;
}
}
+
+ if (len == 1)
+ swizzle[1] = swizzle[2] = swizzle[3] = swizzle[0];
}
@@ -2476,7 +2505,7 @@ static GLuint
parse_src_reg (GLcontext * ctx, const GLubyte ** inst,
struct var_cache **vc_head,
struct arb_program *Program,
- gl_register_file * File, GLint * Index,
+ gl_register_file * File, GLint * Index, GLuint *swizzle,
GLboolean *IsRelOffset )
{
struct var_cache *src;
@@ -2485,6 +2514,8 @@ parse_src_reg (GLcontext * ctx, const GLubyte ** inst,
*IsRelOffset = 0;
+ *swizzle = SWIZZLE_XYZW; /* default */
+
/* And the binding for the src */
switch (*(*inst)++) {
case REGISTER_ATTRIB:
@@ -2540,6 +2571,7 @@ parse_src_reg (GLcontext * ctx, const GLubyte ** inst,
}
*Index = src->param_binding_begin + offset;
+ *swizzle = src->swizzle;
break;
case ARRAY_INDEX_RELATIVE:
@@ -2562,6 +2594,7 @@ parse_src_reg (GLcontext * ctx, const GLubyte ** inst,
/* And store it properly */
*Index = src->param_binding_begin + rel_off;
*IsRelOffset = 1;
+ *swizzle = src->swizzle;
}
break;
}
@@ -2573,6 +2606,7 @@ parse_src_reg (GLcontext * ctx, const GLubyte ** inst,
*File = (gl_register_file) src->param_binding_type;
*Index = src->param_binding_begin;
+ *swizzle = src->swizzle;
break;
}
break;
@@ -2641,6 +2675,21 @@ parse_src_reg (GLcontext * ctx, const GLubyte ** inst,
}
+static GLuint
+swizzle_swizzle(GLuint baseSwizzle, const GLubyte swizzle[4])
+{
+ GLuint i, swz, s[4];
+ for (i = 0; i < 4; i++) {
+ GLuint c = swizzle[i];
+ if (c <= SWIZZLE_W)
+ s[i] = GET_SWZ(baseSwizzle, c);
+ else
+ s[i] = c;
+ }
+ swz = MAKE_SWIZZLE4(s[0], s[1], s[2], s[3]);
+ return swz;
+}
+
/**
* Parse vertex/fragment program vector source register.
*/
@@ -2655,12 +2704,14 @@ parse_vector_src_reg(GLcontext *ctx, const GLubyte **inst,
GLubyte negateMask;
GLubyte swizzle[4];
GLboolean isRelOffset;
+ GLuint baseSwizzle;
/* Grab the sign */
negateMask = (parse_sign (inst) == -1) ? NEGATE_XYZW : NEGATE_NONE;
/* And the src reg */
- if (parse_src_reg(ctx, inst, vc_head, program, &file, &index, &isRelOffset))
+ if (parse_src_reg(ctx, inst, vc_head, program, &file, &index, &baseSwizzle,
+ &isRelOffset))
return 1;
/* finally, the swizzle */
@@ -2668,7 +2719,7 @@ parse_vector_src_reg(GLcontext *ctx, const GLubyte **inst,
reg->File = file;
reg->Index = index;
- reg->Swizzle = MAKE_SWIZZLE4(swizzle[0], swizzle[1], swizzle[2], swizzle[3]);
+ reg->Swizzle = swizzle_swizzle(baseSwizzle, swizzle);
reg->Negate = negateMask;
reg->RelAddr = isRelOffset;
return 0;
@@ -2689,12 +2740,14 @@ parse_scalar_src_reg(GLcontext *ctx, const GLubyte **inst,
GLubyte negateMask;
GLubyte swizzle[4];
GLboolean isRelOffset;
+ GLuint baseSwizzle;
/* Grab the sign */
negateMask = (parse_sign (inst) == -1) ? NEGATE_XYZW : NEGATE_NONE;
/* And the src reg */
- if (parse_src_reg(ctx, inst, vc_head, program, &file, &index, &isRelOffset))
+ if (parse_src_reg(ctx, inst, vc_head, program, &file, &index, &baseSwizzle,
+ &isRelOffset))
return 1;
/* finally, the swizzle */
@@ -2702,7 +2755,7 @@ parse_scalar_src_reg(GLcontext *ctx, const GLubyte **inst,
reg->File = file;
reg->Index = index;
- reg->Swizzle = (swizzle[0] << 0);
+ reg->Swizzle = swizzle_swizzle(baseSwizzle, swizzle);
reg->Negate = negateMask;
reg->RelAddr = isRelOffset;
return 0;
@@ -3013,8 +3066,10 @@ parse_fp_instruction (GLcontext * ctx, const GLubyte ** inst,
GLubyte negateMask;
gl_register_file file;
GLint index;
+ GLuint baseSwizzle;
- if (parse_src_reg(ctx, inst, vc_head, Program, &file, &index, &rel))
+ if (parse_src_reg(ctx, inst, vc_head, Program, &file, &index,
+ &baseSwizzle, &rel))
return 1;
parse_extended_swizzle_mask(inst, swizzle, &negateMask);
fp->SrcReg[0].File = file;
@@ -3354,11 +3409,13 @@ parse_vp_instruction (GLcontext * ctx, const GLubyte ** inst,
GLboolean relAddr;
gl_register_file file;
GLint index;
+ GLuint baseSwizzle;
if (parse_dst_reg(ctx, inst, vc_head, Program, &vp->DstReg))
return 1;
- if (parse_src_reg(ctx, inst, vc_head, Program, &file, &index, &relAddr))
+ if (parse_src_reg(ctx, inst, vc_head, Program, &file, &index,
+ &baseSwizzle, &relAddr))
return 1;
parse_extended_swizzle_mask (inst, swizzle, &negateMask);
vp->SrcReg[0].File = file;
diff --git a/src/mesa/shader/arbprogram.c b/src/mesa/shader/arbprogram.c
index 981565ab8f..317d623a22 100644
--- a/src/mesa/shader/arbprogram.c
+++ b/src/mesa/shader/arbprogram.c
@@ -74,8 +74,6 @@ _mesa_BindProgram(GLenum target, GLuint id)
GET_CURRENT_CONTEXT(ctx);
ASSERT_OUTSIDE_BEGIN_END(ctx);
- FLUSH_VERTICES(ctx, _NEW_PROGRAM);
-
/* Error-check target and get curProg */
if ((target == GL_VERTEX_PROGRAM_ARB) && /* == GL_VERTEX_PROGRAM_NV */
(ctx->Extensions.NV_vertex_program ||
@@ -132,6 +130,9 @@ _mesa_BindProgram(GLenum target, GLuint id)
return;
}
+ /* signal new program (and its new constants) */
+ FLUSH_VERTICES(ctx, _NEW_PROGRAM | _NEW_PROGRAM_CONSTANTS);
+
/* bind newProg */
if (target == GL_VERTEX_PROGRAM_ARB) { /* == GL_VERTEX_PROGRAM_NV */
_mesa_reference_vertprog(ctx, &ctx->VertexProgram.Current,
@@ -489,7 +490,7 @@ _mesa_ProgramEnvParameter4fARB(GLenum target, GLuint index,
GET_CURRENT_CONTEXT(ctx);
ASSERT_OUTSIDE_BEGIN_END(ctx);
- FLUSH_VERTICES(ctx, _NEW_PROGRAM | _NEW_PROGRAM_CONSTANTS);
+ FLUSH_VERTICES(ctx, _NEW_PROGRAM_CONSTANTS);
if (target == GL_FRAGMENT_PROGRAM_ARB
&& ctx->Extensions.ARB_fragment_program) {
@@ -537,7 +538,7 @@ _mesa_ProgramEnvParameters4fvEXT(GLenum target, GLuint index, GLsizei count,
GLfloat * dest;
ASSERT_OUTSIDE_BEGIN_END(ctx);
- FLUSH_VERTICES(ctx, _NEW_PROGRAM | _NEW_PROGRAM_CONSTANTS);
+ FLUSH_VERTICES(ctx, _NEW_PROGRAM_CONSTANTS);
if (count <= 0) {
_mesa_error(ctx, GL_INVALID_VALUE, "glProgramEnvParameters4fv(count)");
@@ -631,7 +632,7 @@ _mesa_ProgramLocalParameter4fARB(GLenum target, GLuint index,
struct gl_program *prog;
ASSERT_OUTSIDE_BEGIN_END(ctx);
- FLUSH_VERTICES(ctx, _NEW_PROGRAM | _NEW_PROGRAM_CONSTANTS);
+ FLUSH_VERTICES(ctx, _NEW_PROGRAM_CONSTANTS);
if ((target == GL_FRAGMENT_PROGRAM_NV
&& ctx->Extensions.NV_fragment_program) ||
@@ -685,7 +686,7 @@ _mesa_ProgramLocalParameters4fvEXT(GLenum target, GLuint index, GLsizei count,
GLint i;
ASSERT_OUTSIDE_BEGIN_END(ctx);
- FLUSH_VERTICES(ctx, _NEW_PROGRAM | _NEW_PROGRAM_CONSTANTS);
+ FLUSH_VERTICES(ctx, _NEW_PROGRAM_CONSTANTS);
if (count <= 0) {
_mesa_error(ctx, GL_INVALID_VALUE, "glProgramLocalParameters4fv(count)");
diff --git a/src/mesa/shader/nvprogram.c b/src/mesa/shader/nvprogram.c
index 5142c2a4a5..8ba521182b 100644
--- a/src/mesa/shader/nvprogram.c
+++ b/src/mesa/shader/nvprogram.c
@@ -706,7 +706,7 @@ _mesa_ProgramNamedParameter4fNV(GLuint id, GLsizei len, const GLubyte *name,
GET_CURRENT_CONTEXT(ctx);
ASSERT_OUTSIDE_BEGIN_END(ctx);
- FLUSH_VERTICES(ctx, _NEW_PROGRAM | _NEW_PROGRAM_CONSTANTS);
+ FLUSH_VERTICES(ctx, _NEW_PROGRAM_CONSTANTS);
prog = _mesa_lookup_program(ctx, id);
if (!prog || prog->Target != GL_FRAGMENT_PROGRAM_NV) {
diff --git a/src/mesa/shader/prog_instruction.c b/src/mesa/shader/prog_instruction.c
index ca7565c091..ae3a003fee 100644
--- a/src/mesa/shader/prog_instruction.c
+++ b/src/mesa/shader/prog_instruction.c
@@ -286,6 +286,56 @@ _mesa_is_tex_instruction(gl_inst_opcode opcode)
/**
+ * Check if there's a potential src/dst register data dependency when
+ * using SOA execution.
+ * Example:
+ * MOV T, T.yxwz;
+ * This would expand into:
+ * MOV t0, t1;
+ * MOV t1, t0;
+ * MOV t2, t3;
+ * MOV t3, t2;
+ * The second instruction will have the wrong value for t0 if executed as-is.
+ */
+GLboolean
+_mesa_check_soa_dependencies(const struct prog_instruction *inst)
+{
+ GLuint i, chan;
+
+ if (inst->DstReg.WriteMask == WRITEMASK_X ||
+ inst->DstReg.WriteMask == WRITEMASK_Y ||
+ inst->DstReg.WriteMask == WRITEMASK_Z ||
+ inst->DstReg.WriteMask == WRITEMASK_W ||
+ inst->DstReg.WriteMask == 0x0) {
+ /* no chance of data dependency */
+ return GL_FALSE;
+ }
+
+ /* loop over src regs */
+ for (i = 0; i < 3; i++) {
+ if (inst->SrcReg[i].File == inst->DstReg.File &&
+ inst->SrcReg[i].Index == inst->DstReg.Index) {
+ /* loop over dest channels */
+ GLuint channelsWritten = 0x0;
+ for (chan = 0; chan < 4; chan++) {
+ if (inst->DstReg.WriteMask & (1 << chan)) {
+ /* check if we're reading a channel that's been written */
+ GLuint swizzle = GET_SWZ(inst->SrcReg[i].Swizzle, chan);
+ if (swizzle <= SWIZZLE_W &&
+ (channelsWritten & (1 << swizzle))) {
+ return GL_TRUE;
+ }
+
+ channelsWritten |= (1 << chan);
+ }
+ }
+ }
+ }
+ return GL_FALSE;
+}
+
+
+/**
* Return string name for given program opcode.
*/
const char *
diff --git a/src/mesa/shader/prog_instruction.h b/src/mesa/shader/prog_instruction.h
index 3109f6cbae..40ad998f79 100644
--- a/src/mesa/shader/prog_instruction.h
+++ b/src/mesa/shader/prog_instruction.h
@@ -428,6 +428,9 @@ _mesa_num_inst_dst_regs(gl_inst_opcode opcode);
extern GLboolean
_mesa_is_tex_instruction(gl_inst_opcode opcode);
+extern GLboolean
+_mesa_check_soa_dependencies(const struct prog_instruction *inst);
+
extern const char *
_mesa_opcode_string(gl_inst_opcode opcode);
diff --git a/src/mesa/shader/prog_optimize.c b/src/mesa/shader/prog_optimize.c
index 6ba2e76ff9..be903106a0 100644
--- a/src/mesa/shader/prog_optimize.c
+++ b/src/mesa/shader/prog_optimize.c
@@ -547,15 +547,13 @@ update_interval(GLint intBegin[], GLint intEnd[], GLuint index, GLuint ic)
/**
- * Find the live intervals for each temporary register in the program.
- * For register R, the interval [A,B] indicates that R is referenced
- * from instruction A through instruction B.
- * Special consideration is needed for loops and subroutines.
- * \return GL_TRUE if success, GL_FALSE if we cannot proceed for some reason
+ * Find first/last instruction that references each temporary register.
*/
-static GLboolean
-find_live_intervals(struct gl_program *prog,
- struct interval_list *liveIntervals)
+GLboolean
+_mesa_find_temp_intervals(const struct prog_instruction *instructions,
+ GLuint numInstructions,
+ GLint intBegin[MAX_PROGRAM_TEMPS],
+ GLint intEnd[MAX_PROGRAM_TEMPS])
{
struct loop_info
{
@@ -563,26 +561,15 @@ find_live_intervals(struct gl_program *prog,
};
struct loop_info loopStack[MAX_LOOP_NESTING];
GLuint loopStackDepth = 0;
- GLint intBegin[MAX_PROGRAM_TEMPS], intEnd[MAX_PROGRAM_TEMPS];
GLuint i;
- /*
- * Note: we'll return GL_FALSE below if we find relative indexing
- * into the TEMP register file. We can't handle that yet.
- * We also give up on subroutines for now.
- */
-
- if (dbg) {
- _mesa_printf("Optimize: Begin find intervals\n");
- }
-
for (i = 0; i < MAX_PROGRAM_TEMPS; i++){
intBegin[i] = intEnd[i] = -1;
}
/* Scan instructions looking for temporary registers */
- for (i = 0; i < prog->NumInstructions; i++) {
- const struct prog_instruction *inst = prog->Instructions + i;
+ for (i = 0; i < numInstructions; i++) {
+ const struct prog_instruction *inst = instructions + i;
if (inst->Opcode == OPCODE_BGNLOOP) {
loopStack[loopStackDepth].Start = i;
loopStack[loopStackDepth].End = inst->BranchTarget;
@@ -595,7 +582,7 @@ find_live_intervals(struct gl_program *prog,
return GL_FALSE;
}
else {
- const GLuint numSrc = _mesa_num_inst_src_regs(inst->Opcode);
+ const GLuint numSrc = 3;/*_mesa_num_inst_src_regs(inst->Opcode);*/
GLuint j;
for (j = 0; j < numSrc; j++) {
if (inst->SrcReg[j].File == PROGRAM_TEMPORARY) {
@@ -624,6 +611,39 @@ find_live_intervals(struct gl_program *prog,
}
}
+ return GL_TRUE;
+}
+
+
+/**
+ * Find the live intervals for each temporary register in the program.
+ * For register R, the interval [A,B] indicates that R is referenced
+ * from instruction A through instruction B.
+ * Special consideration is needed for loops and subroutines.
+ * \return GL_TRUE if success, GL_FALSE if we cannot proceed for some reason
+ */
+static GLboolean
+find_live_intervals(struct gl_program *prog,
+ struct interval_list *liveIntervals)
+{
+ GLint intBegin[MAX_PROGRAM_TEMPS], intEnd[MAX_PROGRAM_TEMPS];
+ GLuint i;
+
+ /*
+ * Note: we'll return GL_FALSE below if we find relative indexing
+ * into the TEMP register file. We can't handle that yet.
+ * We also give up on subroutines for now.
+ */
+
+ if (dbg) {
+ _mesa_printf("Optimize: Begin find intervals\n");
+ }
+
+ /* build intermediate arrays */
+ if (!_mesa_find_temp_intervals(prog->Instructions, prog->NumInstructions,
+ intBegin, intEnd))
+ return GL_FALSE;
+
/* Build live intervals list from intermediate arrays */
liveIntervals->Num = 0;
for (i = 0; i < MAX_PROGRAM_TEMPS; i++) {
@@ -792,8 +812,6 @@ _mesa_reallocate_registers(struct gl_program *prog)
}
-
-
/**
* Apply optimizations to the given program to eliminate unnecessary
* instructions, temp regs, etc.
diff --git a/src/mesa/shader/prog_optimize.h b/src/mesa/shader/prog_optimize.h
index d102cfd9fc..43894a2723 100644
--- a/src/mesa/shader/prog_optimize.h
+++ b/src/mesa/shader/prog_optimize.h
@@ -25,7 +25,19 @@
#ifndef PROG_OPT_H
#define PROG_OPT_H
+
+#include "main/config.h"
+
+
struct gl_program;
+struct prog_instruction;
+
+
+extern GLboolean
+_mesa_find_temp_intervals(const struct prog_instruction *instructions,
+ GLuint numInstructions,
+ GLint intBegin[MAX_PROGRAM_TEMPS],
+ GLint intEnd[MAX_PROGRAM_TEMPS]);
extern void
_mesa_optimize_program(GLcontext *ctx, struct gl_program *program);
diff --git a/src/mesa/shader/prog_print.c b/src/mesa/shader/prog_print.c
index 9967f2978d..e6f9a91069 100644
--- a/src/mesa/shader/prog_print.c
+++ b/src/mesa/shader/prog_print.c
@@ -941,6 +941,10 @@ _mesa_write_shader_to_file(const struct gl_shader *shader)
fprintf(f, "/*\n");
_mesa_fprint_program_opt(f, shader->Program, PROG_PRINT_DEBUG, GL_TRUE);
fprintf(f, "*/\n");
+ fprintf(f, "/* Parameters / constants */\n");
+ fprintf(f, "/*\n");
+ _mesa_fprint_parameter_list(f, shader->Program->Parameters);
+ fprintf(f, "*/\n");
}
fclose(f);
diff --git a/src/mesa/shader/programopt.c b/src/mesa/shader/programopt.c
index ecd98dc85c..f70c75cec8 100644
--- a/src/mesa/shader/programopt.c
+++ b/src/mesa/shader/programopt.c
@@ -45,8 +45,8 @@
* into a vertex program.
* May be used to implement the position_invariant option.
*/
-void
-_mesa_insert_mvp_code(GLcontext *ctx, struct gl_vertex_program *vprog)
+static void
+_mesa_insert_mvp_dp4_code(GLcontext *ctx, struct gl_vertex_program *vprog)
{
struct prog_instruction *newInst;
const GLuint origLen = vprog->Base.NumInstructions;
@@ -113,6 +113,121 @@ _mesa_insert_mvp_code(GLcontext *ctx, struct gl_vertex_program *vprog)
}
+static void
+_mesa_insert_mvp_mad_code(GLcontext *ctx, struct gl_vertex_program *vprog)
+{
+ struct prog_instruction *newInst;
+ const GLuint origLen = vprog->Base.NumInstructions;
+ const GLuint newLen = origLen + 4;
+ GLuint hposTemp;
+ GLuint i;
+
+ /*
+ * Setup state references for the modelview/projection matrix.
+ * XXX we should check if these state vars are already declared.
+ */
+ static const gl_state_index mvpState[4][STATE_LENGTH] = {
+ { STATE_MVP_MATRIX, 0, 0, 0, STATE_MATRIX_TRANSPOSE },
+ { STATE_MVP_MATRIX, 0, 1, 1, STATE_MATRIX_TRANSPOSE },
+ { STATE_MVP_MATRIX, 0, 2, 2, STATE_MATRIX_TRANSPOSE },
+ { STATE_MVP_MATRIX, 0, 3, 3, STATE_MATRIX_TRANSPOSE },
+ };
+ GLint mvpRef[4];
+
+ for (i = 0; i < 4; i++) {
+ mvpRef[i] = _mesa_add_state_reference(vprog->Base.Parameters,
+ mvpState[i]);
+ }
+
+ /* Alloc storage for new instructions */
+ newInst = _mesa_alloc_instructions(newLen);
+ if (!newInst) {
+ _mesa_error(ctx, GL_OUT_OF_MEMORY,
+ "glProgramString(inserting position_invariant code)");
+ return;
+ }
+
+ /* TEMP hposTemp; */
+ hposTemp = vprog->Base.NumTemporaries++;
+
+ /*
+ * Generated instructions:
+ * emit_op2(p, OPCODE_MUL, tmp, 0, swizzle1(src,X), mat[0]);
+ * emit_op3(p, OPCODE_MAD, tmp, 0, swizzle1(src,Y), mat[1], tmp);
+ * emit_op3(p, OPCODE_MAD, tmp, 0, swizzle1(src,Z), mat[2], tmp);
+ * emit_op3(p, OPCODE_MAD, dest, 0, swizzle1(src,W), mat[3], tmp);
+ */
+ _mesa_init_instructions(newInst, 4);
+
+ newInst[0].Opcode = OPCODE_MUL;
+ newInst[0].DstReg.File = PROGRAM_TEMPORARY;
+ newInst[0].DstReg.Index = hposTemp;
+ newInst[0].DstReg.WriteMask = WRITEMASK_XYZW;
+ newInst[0].SrcReg[0].File = PROGRAM_INPUT;
+ newInst[0].SrcReg[0].Index = VERT_ATTRIB_POS;
+ newInst[0].SrcReg[0].Swizzle = SWIZZLE_XXXX;
+ newInst[0].SrcReg[1].File = PROGRAM_STATE_VAR;
+ newInst[0].SrcReg[1].Index = mvpRef[0];
+ newInst[0].SrcReg[1].Swizzle = SWIZZLE_NOOP;
+
+ for (i = 1; i <= 2; i++) {
+ newInst[i].Opcode = OPCODE_MAD;
+ newInst[i].DstReg.File = PROGRAM_TEMPORARY;
+ newInst[i].DstReg.Index = hposTemp;
+ newInst[i].DstReg.WriteMask = WRITEMASK_XYZW;
+ newInst[i].SrcReg[0].File = PROGRAM_INPUT;
+ newInst[i].SrcReg[0].Index = VERT_ATTRIB_POS;
+ newInst[i].SrcReg[0].Swizzle = MAKE_SWIZZLE4(i,i,i,i);
+ newInst[i].SrcReg[1].File = PROGRAM_STATE_VAR;
+ newInst[i].SrcReg[1].Index = mvpRef[i];
+ newInst[i].SrcReg[1].Swizzle = SWIZZLE_NOOP;
+ newInst[i].SrcReg[2].File = PROGRAM_TEMPORARY;
+ newInst[i].SrcReg[2].Index = hposTemp;
+ newInst[1].SrcReg[2].Swizzle = SWIZZLE_NOOP;
+ }
+
+ newInst[3].Opcode = OPCODE_MAD;
+ newInst[3].DstReg.File = PROGRAM_OUTPUT;
+ newInst[3].DstReg.Index = VERT_RESULT_HPOS;
+ newInst[3].DstReg.WriteMask = WRITEMASK_XYZW;
+ newInst[3].SrcReg[0].File = PROGRAM_INPUT;
+ newInst[3].SrcReg[0].Index = VERT_ATTRIB_POS;
+ newInst[3].SrcReg[0].Swizzle = SWIZZLE_WWWW;
+ newInst[3].SrcReg[1].File = PROGRAM_STATE_VAR;
+ newInst[3].SrcReg[1].Index = mvpRef[3];
+ newInst[3].SrcReg[1].Swizzle = SWIZZLE_NOOP;
+ newInst[3].SrcReg[2].File = PROGRAM_TEMPORARY;
+ newInst[3].SrcReg[2].Index = hposTemp;
+ newInst[3].SrcReg[2].Swizzle = SWIZZLE_NOOP;
+
+
+ /* Append original instructions after new instructions */
+ _mesa_copy_instructions (newInst + 4, vprog->Base.Instructions, origLen);
+
+ /* free old instructions */
+ _mesa_free_instructions(vprog->Base.Instructions, origLen);
+
+ /* install new instructions */
+ vprog->Base.Instructions = newInst;
+ vprog->Base.NumInstructions = newLen;
+ vprog->Base.InputsRead |= VERT_BIT_POS;
+ vprog->Base.OutputsWritten |= (1 << VERT_RESULT_HPOS);
+}
+
+
+void
+_mesa_insert_mvp_code(GLcontext *ctx, struct gl_vertex_program *vprog)
+{
+ if (ctx->mvp_with_dp4)
+ _mesa_insert_mvp_dp4_code( ctx, vprog );
+ else
+ _mesa_insert_mvp_mad_code( ctx, vprog );
+}
+
+
+
+
+
/**
* Append extra instructions onto the given fragment program to implement
diff --git a/src/mesa/shader/shader_api.c b/src/mesa/shader/shader_api.c
index 644cd39185..a8390d3094 100644
--- a/src/mesa/shader/shader_api.c
+++ b/src/mesa/shader/shader_api.c
@@ -1487,7 +1487,7 @@ _mesa_use_program(GLcontext *ctx, GLuint program)
return;
}
- FLUSH_VERTICES(ctx, _NEW_PROGRAM);
+ FLUSH_VERTICES(ctx, _NEW_PROGRAM | _NEW_PROGRAM_CONSTANTS);
if (program) {
shProg = _mesa_lookup_shader_program_err(ctx, program, "glUseProgram");
@@ -1509,6 +1509,10 @@ _mesa_use_program(GLcontext *ctx, GLuint program)
shProg->Shaders[i]->Name,
shProg->Shaders[i]->Type);
}
+ if (shProg->VertexProgram)
+ printf(" vert prog %u\n", shProg->VertexProgram->Base.Id);
+ if (shProg->FragmentProgram)
+ printf(" frag prog %u\n", shProg->FragmentProgram->Base.Id);
}
}
else {
@@ -1789,7 +1793,7 @@ _mesa_uniform(GLcontext *ctx, GLint location, GLsizei count,
return;
}
- FLUSH_VERTICES(ctx, _NEW_PROGRAM | _NEW_PROGRAM_CONSTANTS);
+ FLUSH_VERTICES(ctx, _NEW_PROGRAM_CONSTANTS);
uniform = &shProg->Uniforms->Uniforms[location];
@@ -1929,7 +1933,7 @@ _mesa_uniform_matrix(GLcontext *ctx, GLint cols, GLint rows,
return;
}
- FLUSH_VERTICES(ctx, _NEW_PROGRAM | _NEW_PROGRAM_CONSTANTS);
+ FLUSH_VERTICES(ctx, _NEW_PROGRAM_CONSTANTS);
uniform = &shProg->Uniforms->Uniforms[location];
diff --git a/src/mesa/shader/slang/slang_compile.c b/src/mesa/shader/slang/slang_compile.c
index ba2fc4f85c..d7ad879e97 100644
--- a/src/mesa/shader/slang/slang_compile.c
+++ b/src/mesa/shader/slang/slang_compile.c
@@ -2161,6 +2161,12 @@ parse_function(slang_parse_ctx * C, slang_output_ctx * O, int definition,
(O->funs->num_functions + 1)
* sizeof(slang_function));
if (O->funs->functions == NULL) {
+ /* Make sure that there are no functions marked, as the
+ * allocation is currently NULL, in order to avoid
+ * a potental segfault as we clean up later.
+ */
+ O->funs->num_functions = 0;
+
slang_info_log_memory(C->L);
slang_function_destruct(&parsed_func);
return GL_FALSE;
diff --git a/src/mesa/shader/slang/slang_link.c b/src/mesa/shader/slang/slang_link.c
index 1fdf4db054..2bc8809661 100644
--- a/src/mesa/shader/slang/slang_link.c
+++ b/src/mesa/shader/slang/slang_link.c
@@ -97,7 +97,8 @@ bits_agree(GLbitfield flags1, GLbitfield flags2, GLbitfield bit)
* which inputs are centroid-sampled, invariant, etc.
*/
static GLboolean
-link_varying_vars(struct gl_shader_program *shProg, struct gl_program *prog)
+link_varying_vars(GLcontext *ctx,
+ struct gl_shader_program *shProg, struct gl_program *prog)
{
GLuint *map, i, firstVarying, newFile;
GLbitfield *inOutFlags;
@@ -156,8 +157,12 @@ link_varying_vars(struct gl_shader_program *shProg, struct gl_program *prog)
var->Flags);
}
+ if (shProg->Varying->NumParameters > ctx->Const.MaxVarying) {
+ link_error(shProg, "Too many varying variables");
+ return GL_FALSE;
+ }
+
/* Map varying[i] to varying[j].
- * Plus, set prog->Input/OutputFlags[] as described above.
* Note: the loop here takes care of arrays or large (sz>4) vars.
*/
{
@@ -712,6 +717,8 @@ _slang_link(GLcontext *ctx,
struct gl_vertex_program *linked_vprog =
vertex_program(_mesa_clone_program(ctx, &vertProg->Base));
shProg->VertexProgram = linked_vprog; /* refcount OK */
+ /* vertex program ID not significant; just set Id for debugging purposes */
+ shProg->VertexProgram->Base.Id = shProg->Name;
ASSERT(shProg->VertexProgram->Base.RefCount == 1);
}
@@ -720,16 +727,18 @@ _slang_link(GLcontext *ctx,
struct gl_fragment_program *linked_fprog =
fragment_program(_mesa_clone_program(ctx, &fragProg->Base));
shProg->FragmentProgram = linked_fprog; /* refcount OK */
+ /* vertex program ID not significant; just set Id for debugging purposes */
+ shProg->FragmentProgram->Base.Id = shProg->Name;
ASSERT(shProg->FragmentProgram->Base.RefCount == 1);
}
/* link varying vars */
if (shProg->VertexProgram) {
- if (!link_varying_vars(shProg, &shProg->VertexProgram->Base))
+ if (!link_varying_vars(ctx, shProg, &shProg->VertexProgram->Base))
return;
}
if (shProg->FragmentProgram) {
- if (!link_varying_vars(shProg, &shProg->FragmentProgram->Base))
+ if (!link_varying_vars(ctx, shProg, &shProg->FragmentProgram->Base))
return;
}
generated by cgit v1.2.3 (git 2.25.1) at 2024-12-26 19:31:31 +0000