Dnia niedziela, 7 czerwca 2009 o 23:04:37 Maciej Cencora napisał(a):
> Dnia sobota, 6 czerwca 2009 o 13:47:37 Maciej Cencora napisał(a):
> > Hi,
> >
> > just as title says.
> >
> > Please test and review.
> >
> > Regards,
> > Maciej Cencora
>
> please don't commit it yet. I've found one regression in 2nd patch -
> rendering errors when two sided lighting is enabled.
>
> I've got the fix already, but this time I will give it some more testing.
>
> Maciej Cencora
Ok, I think I have tested it enough. There's one more patch compared to
initial patchset - it fixes GPU lockups for certain vertex shader and pixel
shader combinations.
Maciej Cencora
From b343bb1a1fb16ea345110f2c2108cee3709d3bf8 Mon Sep 17 00:00:00 2001
From: Maciej Cencora <[email protected]>
Date: Fri, 5 Jun 2009 18:00:58 +0200
Subject: [PATCH] r300: hw doesn't support saturation for tex instructions
---
src/mesa/drivers/dri/r300/r300_fragprog.c | 3 +++
1 files changed, 3 insertions(+), 0 deletions(-)
diff --git a/src/mesa/drivers/dri/r300/r300_fragprog.c b/src/mesa/drivers/dri/r300/r300_fragprog.c
index ea530fd..55c1cfe 100644
--- a/src/mesa/drivers/dri/r300/r300_fragprog.c
+++ b/src/mesa/drivers/dri/r300/r300_fragprog.c
@@ -140,6 +140,8 @@ GLboolean r300_transform_TEX(
inst.DstReg.Index = tempreg;
inst.DstReg.WriteMask = WRITEMASK_XYZW;
destredirect = GL_TRUE;
+ } else if (inst.SaturateMode) {
+ destredirect = GL_TRUE;
}
}
@@ -219,6 +221,7 @@ GLboolean r300_transform_TEX(
tgt->Opcode = OPCODE_MOV;
tgt->DstReg = orig_inst->DstReg;
+ tgt->SaturateMode = inst.SaturateMode;
tgt->SrcReg[0].File = PROGRAM_TEMPORARY;
tgt->SrcReg[0].Index = inst.DstReg.Index;
}
--
1.6.0.4
From df37d9fb21838769677e98e040249c7338fa5e05 Mon Sep 17 00:00:00 2001
From: Maciej Cencora <[email protected]>
Date: Fri, 5 Jun 2009 18:32:05 +0200
Subject: [PATCH] r300: fix vertex program bug
If the vertex program didn't write position attribute, the position invariant function would add necessary instructions, but the vertex position would be overwritten by artificial outputs insts added to satisfy fragment program requirements.
Fixes "whole screen is gray" problem for HW TCL path in sauerbraten when shaders are enabled, and whole slew of wine d3d9 tests.
---
src/mesa/drivers/dri/r300/r300_vertprog.c | 16 ++++++++++------
1 files changed, 10 insertions(+), 6 deletions(-)
diff --git a/src/mesa/drivers/dri/r300/r300_vertprog.c b/src/mesa/drivers/dri/r300/r300_vertprog.c
index 66750b1..c41a8fd 100644
--- a/src/mesa/drivers/dri/r300/r300_vertprog.c
+++ b/src/mesa/drivers/dri/r300/r300_vertprog.c
@@ -1456,7 +1456,12 @@ void r300SelectVertexShader(r300ContextPtr r300)
wpos_idx = i;
}
- add_outputs(&wanted_key, VERT_RESULT_HPOS);
+ if (vpc->mesa_program.IsPositionInvariant) {
+ wanted_key.InputsRead |= (1 << VERT_ATTRIB_POS);
+ wanted_key.OutputsWritten |= (1 << VERT_RESULT_HPOS);
+ } else {
+ add_outputs(&wanted_key, VERT_RESULT_HPOS);
+ }
if (InputsRead & FRAG_BIT_COL0) {
add_outputs(&wanted_key, VERT_RESULT_COL0);
@@ -1466,17 +1471,16 @@ void r300SelectVertexShader(r300ContextPtr r300)
add_outputs(&wanted_key, VERT_RESULT_COL1);
}
+ if (InputsRead & FRAG_BIT_FOGC) {
+ add_outputs(&wanted_key, VERT_RESULT_FOGC);
+ }
+
for (i = 0; i < ctx->Const.MaxTextureUnits; i++) {
if (InputsRead & (FRAG_BIT_TEX0 << i)) {
add_outputs(&wanted_key, VERT_RESULT_TEX0 + i);
}
}
- if (vpc->mesa_program.IsPositionInvariant) {
- /* we wan't position don't we ? */
- wanted_key.InputsRead |= (1 << VERT_ATTRIB_POS);
- }
-
for (vp = vpc->progs; vp; vp = vp->next)
if (_mesa_memcmp(&vp->key, &wanted_key, sizeof(wanted_key))
== 0) {
--
1.6.0.4
From 0ff17c915ba4ef9899c24094f3464b216893b174 Mon Sep 17 00:00:00 2001
From: Maciej Cencora <[email protected]>
Date: Fri, 5 Jun 2009 18:14:15 +0200
Subject: [PATCH] r300: fix output register allocation for vertex shaders
If the vertex program wrote secondary color without primary color, the secondary color output register index would be 0 which resulted in overwriting vertex position in some cases.
---
src/mesa/drivers/dri/r300/r300_vertprog.c | 28 +++++++++++++++++++---------
1 files changed, 19 insertions(+), 9 deletions(-)
diff --git a/src/mesa/drivers/dri/r300/r300_vertprog.c b/src/mesa/drivers/dri/r300/r300_vertprog.c
index a74b915..1cacebf 100644
--- a/src/mesa/drivers/dri/r300/r300_vertprog.c
+++ b/src/mesa/drivers/dri/r300/r300_vertprog.c
@@ -955,26 +955,36 @@ static void t_inputs_outputs(struct r300_vertex_program *vp)
vp->outputs[VERT_RESULT_PSIZ] = cur_reg++;
}
+ /* If we're writing back facing colors we need to send
+ * four colors to make front/back face colors selection work.
+ * If the vertex program doesn't write all 4 colors, lets
+ * pretend it does by skipping output index reg so the colors
+ * get written into appropriate output vectors.
+ */
if (vp->key.OutputsWritten & (1 << VERT_RESULT_COL0)) {
vp->outputs[VERT_RESULT_COL0] = cur_reg++;
+ } else if (vp->key.OutputsWritten & (1 << VERT_RESULT_BFC0) ||
+ vp->key.OutputsWritten & (1 << VERT_RESULT_BFC1)) {
+ cur_reg++;
}
if (vp->key.OutputsWritten & (1 << VERT_RESULT_COL1)) {
- vp->outputs[VERT_RESULT_COL1] =
- vp->outputs[VERT_RESULT_COL0] + 1;
- cur_reg = vp->outputs[VERT_RESULT_COL1] + 1;
+ vp->outputs[VERT_RESULT_COL1] = cur_reg++;
+ } else if (vp->key.OutputsWritten & (1 << VERT_RESULT_BFC0) ||
+ vp->key.OutputsWritten & (1 << VERT_RESULT_BFC1)) {
+ cur_reg++;
}
if (vp->key.OutputsWritten & (1 << VERT_RESULT_BFC0)) {
- vp->outputs[VERT_RESULT_BFC0] =
- vp->outputs[VERT_RESULT_COL0] + 2;
- cur_reg = vp->outputs[VERT_RESULT_BFC0] + 2;
+ vp->outputs[VERT_RESULT_BFC0] = cur_reg++;
+ } else if (vp->key.OutputsWritten & (1 << VERT_RESULT_BFC1)) {
+ cur_reg++;
}
if (vp->key.OutputsWritten & (1 << VERT_RESULT_BFC1)) {
- vp->outputs[VERT_RESULT_BFC1] =
- vp->outputs[VERT_RESULT_COL0] + 3;
- cur_reg = vp->outputs[VERT_RESULT_BFC1] + 1;
+ vp->outputs[VERT_RESULT_BFC1] = cur_reg++;
+ } else if (vp->key.OutputsWritten & (1 << VERT_RESULT_BFC0)) {
+ cur_reg++;
}
for (i = VERT_RESULT_TEX0; i <= VERT_RESULT_TEX7; i++) {
--
1.6.0.4
From b41cf9080795f059a36a8f7d89f0868093b60108 Mon Sep 17 00:00:00 2001
From: Maciej Cencora <[email protected]>
Date: Fri, 5 Jun 2009 18:23:55 +0200
Subject: [PATCH] r300: print vertex program when debugging is enabled
---
src/mesa/drivers/dri/r300/r300_vertprog.c | 17 ++++++++++++++---
1 files changed, 14 insertions(+), 3 deletions(-)
diff --git a/src/mesa/drivers/dri/r300/r300_vertprog.c b/src/mesa/drivers/dri/r300/r300_vertprog.c
index 1cacebf..46e1527 100644
--- a/src/mesa/drivers/dri/r300/r300_vertprog.c
+++ b/src/mesa/drivers/dri/r300/r300_vertprog.c
@@ -34,6 +34,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
#include "shader/program.h"
#include "shader/prog_instruction.h"
#include "shader/prog_parameter.h"
+#include "shader/prog_print.h"
#include "shader/prog_statevars.h"
#include "tnl/tnl.h"
@@ -1383,6 +1384,12 @@ static struct r300_vertex_program *build_program(struct r300_vertex_program_key
pos_as_texcoord(vp, &mesa_vp->Base);
}
+ if (RADEON_DEBUG & DEBUG_VERTS) {
+ fprintf(stderr, "Vertex program after native rewrite:\n");
+ _mesa_print_program(&mesa_vp->Base);
+ fflush(stdout);
+ }
+
assert(mesa_vp->Base.NumInstructions);
vp->num_temporaries = mesa_vp->Base.NumTemporaries;
r300TranslateVertexShader(vp, mesa_vp->Base.Instructions);
@@ -1456,7 +1463,12 @@ void r300SelectVertexShader(r300ContextPtr r300)
r300->selected_vp = vp;
return;
}
- //_mesa_print_program(&vpc->mesa_program.Base);
+
+ if (RADEON_DEBUG & DEBUG_VERTS) {
+ fprintf(stderr, "Initial vertex program:\n");
+ _mesa_print_program(&vpc->mesa_program.Base);
+ fflush(stdout);
+ }
vp = build_program(&wanted_key, &vpc->mesa_program, wpos_idx);
vp->next = vpc->progs;
@@ -1518,8 +1530,7 @@ void r300SetupVertexProgram(r300ContextPtr rmesa)
param_count = r300VertexProgUpdateParams(ctx,
(struct r300_vertex_program_cont *)
ctx->VertexProgram._Current,
- (float *)&rmesa->hw.vpp.
- cmd[R300_VPP_PARAM_0]);
+ (float *)&rmesa->hw.vpp.cmd[R300_VPP_PARAM_0]);
bump_vpu_count(rmesa->hw.vpp.cmd, param_count);
param_count /= 4;
--
1.6.0.4
From 1626b837f147f2a320d157b2e57dd01fa5d8d495 Mon Sep 17 00:00:00 2001
From: Maciej Cencora <[email protected]>
Date: Sun, 7 Jun 2009 21:34:44 +0200
Subject: [PATCH] r300: fix a GPU lock up
Sending from VAP more texture coordinates than RS expects results in GPU hang.
Fixes BumpSelfShadow from DirectX8 SDK.
---
src/mesa/drivers/dri/r300/r300_emit.c | 33 +++++++++++++++++--------------
src/mesa/drivers/dri/r300/r300_emit.h | 4 +-
src/mesa/drivers/dri/r300/r300_state.c | 8 +++---
3 files changed, 24 insertions(+), 21 deletions(-)
diff --git a/src/mesa/drivers/dri/r300/r300_emit.c b/src/mesa/drivers/dri/r300/r300_emit.c
index cf31596..4017224 100644
--- a/src/mesa/drivers/dri/r300/r300_emit.c
+++ b/src/mesa/drivers/dri/r300/r300_emit.c
@@ -81,49 +81,52 @@ GLuint r300VAPInputCntl1(GLcontext * ctx, GLuint InputsRead)
return vic_1;
}
-GLuint r300VAPOutputCntl0(GLcontext * ctx, GLuint OutputsWritten)
+GLuint r300VAPOutputCntl0(GLcontext * ctx, GLuint vp_writes, GLuint fp_reads)
{
GLuint ret = 0;
- if (OutputsWritten & (1 << VERT_RESULT_HPOS))
+ if (vp_writes & (1 << VERT_RESULT_HPOS))
ret |= R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT;
- if (OutputsWritten & (1 << VERT_RESULT_COL0))
+ if (vp_writes & (1 << VERT_RESULT_COL0) && fp_reads & FRAG_BIT_COL0)
ret |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT;
- if (OutputsWritten & (1 << VERT_RESULT_COL1))
+ if (vp_writes & (1 << VERT_RESULT_COL1) && fp_reads & FRAG_BIT_COL1)
ret |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_1_PRESENT;
- if (OutputsWritten & (1 << VERT_RESULT_BFC0)
- || OutputsWritten & (1 << VERT_RESULT_BFC1))
- ret |=
- R300_VAP_OUTPUT_VTX_FMT_0__COLOR_1_PRESENT |
- R300_VAP_OUTPUT_VTX_FMT_0__COLOR_2_PRESENT |
- R300_VAP_OUTPUT_VTX_FMT_0__COLOR_3_PRESENT;
+ /* Two sided lighting works only if all 4 colors are written */
+ if (vp_writes & (1 << VERT_RESULT_BFC0) || vp_writes & (1 << VERT_RESULT_BFC1))
+ ret |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT | R300_VAP_OUTPUT_VTX_FMT_0__COLOR_1_PRESENT |
+ R300_VAP_OUTPUT_VTX_FMT_0__COLOR_2_PRESENT | R300_VAP_OUTPUT_VTX_FMT_0__COLOR_3_PRESENT;
- if (OutputsWritten & (1 << VERT_RESULT_PSIZ))
+ if (vp_writes & (1 << VERT_RESULT_PSIZ))
ret |= R300_VAP_OUTPUT_VTX_FMT_0__PT_SIZE_PRESENT;
return ret;
}
-GLuint r300VAPOutputCntl1(GLcontext * ctx, GLuint OutputsWritten)
+GLuint r300VAPOutputCntl1(GLcontext * ctx, GLuint vp_writes, GLuint fp_reads)
{
GLuint i, ret = 0, first_free_texcoord = 0;
for (i = 0; i < ctx->Const.MaxTextureUnits; i++) {
- if (OutputsWritten & (1 << (VERT_RESULT_TEX0 + i))) {
+ if (vp_writes & (1 << (VERT_RESULT_TEX0 + i)) && fp_reads & FRAG_BIT_TEX(i)) {
ret |= (4 << (3 * i));
++first_free_texcoord;
}
}
- if (OutputsWritten & (1 << VERT_RESULT_FOGC)) {
+ if (fp_reads & FRAG_BIT_WPOS) {
+ ret |= (4 << (3 * first_free_texcoord));
+ ++first_free_texcoord;
+ }
+
+ if (vp_writes & (1 << VERT_RESULT_FOGC) && fp_reads & FRAG_BIT_FOGC) {
if (first_free_texcoord > 8) {
fprintf(stderr, "\tout of free texcoords to write fog coord\n");
_mesa_exit(-1);
}
- ret |= 1 << (3 * first_free_texcoord);
+ ret |= 4 << (3 * first_free_texcoord);
}
return ret;
diff --git a/src/mesa/drivers/dri/r300/r300_emit.h b/src/mesa/drivers/dri/r300/r300_emit.h
index d88914c..2fb8b82 100644
--- a/src/mesa/drivers/dri/r300/r300_emit.h
+++ b/src/mesa/drivers/dri/r300/r300_emit.h
@@ -225,7 +225,7 @@ extern void r300EmitCacheFlush(r300ContextPtr rmesa);
extern GLuint r300VAPInputCntl0(GLcontext * ctx, GLuint InputsRead);
extern GLuint r300VAPInputCntl1(GLcontext * ctx, GLuint InputsRead);
-extern GLuint r300VAPOutputCntl0(GLcontext * ctx, GLuint OutputsWritten);
-extern GLuint r300VAPOutputCntl1(GLcontext * ctx, GLuint OutputsWritten);
+extern GLuint r300VAPOutputCntl0(GLcontext * ctx, GLuint vp_writes, GLuint fp_reads);
+extern GLuint r300VAPOutputCntl1(GLcontext * ctx, GLuint vp_writes, GLuint fp_reads);
#endif
diff --git a/src/mesa/drivers/dri/r300/r300_state.c b/src/mesa/drivers/dri/r300/r300_state.c
index 2fff89c..b509c13 100644
--- a/src/mesa/drivers/dri/r300/r300_state.c
+++ b/src/mesa/drivers/dri/r300/r300_state.c
@@ -1536,7 +1536,7 @@ static void r300SetupRSUnit(GLcontext * ctx)
r300->hw.ri.cmd[R300_RI_INTERP_0 + tex_ip] |= R300_RS_SEL_Q(R300_RS_SEL_K1) | R300_RS_TEX_PTR(rs_tex_count);
r300->hw.rr.cmd[R300_RR_INST_0 + tex_ip] |= R300_RS_INST_TEX_ID(tex_ip) | R300_RS_INST_TEX_CN_WRITE | R300_RS_INST_TEX_ADDR(fp_reg);
InputsRead &= ~FRAG_BIT_FOGC;
- rs_tex_count += 1;
+ rs_tex_count += 4;
++tex_ip;
++fp_reg;
} else {
@@ -1662,7 +1662,7 @@ static void r500SetupRSUnit(GLcontext * ctx)
r300->hw.rr.cmd[R300_RR_INST_0 + tex_ip] |= R500_RS_INST_TEX_ID(tex_ip) | R500_RS_INST_TEX_CN_WRITE | R500_RS_INST_TEX_ADDR(fp_reg);
InputsRead &= ~FRAG_BIT_FOGC;
- rs_tex_count += 1;
+ rs_tex_count += 4;
++tex_ip;
++fp_reg;
} else {
@@ -2270,8 +2270,8 @@ void r300SetupVAP(GLcontext *ctx, GLuint InputsRead, GLuint OutputsWritten)
rmesa->hw.vic.cmd[R300_VIC_CNTL_0] = r300VAPInputCntl0(ctx, InputsRead);
rmesa->hw.vic.cmd[R300_VIC_CNTL_1] = r300VAPInputCntl1(ctx, InputsRead);
- rmesa->hw.vof.cmd[R300_VOF_CNTL_0] = r300VAPOutputCntl0(ctx, OutputsWritten);
- rmesa->hw.vof.cmd[R300_VOF_CNTL_1] = r300VAPOutputCntl1(ctx, OutputsWritten);
+ rmesa->hw.vof.cmd[R300_VOF_CNTL_0] = r300VAPOutputCntl0(ctx, OutputsWritten, ctx->FragmentProgram._Current->Base.InputsRead);
+ rmesa->hw.vof.cmd[R300_VOF_CNTL_1] = r300VAPOutputCntl1(ctx, OutputsWritten, ctx->FragmentProgram._Current->Base.InputsRead);
}
void r300UpdateShaderStates(r300ContextPtr rmesa)
--
1.6.0.4
From 840f3f38941818753c44efaadf79f690eeea9c68 Mon Sep 17 00:00:00 2001
From: Maciej Cencora <[email protected]>
Date: Fri, 5 Jun 2009 18:27:00 +0200
Subject: [PATCH] r300: move some code for easier debugging
---
src/mesa/drivers/dri/r300/r300_vertprog.c | 54 ++++++++++++++++++++---------
1 files changed, 37 insertions(+), 17 deletions(-)
diff --git a/src/mesa/drivers/dri/r300/r300_vertprog.c b/src/mesa/drivers/dri/r300/r300_vertprog.c
index 46e1527..66750b1 100644
--- a/src/mesa/drivers/dri/r300/r300_vertprog.c
+++ b/src/mesa/drivers/dri/r300/r300_vertprog.c
@@ -1185,23 +1185,6 @@ static void r300TranslateVertexShader(struct r300_vertex_program *vp,
}
}
- /* Some outputs may be artificially added, to match the inputs
- of the fragment program. Blank the outputs here. */
- for (i = 0; i < VERT_RESULT_MAX; i++) {
- if (vp->key.OutputsAdded & (1 << i)) {
- inst[0] = PVS_OP_DST_OPERAND(VE_ADD,
- GL_FALSE,
- GL_FALSE,
- vp->outputs[i],
- VSF_FLAG_ALL,
- PVS_DST_REG_OUT);
- inst[1] = __CONST(0, SWIZZLE_ZERO);
- inst[2] = __CONST(0, SWIZZLE_ZERO);
- inst[3] = __CONST(0, SWIZZLE_ZERO);
- inst += 4;
- }
- }
-
vp->hw_code.length = (inst - vp->hw_code.body.d);
if (vp->hw_code.length >= VSF_MAX_FRAGMENT_LENGTH) {
vp->error = GL_TRUE;
@@ -1390,6 +1373,43 @@ static struct r300_vertex_program *build_program(struct r300_vertex_program_key
fflush(stdout);
}
+ /* Some outputs may be artificially added, to match the inputs of the fragment program.
+ * Issue 16 of vertex program spec says that all vertex attributes that are unwritten by
+ * vertex program are undefined, so just use MOV [vertex_result], CONST[0]
+ */
+ {
+ int i, count = 0;
+ for (i = 0; i < VERT_RESULT_MAX; ++i) {
+ if (vp->key.OutputsAdded & (1 << i)) {
+ ++count;
+ }
+ }
+
+ if (count > 0) {
+ struct prog_instruction *inst;
+
+ _mesa_insert_instructions(&mesa_vp->Base, mesa_vp->Base.NumInstructions - 1, count);
+ inst = &mesa_vp->Base.Instructions[mesa_vp->Base.NumInstructions - 1 - count];
+
+ for (i = 0; i < VERT_RESULT_MAX; ++i) {
+ if (vp->key.OutputsAdded & (1 << i)) {
+ inst->Opcode = OPCODE_MOV;
+
+ inst->DstReg.File = PROGRAM_OUTPUT;
+ inst->DstReg.Index = i;
+ inst->DstReg.WriteMask = WRITEMASK_XYZW;
+ inst->DstReg.CondMask = COND_TR;
+
+ inst->SrcReg[0].File = PROGRAM_CONSTANT;
+ inst->SrcReg[0].Index = 0;
+ inst->SrcReg[0].Swizzle = SWIZZLE_XYZW;
+
+ ++inst;
+ }
+ }
+ }
+ }
+
assert(mesa_vp->Base.NumInstructions);
vp->num_temporaries = mesa_vp->Base.NumTemporaries;
r300TranslateVertexShader(vp, mesa_vp->Base.Instructions);
--
1.6.0.4
------------------------------------------------------------------------------
Crystal Reports - New Free Runtime and 30 Day Trial
Check out the new simplified licensing option that enables unlimited
royalty-free distribution of the report engine for externally facing
server and web deployment.
http://p.sf.net/sfu/businessobjects
_______________________________________________
Mesa3d-dev mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/mesa3d-dev
