[r300] Fix KIL instruction and swizzling of texture coordinates

The KIL instruction only works if at least one texture unit is enabled
in hardware.

Texture instructions do not support swizzles, negations etc. natively,
so we now emit an explicit swizzling etc. operation when the texture coordinate
requires it.

This fixes the Piglit fp-kil test.

2 files changed, 72 insertions, 57 deletions

diff --git a/src/mesa/drivers/dri/r300/r300_fragprog.c b/src/mesa/drivers/dri/r300/r300_fragprog.c
index a28841dda8..ac45da4404 100644
--- a/src/mesa/drivers/dri/r300/r300_fragprog.c
+++ b/src/mesa/drivers/dri/r300/r300_fragprog.c
@@ -903,49 +903,59 @@ static void emit_tex(struct r300_fragment_program *fp,
 	int hwsrc, hwdest;
 	GLuint tempreg = 0;
 
+	/**
+	 * Hardware uses [0..1]x[0..1] range for rectangle textures
+	 * instead of [0..Width]x[0..Height].
+	 * Add a scaling instruction.
+	 *
+	 * \todo Refactor this once we have proper rewriting/optimization
+	 * support for programs.
+	 */
+	if (opcode != R300_TEX_OP_KIL && fpi->TexSrcTarget == TEXTURE_RECT_INDEX) {
+		gl_state_index tokens[STATE_LENGTH] = {
+			STATE_INTERNAL, STATE_R300_TEXRECT_FACTOR, 0, 0,
+			0
+		};
+		int factor_index;
+		GLuint factorreg;
+
+		tokens[2] = unit;
+		factor_index =
+			_mesa_add_state_reference(fp->mesa_program.Base.
+						Parameters, tokens);
+		factorreg =
+			emit_const4fv(fp,
+				fp->mesa_program.Base.Parameters->
+				ParameterValues[factor_index]);
+		tempreg = keep(get_temp_reg(fp));
+
+		emit_arith(fp, PFS_OP_MAD, tempreg, WRITEMASK_XYZW,
+			coord, factorreg, pfs_zero, 0);
+
+		coord = tempreg;
+	}
+
+	/* Texture operations do not support swizzles etc. in hardware,
+	 * so emit an additional arithmetic operation if necessary.
+	 */
+	if (REG_GET_VSWZ(coord) != SWIZZLE_XYZ ||
+	    REG_GET_SSWZ(coord) != SWIZZLE_W ||
+	    coord & (REG_NEGV_MASK | REG_NEGS_MASK | REG_ABS_MASK)) {
+		assert(tempreg == 0);
+		tempreg = keep(get_temp_reg(fp));
+		emit_arith(fp, PFS_OP_MAD, tempreg, WRITEMASK_XYZW,
+			coord, pfs_one, pfs_zero, 0);
+		coord = tempreg;
+	}
+
+	/* Ensure correct node indirection */
 	uin = cs->used_in_node;
 	din = cs->dest_in_node;
 
 	/* Resolve source/dest to hardware registers */
-	if (opcode != R300_TEX_OP_KIL) {
-		if (fpi->TexSrcTarget == TEXTURE_RECT_INDEX) {
-			/**
-			 * Hardware uses [0..1]x[0..1] range for rectangle textures
-			 * instead of [0..Width]x[0..Height].
-			 * Add a scaling instruction.
-			 *
-			 * \todo Refactor this once we have proper rewriting/optimization
-			 * support for programs.
-			 */
-			gl_state_index tokens[STATE_LENGTH] = {
-				STATE_INTERNAL, STATE_R300_TEXRECT_FACTOR, 0, 0,
-				0
-			};
-			int factor_index;
-			GLuint factorreg;
-
-			tokens[2] = unit;
-			factor_index =
-			    _mesa_add_state_reference(fp->mesa_program.Base.
-						      Parameters, tokens);
-			factorreg =
-			    emit_const4fv(fp,
-					  fp->mesa_program.Base.Parameters->
-					  ParameterValues[factor_index]);
-			tempreg = keep(get_temp_reg(fp));
-
-			emit_arith(fp, PFS_OP_MAD, tempreg, WRITEMASK_XYZW,
-				   coord, factorreg, pfs_zero, 0);
-
-			/* Ensure correct node indirection */
-			uin = cs->used_in_node;
-			din = cs->dest_in_node;
-
-			hwsrc = t_hw_src(fp, tempreg, GL_TRUE);
-		} else {
-			hwsrc = t_hw_src(fp, coord, GL_TRUE);
-		}
+	hwsrc = t_hw_src(fp, coord, GL_TRUE);
 
+	if (opcode != R300_TEX_OP_KIL) {
 		dest = t_dst(fp, fpi->DstReg);
 
 		/* r300 doesn't seem to be able to do TEX->output reg */
@@ -972,7 +982,6 @@ static void emit_tex(struct r300_fragment_program *fp,
 	} else {
 		hwdest = 0;
 		unit = 0;
-		hwsrc = t_hw_src(fp, coord, GL_TRUE);
 	}
 
 	/* Indirection if source has been written in this node, or if the
diff --git a/src/mesa/drivers/dri/r300/r300_state.c b/src/mesa/drivers/dri/r300/r300_state.c
index dc14759c08..df4573d6b2 100644
--- a/src/mesa/drivers/dri/r300/r300_state.c
+++ b/src/mesa/drivers/dri/r300/r300_state.c
@@ -1337,13 +1337,13 @@ static void r300SetupFragmentShaderTextures(GLcontext *ctx, int *tmu_mappings)
 		int unit;
 		int opcode;
 		unsigned long val;
-			
+
 		unit = fp->tex.inst[i] >> R300_TEX_ID_SHIFT;
 		unit &= 15;
-			
+
 		val = fp->tex.inst[i];
 		val &= ~R300_TEX_ID_MASK;
-			
+
 		opcode =
 			(val & R300_TEX_INST_MASK) >> R300_TEX_INST_SHIFT;
 		if (opcode == R300_TEX_OP_KIL) {
@@ -1361,10 +1361,9 @@ static void r300SetupFragmentShaderTextures(GLcontext *ctx, int *tmu_mappings)
 			}
 		}
 	}
-	
+
 	r300->hw.fpt.cmd[R300_FPT_CMD_0] =
 		cmdpacket0(R300_US_TEX_INST_0, fp->tex.length);
-	
 }
 
 static void r500SetupFragmentShaderTextures(GLcontext *ctx, int *tmu_mappings)
@@ -1384,7 +1383,7 @@ static void r500SetupFragmentShaderTextures(GLcontext *ctx, int *tmu_mappings)
 			unit = (val >> 16) & 0xf;
 
 			val &= ~(0xf << 16);
-			
+
 			opcode = val & (0x7 << 22);
 			if (opcode == R500_TEX_INST_TEXKILL) {
 				new_unit = 0;
@@ -1515,10 +1514,17 @@ static void r300SetupTextures(GLcontext * ctx)
 	if (!fp)		/* should only happenen once, just after context is created */
 		return;
 
-
-        if (r300->radeon.radeonScreen->chip_family < CHIP_FAMILY_RV515)
+	if (r300->radeon.radeonScreen->chip_family < CHIP_FAMILY_RV515) {
+		if (fp->mesa_program.UsesKill && last_hw_tmu < 0) {
+			// The KILL operation requires the first texture unit
+			// to be enabled.
+			r300->hw.txe.cmd[R300_TXE_ENABLE] |= 1;
+			r300->hw.tex.filter.cmd[R300_TEX_VALUE_0] = 0;
+			r300->hw.tex.filter.cmd[R300_TEX_CMD_0] =
+				cmdpacket0(R300_TX_FILTER0_0, 1);
+		}
 		r300SetupFragmentShaderTextures(ctx, tmu_mappings);
-	else 
+	} else
 		r500SetupFragmentShaderTextures(ctx, tmu_mappings);
 
 	if (RADEON_DEBUG & DEBUG_STATE)
@@ -1623,7 +1629,7 @@ static void r300SetupRSUnit(GLcontext * ctx)
 		};
 
 		r300->hw.ri.cmd[R300_RI_INTERP_0 + i] |= swiz;
- 
+
 		r300->hw.rr.cmd[R300_RR_INST_0 + fp_reg] = 0;
 		if (InputsRead & (FRAG_BIT_TEX0 << i)) {
 
@@ -1756,15 +1762,15 @@ static void r500SetupRSUnit(GLcontext * ctx)
 
 		/* with TCL we always seem to route 4 components */
 		if (InputsRead & (FRAG_BIT_TEX0 << i)) {
-		  
+
 		  if (hw_tcl_on)
 		    count = 4;
 		  else
 		    count = VB->AttribPtr[_TNL_ATTRIB_TEX(i)]->size;
-		  
+
 		  /* always have on texcoord */
 		  swiz |= in_texcoords++ << R500_RS_IP_TEX_PTR_S_SHIFT;
-		  if (count >= 2) 
+		  if (count >= 2)
 		    swiz |= in_texcoords++ << R500_RS_IP_TEX_PTR_T_SHIFT;
 		  else
 		    swiz |= R500_RS_IP_PTR_K0 << R500_RS_IP_TEX_PTR_T_SHIFT;
@@ -1774,11 +1780,11 @@ static void r500SetupRSUnit(GLcontext * ctx)
 		  else
 		    swiz |= R500_RS_IP_PTR_K0 << R500_RS_IP_TEX_PTR_R_SHIFT;
 
-		  if (count == 4) 
+		  if (count == 4)
 		    swiz |= in_texcoords++ << R500_RS_IP_TEX_PTR_Q_SHIFT;
 		  else
 		    swiz |= R500_RS_IP_PTR_K1 << R500_RS_IP_TEX_PTR_Q_SHIFT;
-		  
+
 		} else
 		   swiz = (R500_RS_IP_PTR_K0 << R500_RS_IP_TEX_PTR_S_SHIFT) |
 		          (R500_RS_IP_PTR_K0 << R500_RS_IP_TEX_PTR_T_SHIFT) |
@@ -1920,7 +1926,7 @@ static void r300VapCntl(r300ContextPtr rmesa, GLuint input_count,
 
     R300_STATECHANGE(rmesa, vap_cntl);
     if (rmesa->radeon.radeonScreen->chip_flags & RADEON_CHIPSET_TCL) {
-	rmesa->hw.vap_cntl.cmd[R300_VAP_CNTL_INSTR] = 
+	rmesa->hw.vap_cntl.cmd[R300_VAP_CNTL_INSTR] =
 	    (pvs_num_slots << R300_PVS_NUM_SLOTS_SHIFT) |
 	    (pvs_num_cntrls << R300_PVS_NUM_CNTLRS_SHIFT) |
 	    (12 << R300_VF_MAX_VTX_NUM_SHIFT);
@@ -2640,11 +2646,11 @@ void r300UpdateClipPlanes( GLcontext *ctx )
 {
 	r300ContextPtr rmesa = R300_CONTEXT(ctx);
 	GLuint p;
-	
+
 	for (p = 0; p < ctx->Const.MaxClipPlanes; p++) {
 		if (ctx->Transform.ClipPlanesEnabled & (1 << p)) {
 			GLint *ip = (GLint *)ctx->Transform._ClipUserPlane[p];
-			
+
 			R300_STATECHANGE( rmesa, vpucp[p] );
 			rmesa->hw.vpucp[p].cmd[R300_VPUCP_X] = ip[0];
 			rmesa->hw.vpucp[p].cmd[R300_VPUCP_Y] = ip[1];