r300_fragprog: Emulate trigonometric functions in radeon_program_alu

4 files changed, 153 insertions, 207 deletions

diff --git a/src/mesa/drivers/dri/r300/r300_fragprog.c b/src/mesa/drivers/dri/r300/r300_fragprog.c
index 57987f5d0fc..8a1d690ae4e 100644
--- a/src/mesa/drivers/dri/r300/r300_fragprog.c
+++ b/src/mesa/drivers/dri/r300/r300_fragprog.c
@@ -408,12 +408,13 @@ void r300TranslateFragmentShader(r300ContextPtr r300,
 
 		struct radeon_program_transformation transformations[] = {
 			{ &transform_TEX, &compiler },
-			{ &radeonTransformALU, 0 }
+			{ &radeonTransformALU, 0 },
+			{ &radeonTransformTrigSimple, 0 }
 		};
 		radeonLocalTransform(
 			r300->radeon.glCtx,
 			compiler.program,
-			2, transformations);
+			3, transformations);
 
 		if (RADEON_DEBUG & DEBUG_PIXEL) {
 			_mesa_printf("Fragment Program: After transformations:\n");
diff --git a/src/mesa/drivers/dri/r300/r300_fragprog_emit.c b/src/mesa/drivers/dri/r300/r300_fragprog_emit.c
index 30f513b5a31..4786b4554df 100644
--- a/src/mesa/drivers/dri/r300/r300_fragprog_emit.c
+++ b/src/mesa/drivers/dri/r300/r300_fragprog_emit.c
@@ -1423,40 +1423,11 @@ static void emit_arith(struct r300_pfs_compile_state *cs,
 	return;
 }
 
-static GLfloat SinCosConsts[2][4] = {
-	{
-	 1.273239545,		// 4/PI
-	 -0.405284735,		// -4/(PI*PI)
-	 3.141592654,		// PI
-	 0.2225			// weight
-	 },
-	{
-	 0.75,
-	 0.0,
-	 0.159154943,		// 1/(2*PI)
-	 6.283185307		// 2*PI
-	 }
-};
-
-static GLuint emit_sincosconsts(struct r300_pfs_compile_state *cs, int i)
-{
-	struct prog_src_register srcreg;
-	GLuint constant_swizzle;
-
-	srcreg.File = PROGRAM_CONSTANT;
-	srcreg.Index = _mesa_add_unnamed_constant(cs->compiler->program->Parameters,
-		SinCosConsts[i], 4, &constant_swizzle);
-	srcreg.Swizzle = constant_swizzle;
-
-	return emit_const4fv(cs, srcreg);
-}
-
 static void emit_instruction(struct r300_pfs_compile_state *cs, struct prog_instruction *fpi)
 {
 	COMPILE_STATE;
-	GLuint src[3], dest, temp[2];
+	GLuint src[3], dest;
 	int flags, mask = 0;
-	int const_sin[2];
 
 	if (fpi->SaturateMode == SATURATE_ZERO_ONE)
 		flags = PFS_FLAG_SAT;
@@ -1485,60 +1456,6 @@ static void emit_instruction(struct r300_pfs_compile_state *cs, struct prog_inst
 		emit_arith(cs, PFS_OP_CMP, dest, mask,
 				src[2], src[1], src[0], flags);
 		break;
-	case OPCODE_COS:
-		/*
-			* cos using a parabola (see SIN):
-			* cos(x):
-			*   x = (x/(2*PI))+0.75
-			*   x = frac(x)
-			*   x = (x*2*PI)-PI
-			*   result = sin(x)
-			*/
-		temp[0] = get_temp_reg(cs);
-		const_sin[0] = emit_sincosconsts(cs, 0);
-		const_sin[1] = emit_sincosconsts(cs, 1);
-		src[0] = t_scalar_src(cs, fpi->SrcReg[0]);
-
-		/* add 0.5*PI and do range reduction */
-
-		emit_arith(cs, PFS_OP_MAD, temp[0], WRITEMASK_X,
-				swizzle(src[0], X, X, X, X),
-				swizzle(const_sin[1], Z, Z, Z, Z),
-				swizzle(const_sin[1], X, X, X, X), 0);
-
-		emit_arith(cs, PFS_OP_FRC, temp[0], WRITEMASK_X,
-				swizzle(temp[0], X, X, X, X),
-				undef, undef, 0);
-
-		emit_arith(cs, PFS_OP_MAD, temp[0], WRITEMASK_Z, swizzle(temp[0], X, X, X, X), swizzle(const_sin[1], W, W, W, W),	//2*PI
-				negate(swizzle(const_sin[0], Z, Z, Z, Z)),	//-PI
-				0);
-
-		/* SIN */
-
-		emit_arith(cs, PFS_OP_MAD, temp[0],
-				WRITEMASK_X | WRITEMASK_Y, swizzle(temp[0],
-								Z, Z, Z,
-								Z),
-				const_sin[0], pfs_zero, 0);
-
-		emit_arith(cs, PFS_OP_MAD, temp[0], WRITEMASK_X,
-				swizzle(temp[0], Y, Y, Y, Y),
-				absolute(swizzle(temp[0], Z, Z, Z, Z)),
-				swizzle(temp[0], X, X, X, X), 0);
-
-		emit_arith(cs, PFS_OP_MAD, temp[0], WRITEMASK_Y,
-				swizzle(temp[0], X, X, X, X),
-				absolute(swizzle(temp[0], X, X, X, X)),
-				negate(swizzle(temp[0], X, X, X, X)), 0);
-
-		emit_arith(cs, PFS_OP_MAD, dest, mask,
-				swizzle(temp[0], Y, Y, Y, Y),
-				swizzle(const_sin[0], W, W, W, W),
-				swizzle(temp[0], X, X, X, X), flags);
-
-		free_temp(cs, temp[0]);
-		break;
 	case OPCODE_DP3:
 		src[0] = t_src(cs, fpi->SrcReg[0]);
 		src[1] = t_src(cs, fpi->SrcReg[1]);
@@ -1609,127 +1526,6 @@ static void emit_instruction(struct r300_pfs_compile_state *cs, struct prog_inst
 		emit_arith(cs, PFS_OP_RSQ, dest, mask,
 				absolute(src[0]), pfs_zero, pfs_zero, flags);
 		break;
-	case OPCODE_SCS:
-		/*
-			* scs using a parabola :
-			* scs(x):
-			*   result.x = sin(-abs(x)+0.5*PI)  (cos)
-			*   result.y = sin(x)               (sin)
-			*
-			*/
-		temp[0] = get_temp_reg(cs);
-		temp[1] = get_temp_reg(cs);
-		const_sin[0] = emit_sincosconsts(cs, 0);
-		const_sin[1] = emit_sincosconsts(cs, 1);
-		src[0] = t_scalar_src(cs, fpi->SrcReg[0]);
-
-		/* x = -abs(x)+0.5*PI */
-		emit_arith(cs, PFS_OP_MAD, temp[0], WRITEMASK_Z, swizzle(const_sin[0], Z, Z, Z, Z),	//PI
-				pfs_half,
-				negate(abs
-					(swizzle(keep(src[0]), X, X, X, X))),
-				0);
-
-		/* C*x (sin) */
-		emit_arith(cs, PFS_OP_MAD, temp[0], WRITEMASK_W,
-				swizzle(const_sin[0], Y, Y, Y, Y),
-				swizzle(keep(src[0]), X, X, X, X),
-				pfs_zero, 0);
-
-		/* B*x, C*x (cos) */
-		emit_arith(cs, PFS_OP_MAD, temp[0],
-				WRITEMASK_X | WRITEMASK_Y, swizzle(temp[0],
-								Z, Z, Z,
-								Z),
-				const_sin[0], pfs_zero, 0);
-
-		/* B*x (sin) */
-		emit_arith(cs, PFS_OP_MAD, temp[1], WRITEMASK_W,
-				swizzle(const_sin[0], X, X, X, X),
-				keep(src[0]), pfs_zero, 0);
-
-		/* y = B*x + C*x*abs(x) (sin) */
-		emit_arith(cs, PFS_OP_MAD, temp[1], WRITEMASK_Z,
-				absolute(src[0]),
-				swizzle(temp[0], W, W, W, W),
-				swizzle(temp[1], W, W, W, W), 0);
-
-		/* y = B*x + C*x*abs(x) (cos) */
-		emit_arith(cs, PFS_OP_MAD, temp[1], WRITEMASK_W,
-				swizzle(temp[0], Y, Y, Y, Y),
-				absolute(swizzle(temp[0], Z, Z, Z, Z)),
-				swizzle(temp[0], X, X, X, X), 0);
-
-		/* y*abs(y) - y (cos), y*abs(y) - y (sin) */
-		emit_arith(cs, PFS_OP_MAD, temp[0],
-				WRITEMASK_X | WRITEMASK_Y, swizzle(temp[1],
-								W, Z, Y,
-								X),
-				absolute(swizzle(temp[1], W, Z, Y, X)),
-				negate(swizzle(temp[1], W, Z, Y, X)), 0);
-
-		/* dest.xy = mad(temp.xy, P, temp2.wz) */
-		emit_arith(cs, PFS_OP_MAD, dest,
-				mask & (WRITEMASK_X | WRITEMASK_Y), temp[0],
-				swizzle(const_sin[0], W, W, W, W),
-				swizzle(temp[1], W, Z, Y, X), flags);
-
-		free_temp(cs, temp[0]);
-		free_temp(cs, temp[1]);
-		break;
-	case OPCODE_SIN:
-		/*
-			*  using a parabola:
-			* sin(x) = 4/pi * x + -4/(pi*pi) * x * abs(x)
-			* extra precision is obtained by weighting against
-			* itself squared.
-			*/
-
-		temp[0] = get_temp_reg(cs);
-		const_sin[0] = emit_sincosconsts(cs, 0);
-		const_sin[1] = emit_sincosconsts(cs, 1);
-		src[0] = t_scalar_src(cs, fpi->SrcReg[0]);
-
-		/* do range reduction */
-
-		emit_arith(cs, PFS_OP_MAD, temp[0], WRITEMASK_X,
-				swizzle(keep(src[0]), X, X, X, X),
-				swizzle(const_sin[1], Z, Z, Z, Z),
-				pfs_half, 0);
-
-		emit_arith(cs, PFS_OP_FRC, temp[0], WRITEMASK_X,
-				swizzle(temp[0], X, X, X, X),
-				undef, undef, 0);
-
-		emit_arith(cs, PFS_OP_MAD, temp[0], WRITEMASK_Z, swizzle(temp[0], X, X, X, X), swizzle(const_sin[1], W, W, W, W),	//2*PI
-				negate(swizzle(const_sin[0], Z, Z, Z, Z)),	//PI
-				0);
-
-		/* SIN */
-
-		emit_arith(cs, PFS_OP_MAD, temp[0],
-				WRITEMASK_X | WRITEMASK_Y, swizzle(temp[0],
-								Z, Z, Z,
-								Z),
-				const_sin[0], pfs_zero, 0);
-
-		emit_arith(cs, PFS_OP_MAD, temp[0], WRITEMASK_X,
-				swizzle(temp[0], Y, Y, Y, Y),
-				absolute(swizzle(temp[0], Z, Z, Z, Z)),
-				swizzle(temp[0], X, X, X, X), 0);
-
-		emit_arith(cs, PFS_OP_MAD, temp[0], WRITEMASK_Y,
-				swizzle(temp[0], X, X, X, X),
-				absolute(swizzle(temp[0], X, X, X, X)),
-				negate(swizzle(temp[0], X, X, X, X)), 0);
-
-		emit_arith(cs, PFS_OP_MAD, dest, mask,
-				swizzle(temp[0], Y, Y, Y, Y),
-				swizzle(const_sin[0], W, W, W, W),
-				swizzle(temp[0], X, X, X, X), flags);
-
-		free_temp(cs, temp[0]);
-		break;
 	case OPCODE_TEX:
 		emit_tex(cs, fpi, R300_TEX_OP_LD);
 		break;
diff --git a/src/mesa/drivers/dri/r300/radeon_program_alu.c b/src/mesa/drivers/dri/r300/radeon_program_alu.c
index 4a40d3e44d8..fa6a67f0c19 100644
--- a/src/mesa/drivers/dri/r300/radeon_program_alu.c
+++ b/src/mesa/drivers/dri/r300/radeon_program_alu.c
@@ -149,6 +149,14 @@ static struct prog_src_register srcregswz(int file, int index, int swz)
 	return src;
 }
 
+static struct prog_src_register absolute(struct prog_src_register reg)
+{
+	struct prog_src_register newreg = reg;
+	newreg.Abs = 1;
+	newreg.NegateAbs = 0;
+	return newreg;
+}
+
 static struct prog_src_register negate(struct prog_src_register reg)
 {
 	struct prog_src_register newreg = reg;
@@ -412,3 +420,139 @@ GLboolean radeonTransformALU(struct radeon_transform_context* t,
 		return GL_FALSE;
 	}
 }
+
+
+static void sincos_constants(struct radeon_transform_context* t, GLuint *constants)
+{
+	static const GLfloat SinCosConsts[2][4] = {
+		{
+			1.273239545,		// 4/PI
+			-0.405284735,		// -4/(PI*PI)
+			3.141592654,		// PI
+			0.2225			// weight
+		},
+		{
+			0.75,
+			0.5,
+			0.159154943,		// 1/(2*PI)
+			6.283185307		// 2*PI
+		}
+	};
+	int i;
+
+	for(i = 0; i < 2; ++i) {
+		GLuint swz;
+		constants[i] = _mesa_add_unnamed_constant(t->Program->Parameters, SinCosConsts[i], 4, &swz);
+		ASSERT(swz == SWIZZLE_NOOP);
+	}
+}
+
+/**
+ * Approximate sin(x), where x is clamped to (-pi/2, pi/2).
+ *
+ * MUL tmp.xy, src, { 4/PI, -4/(PI^2) }
+ * MAD tmp.x, tmp.y, |src|, tmp.x
+ * MAD tmp.y, tmp.x, |tmp.x|, -tmp.x
+ * MAD dest, tmp.y, weight, tmp.x
+ */
+static void sin_approx(struct radeon_transform_context* t,
+	struct prog_dst_register dst, struct prog_src_register src, const GLuint* constants)
+{
+	GLuint tempreg = radeonFindFreeTemporary(t);
+
+	emit2(t->Program, OPCODE_MUL, dstregtmpmask(tempreg, WRITEMASK_XY),
+		swizzle(src, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X),
+		srcreg(PROGRAM_CONSTANT, constants[0]));
+	emit3(t->Program, OPCODE_MAD, dstregtmpmask(tempreg, WRITEMASK_X),
+		swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y),
+		absolute(swizzle(src, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X)),
+		swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X));
+	emit3(t->Program, OPCODE_MAD, dstregtmpmask(tempreg, WRITEMASK_Y),
+		swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X),
+		absolute(swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X)),
+		negate(swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X)));
+	emit3(t->Program, OPCODE_MAD, dst,
+		swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y),
+		swizzle(srcreg(PROGRAM_CONSTANT, constants[0]), SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W),
+		swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X));
+}
+
+/**
+ * Translate the trigonometric functions COS, SIN, and SCS
+ * using only the basic instructions
+ *  MOV, ADD, MUL, MAD, FRC
+ */
+GLboolean radeonTransformTrigSimple(struct radeon_transform_context* t,
+	struct prog_instruction* inst,
+	void* unused)
+{
+	if (inst->Opcode != OPCODE_COS &&
+	    inst->Opcode != OPCODE_SIN &&
+	    inst->Opcode != OPCODE_SCS)
+		return GL_FALSE;
+
+	GLuint constants[2];
+	GLuint tempreg = radeonFindFreeTemporary(t);
+
+	sincos_constants(t, constants);
+
+	if (inst->Opcode == OPCODE_COS) {
+		// MAD tmp.x, src, 1/(2*PI), 0.75
+		// FRC tmp.x, tmp.x
+		// MAD tmp.z, tmp.x, 2*PI, -PI
+		emit3(t->Program, OPCODE_MAD, dstregtmpmask(tempreg, WRITEMASK_W),
+			swizzle(inst->SrcReg[0], SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X),
+			swizzle(srcreg(PROGRAM_CONSTANT, constants[1]), SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z),
+			swizzle(srcreg(PROGRAM_CONSTANT, constants[1]), SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X));
+		emit1(t->Program, OPCODE_FRC, dstregtmpmask(tempreg, WRITEMASK_W),
+			swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W));
+		emit3(t->Program, OPCODE_MAD, dstregtmpmask(tempreg, WRITEMASK_W),
+			swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W),
+			swizzle(srcreg(PROGRAM_CONSTANT, constants[1]), SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W),
+			negate(swizzle(srcreg(PROGRAM_CONSTANT, constants[0]), SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z)));
+
+		sin_approx(t, inst->DstReg,
+			swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W),
+			constants);
+	} else if (inst->Opcode == OPCODE_SIN) {
+		emit3(t->Program, OPCODE_MAD, dstregtmpmask(tempreg, WRITEMASK_W),
+			swizzle(inst->SrcReg[0], SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X),
+			swizzle(srcreg(PROGRAM_CONSTANT, constants[1]), SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z),
+			swizzle(srcreg(PROGRAM_CONSTANT, constants[1]), SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y));
+		emit1(t->Program, OPCODE_FRC, dstregtmpmask(tempreg, WRITEMASK_W),
+			swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W));
+		emit3(t->Program, OPCODE_MAD, dstregtmpmask(tempreg, WRITEMASK_W),
+			swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W),
+			swizzle(srcreg(PROGRAM_CONSTANT, constants[1]), SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W),
+			negate(swizzle(srcreg(PROGRAM_CONSTANT, constants[0]), SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z)));
+
+		sin_approx(t, inst->DstReg,
+			swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W),
+			constants);
+	} else {
+		emit3(t->Program, OPCODE_MAD, dstregtmpmask(tempreg, WRITEMASK_XY),
+			swizzle(inst->SrcReg[0], SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X),
+			swizzle(srcreg(PROGRAM_CONSTANT, constants[1]), SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z),
+			swizzle(srcreg(PROGRAM_CONSTANT, constants[1]), SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_W));
+		emit1(t->Program, OPCODE_FRC, dstregtmpmask(tempreg, WRITEMASK_XY),
+			srcreg(PROGRAM_TEMPORARY, tempreg));
+		emit3(t->Program, OPCODE_MAD, dstregtmpmask(tempreg, WRITEMASK_XY),
+			srcreg(PROGRAM_TEMPORARY, tempreg),
+			swizzle(srcreg(PROGRAM_CONSTANT, constants[1]), SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W),
+			negate(swizzle(srcreg(PROGRAM_CONSTANT, constants[0]), SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z)));
+
+		struct prog_dst_register dst = inst->DstReg;
+
+		dst.WriteMask = inst->DstReg.WriteMask & WRITEMASK_X;
+		sin_approx(t, dst,
+			swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X),
+			constants);
+
+		dst.WriteMask = inst->DstReg.WriteMask & WRITEMASK_Y;
+		sin_approx(t, dst,
+			swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y),
+			constants);
+	}
+
+	return GL_TRUE;
+}
diff --git a/src/mesa/drivers/dri/r300/radeon_program_alu.h b/src/mesa/drivers/dri/r300/radeon_program_alu.h
index 858c5ed0b8c..3fe6153fd82 100644
--- a/src/mesa/drivers/dri/r300/radeon_program_alu.h
+++ b/src/mesa/drivers/dri/r300/radeon_program_alu.h
@@ -35,4 +35,9 @@ GLboolean radeonTransformALU(
 	struct prog_instruction*,
 	void*);
 
+GLboolean radeonTransformTrigSimple(
+	struct radeon_transform_context *t,
+	struct prog_instruction*,
+	void*);
+
 #endif /* __RADEON_PROGRAM_ALU_H_ */