r300: fix a GPU lock up

Sending from VAP more texture coordinates than RS expects results in GPU hang.

Fixes BumpSelfShadow from DirectX8 SDK.

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 cf31596d9b..4017224b10 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 d88914ce95..2fb8b82d3a 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 2fff89c473..b509c13b43 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)