#include "cell_gen_fp.h"
-/** Set to 1 to enable debug/disassembly printfs */
-#define DISASSEM 0
-
-
#define MAX_TEMPS 16
#define MAX_IMMED 8
+#define CHAN_X 0
+#define CHAN_Y 1
+#define CHAN_Z 2
+#define CHAN_W 3
/**
* Context needed during code generation.
/** Per-instruction temps / intermediate temps */
int num_itemps;
- int itemps[3];
+ int itemps[10];
/** Current IF/ELSE/ENDIF nesting level */
int if_nesting;
{
int reg = -1;
int swizzle = tgsi_util_get_full_src_register_extswizzle(src, channel);
+ boolean reg_is_itemp = FALSE;
+ uint sign_op;
- assert(swizzle >= 0);
- assert(swizzle <= 3);
-
- channel = swizzle;
+ assert(swizzle >= TGSI_SWIZZLE_X);
+ assert(swizzle <= TGSI_EXTSWIZZLE_ONE);
switch (src->SrcRegister.File) {
case TGSI_FILE_TEMPORARY:
- reg = gen->temp_regs[src->SrcRegister.Index][channel];
+ reg = gen->temp_regs[src->SrcRegister.Index][swizzle];
break;
case TGSI_FILE_INPUT:
{
- /* offset is measured in quadwords, not bytes */
- int offset = src->SrcRegister.Index * 4 + channel;
- reg = get_itemp(gen);
- /* Load: reg = memory[(machine_reg) + offset] */
- spe_lqd(gen->f, reg, gen->inputs_reg, offset);
+ if(swizzle == TGSI_EXTSWIZZLE_ONE)
+ {
+ /* Load const one float and early out */
+ reg = get_const_one_reg(gen);
+ }
+ else if(swizzle == TGSI_EXTSWIZZLE_ZERO)
+ {
+ /* Load const zero float and early out */
+ reg = get_itemp(gen);
+ spe_xor(gen->f, reg, reg, reg);
+ }
+ else
+ {
+ /* offset is measured in quadwords, not bytes */
+ int offset = src->SrcRegister.Index * 4 + swizzle;
+ reg = get_itemp(gen);
+ reg_is_itemp = TRUE;
+ /* Load: reg = memory[(machine_reg) + offset] */
+ spe_lqd(gen->f, reg, gen->inputs_reg, offset);
+ }
}
break;
case TGSI_FILE_IMMEDIATE:
- reg = gen->imm_regs[src->SrcRegister.Index][channel];
+ reg = gen->imm_regs[src->SrcRegister.Index][swizzle];
break;
case TGSI_FILE_CONSTANT:
/* xxx fall-through for now / fix */
assert(0);
}
+ /*
+ * Handle absolute value, negate or set-negative of src register.
+ */
+ sign_op = tgsi_util_get_full_src_register_sign_mode(src, channel);
+ if (sign_op != TGSI_UTIL_SIGN_KEEP) {
+ /*
+ * All sign ops are done by manipulating bit 31, the IEEE float sign bit.
+ */
+ const int bit31mask_reg = get_itemp(gen);
+ int result_reg;
+
+ if (reg_is_itemp) {
+ /* re-use 'reg' for the result */
+ result_reg = reg;
+ }
+ else {
+ /* alloc a new reg for the result */
+ result_reg = get_itemp(gen);
+ }
+
+ /* mask with bit 31 set, the rest cleared */
+ spe_load_int(gen->f, bit31mask_reg, (1 << 31));
+
+ if (sign_op == TGSI_UTIL_SIGN_CLEAR) {
+ spe_andc(gen->f, result_reg, reg, bit31mask_reg);
+ }
+ else if (sign_op == TGSI_UTIL_SIGN_SET) {
+ spe_and(gen->f, result_reg, reg, bit31mask_reg);
+ }
+ else {
+ assert(sign_op == TGSI_UTIL_SIGN_TOGGLE);
+ spe_xor(gen->f, result_reg, reg, bit31mask_reg);
+ }
+
+ reg = result_reg;
+ }
+
return reg;
}
return true;
}
-
-
/**
* Emit addition instructions. Recall that a single TGSI_OPCODE_ADD
* becomes (up to) four SPU "fa" instructions because we're doing SOA
return true;
}
+/**
+ * Emit subtract. See emit_ADD for comments.
+ */
+static boolean
+emit_SUB(struct codegen *gen, const struct tgsi_full_instruction *inst)
+{
+ int ch;
+ spe_comment(gen->f, -4, "SUB:");
+ /* Loop over Red/Green/Blue/Alpha channels */
+ for (ch = 0; ch < 4; ch++) {
+ /* If the dest R, G, B or A writemask is enabled... */
+ if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) {
+ /* get indexes of the two src, one dest SPE registers */
+ int s1_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]);
+ int s2_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[1]);
+ int d_reg = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]);
+
+ /* Emit actual SPE instruction: d = s1 - s2 */
+ spe_fs(gen->f, d_reg, s1_reg, s2_reg);
+
+ /* Store the result (a no-op for TGSI_FILE_TEMPORARY dests) */
+ store_dest_reg(gen, d_reg, ch, &inst->FullDstRegisters[0]);
+ /* Free any intermediate temps we allocated */
+ free_itemps(gen);
+ }
+ }
+ return true;
+}
+
+/**
+ * Emit multiply add. See emit_ADD for comments.
+ */
+static boolean
+emit_MAD(struct codegen *gen, const struct tgsi_full_instruction *inst)
+{
+ int ch;
+ spe_comment(gen->f, -4, "MAD:");
+ for (ch = 0; ch < 4; ch++) {
+ if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) {
+ int s1_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]);
+ int s2_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[1]);
+ int s3_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[2]);
+ int d_reg = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]);
+ /* d = s1 * s2 + s3 */
+ spe_fma(gen->f, d_reg, s1_reg, s2_reg, s3_reg);
+ store_dest_reg(gen, d_reg, ch, &inst->FullDstRegisters[0]);
+ free_itemps(gen);
+ }
+ }
+ return true;
+}
+
+
+/**
+ * Emit linear interpolate. See emit_ADD for comments.
+ */
+static boolean
+emit_LERP(struct codegen *gen, const struct tgsi_full_instruction *inst)
+{
+ int ch;
+ spe_comment(gen->f, -4, "LERP:");
+ for (ch = 0; ch < 4; ch++) {
+ if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) {
+ int s1_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]);
+ int s2_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[1]);
+ int s3_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[2]);
+ int d_reg = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]);
+ /* d = s3 + s1(s2 - s3) */
+ spe_fs(gen->f, d_reg, s2_reg, s3_reg);
+ spe_fma(gen->f, d_reg, d_reg, s1_reg, s3_reg);
+ store_dest_reg(gen, d_reg, ch, &inst->FullDstRegisters[0]);
+ free_itemps(gen);
+ }
+ }
+ return true;
+}
/**
* Emit multiply. See emit_ADD for comments.
return true;
}
+/**
+ * Emit reciprocal. See emit_ADD for comments.
+ */
+static boolean
+emit_RCP(struct codegen *gen, const struct tgsi_full_instruction *inst)
+{
+ int ch;
+ spe_comment(gen->f, -4, "RCP:");
+ for (ch = 0; ch < 4; ch++) {
+ if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) {
+ int s1_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]);
+ int d_reg = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]);
+ /* d = 1/s1 */
+ spe_frest(gen->f, d_reg, s1_reg);
+ spe_fi(gen->f, d_reg, s1_reg, d_reg);
+ store_dest_reg(gen, d_reg, ch, &inst->FullDstRegisters[0]);
+ free_itemps(gen);
+ }
+ }
+ return true;
+}
+
+/**
+ * Emit reciprocal sqrt. See emit_ADD for comments.
+ */
+static boolean
+emit_RSQ(struct codegen *gen, const struct tgsi_full_instruction *inst)
+{
+ int ch;
+ spe_comment(gen->f, -4, "RSQ:");
+ for (ch = 0; ch < 4; ch++) {
+ if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) {
+ int s1_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]);
+ int d_reg = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]);
+ /* d = 1/s1 */
+ spe_frsqest(gen->f, d_reg, s1_reg);
+ spe_fi(gen->f, d_reg, s1_reg, d_reg);
+ store_dest_reg(gen, d_reg, ch, &inst->FullDstRegisters[0]);
+ free_itemps(gen);
+ }
+ }
+ return true;
+}
+
+/**
+ * Emit absolute value. See emit_ADD for comments.
+ */
+static boolean
+emit_ABS(struct codegen *gen, const struct tgsi_full_instruction *inst)
+{
+ int ch;
+ spe_comment(gen->f, -4, "ABS:");
+ for (ch = 0; ch < 4; ch++) {
+ if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) {
+ int s1_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]);
+ int d_reg = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]);
+ const int bit31mask_reg = get_itemp(gen);
+
+ /* mask with bit 31 set, the rest cleared */
+ spe_load_int(gen->f, bit31mask_reg, (1 << 31));
+
+ /* d = sign bit cleared in s1 */
+ spe_andc(gen->f, d_reg, s1_reg, bit31mask_reg);
+
+ store_dest_reg(gen, d_reg, ch, &inst->FullDstRegisters[0]);
+ free_itemps(gen);
+ }
+ }
+ return true;
+}
+
+/**
+ * Emit 3 component dot product. See emit_ADD for comments.
+ */
+static boolean
+emit_DP3(struct codegen *gen, const struct tgsi_full_instruction *inst)
+{
+ int ch;
+ spe_comment(gen->f, -4, "DP3:");
+
+ int s1_reg = get_src_reg(gen, CHAN_X, &inst->FullSrcRegisters[0]);
+ int s2_reg = get_src_reg(gen, CHAN_X, &inst->FullSrcRegisters[1]);
+ int tmp_reg = get_itemp(gen);
+ /* t = x0 * x1 */
+ spe_fm(gen->f, tmp_reg, s1_reg, s2_reg);
+
+ s1_reg = get_src_reg(gen, CHAN_Y, &inst->FullSrcRegisters[0]);
+ s2_reg = get_src_reg(gen, CHAN_Y, &inst->FullSrcRegisters[1]);
+ /* t = y0 * y1 + t */
+ spe_fma(gen->f, tmp_reg, s1_reg, s2_reg, tmp_reg);
+
+ s1_reg = get_src_reg(gen, CHAN_Z, &inst->FullSrcRegisters[0]);
+ s2_reg = get_src_reg(gen, CHAN_Z, &inst->FullSrcRegisters[1]);
+ /* t = z0 * z1 + t */
+ spe_fma(gen->f, tmp_reg, s1_reg, s2_reg, tmp_reg);
+
+ for (ch = 0; ch < 4; ch++) {
+ if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) {
+ store_dest_reg(gen, tmp_reg, ch, &inst->FullDstRegisters[0]);
+ }
+ }
+
+ free_itemps(gen);
+ return true;
+}
+
+/**
+ * Emit 4 component dot product. See emit_ADD for comments.
+ */
+static boolean
+emit_DP4(struct codegen *gen, const struct tgsi_full_instruction *inst)
+{
+ int ch;
+ spe_comment(gen->f, -4, "DP4:");
+
+ int s1_reg = get_src_reg(gen, CHAN_X, &inst->FullSrcRegisters[0]);
+ int s2_reg = get_src_reg(gen, CHAN_X, &inst->FullSrcRegisters[1]);
+ int tmp_reg = get_itemp(gen);
+ /* t = x0 * x1 */
+ spe_fm(gen->f, tmp_reg, s1_reg, s2_reg);
+
+ s1_reg = get_src_reg(gen, CHAN_Y, &inst->FullSrcRegisters[0]);
+ s2_reg = get_src_reg(gen, CHAN_Y, &inst->FullSrcRegisters[1]);
+ /* t = y0 * y1 + t */
+ spe_fma(gen->f, tmp_reg, s1_reg, s2_reg, tmp_reg);
+
+ s1_reg = get_src_reg(gen, CHAN_Z, &inst->FullSrcRegisters[0]);
+ s2_reg = get_src_reg(gen, CHAN_Z, &inst->FullSrcRegisters[1]);
+ /* t = z0 * z1 + t */
+ spe_fma(gen->f, tmp_reg, s1_reg, s2_reg, tmp_reg);
+
+ s1_reg = get_src_reg(gen, CHAN_W, &inst->FullSrcRegisters[0]);
+ s2_reg = get_src_reg(gen, CHAN_W, &inst->FullSrcRegisters[1]);
+ /* t = w0 * w1 + t */
+ spe_fma(gen->f, tmp_reg, s1_reg, s2_reg, tmp_reg);
+
+ for (ch = 0; ch < 4; ch++) {
+ if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) {
+ store_dest_reg(gen, tmp_reg, ch, &inst->FullDstRegisters[0]);
+ }
+ }
+
+ free_itemps(gen);
+ return true;
+}
+
+/**
+ * Emit homogeneous dot product. See emit_ADD for comments.
+ */
+static boolean
+emit_DPH(struct codegen *gen, const struct tgsi_full_instruction *inst)
+{
+ int ch;
+ spe_comment(gen->f, -4, "DPH:");
+
+ int s1_reg = get_src_reg(gen, CHAN_X, &inst->FullSrcRegisters[0]);
+ int s2_reg = get_src_reg(gen, CHAN_X, &inst->FullSrcRegisters[1]);
+ int tmp_reg = get_itemp(gen);
+
+ /* t = x0 * x1 */
+ spe_fm(gen->f, tmp_reg, s1_reg, s2_reg);
+
+ s1_reg = get_src_reg(gen, CHAN_Y, &inst->FullSrcRegisters[0]);
+ s2_reg = get_src_reg(gen, CHAN_Y, &inst->FullSrcRegisters[1]);
+ /* t = y0 * y1 + t */
+ spe_fma(gen->f, tmp_reg, s1_reg, s2_reg, tmp_reg);
+
+ s1_reg = get_src_reg(gen, CHAN_Z, &inst->FullSrcRegisters[0]);
+ s2_reg = get_src_reg(gen, CHAN_Z, &inst->FullSrcRegisters[1]);
+ /* t = z0 * z1 + t */
+ spe_fma(gen->f, tmp_reg, s1_reg, s2_reg, tmp_reg);
+
+ s2_reg = get_src_reg(gen, CHAN_W, &inst->FullSrcRegisters[1]);
+ /* t = w1 + t */
+ spe_fa(gen->f, tmp_reg, s2_reg, tmp_reg);
+
+ for (ch = 0; ch < 4; ch++) {
+ if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) {
+ store_dest_reg(gen, tmp_reg, ch, &inst->FullDstRegisters[0]);
+ }
+ }
+
+ free_itemps(gen);
+ return true;
+}
+
+/**
+ * Emit cross product. See emit_ADD for comments.
+ */
+static boolean
+emit_XPD(struct codegen *gen, const struct tgsi_full_instruction *inst)
+{
+ spe_comment(gen->f, -4, "XPD:");
+
+ int s1_reg = get_src_reg(gen, CHAN_Z, &inst->FullSrcRegisters[0]);
+ int s2_reg = get_src_reg(gen, CHAN_Y, &inst->FullSrcRegisters[1]);
+ int tmp_reg = get_itemp(gen);
+
+ /* t = z0 * y1 */
+ spe_fm(gen->f, tmp_reg, s1_reg, s2_reg);
+
+ s1_reg = get_src_reg(gen, CHAN_Y, &inst->FullSrcRegisters[0]);
+ s2_reg = get_src_reg(gen, CHAN_Z, &inst->FullSrcRegisters[1]);
+ /* t = y0 * z1 - t */
+ spe_fms(gen->f, tmp_reg, s1_reg, s2_reg, tmp_reg);
+
+ if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << CHAN_X)) {
+ store_dest_reg(gen, tmp_reg, CHAN_X, &inst->FullDstRegisters[0]);
+ }
+
+ s1_reg = get_src_reg(gen, CHAN_X, &inst->FullSrcRegisters[0]);
+ s2_reg = get_src_reg(gen, CHAN_Z, &inst->FullSrcRegisters[1]);
+ /* t = x0 * z1 */
+ spe_fm(gen->f, tmp_reg, s1_reg, s2_reg);
+
+ s1_reg = get_src_reg(gen, CHAN_Z, &inst->FullSrcRegisters[0]);
+ s2_reg = get_src_reg(gen, CHAN_X, &inst->FullSrcRegisters[1]);
+ /* t = z0 * x1 - t */
+ spe_fms(gen->f, tmp_reg, s1_reg, s2_reg, tmp_reg);
+
+ if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << CHAN_Y)) {
+ store_dest_reg(gen, tmp_reg, CHAN_Y, &inst->FullDstRegisters[0]);
+ }
+
+ s1_reg = get_src_reg(gen, CHAN_Y, &inst->FullSrcRegisters[0]);
+ s2_reg = get_src_reg(gen, CHAN_X, &inst->FullSrcRegisters[1]);
+ /* t = y0 * x1 */
+ spe_fm(gen->f, tmp_reg, s1_reg, s2_reg);
+
+ s1_reg = get_src_reg(gen, CHAN_X, &inst->FullSrcRegisters[0]);
+ s2_reg = get_src_reg(gen, CHAN_Y, &inst->FullSrcRegisters[1]);
+ /* t = x0 * y1 - t */
+ spe_fms(gen->f, tmp_reg, s1_reg, s2_reg, tmp_reg);
+
+ if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << CHAN_Z)) {
+ store_dest_reg(gen, tmp_reg, CHAN_Z, &inst->FullDstRegisters[0]);
+ }
+
+ free_itemps(gen);
+ return true;
+}
/**
* Emit set-if-greater-than.
return true;
}
+/**
+ * Emit set-if_less-then. See emit_SGT for comments.
+ */
+static boolean
+emit_SLT(struct codegen *gen, const struct tgsi_full_instruction *inst)
+{
+ int ch;
+
+ spe_comment(gen->f, -4, "SLT:");
+
+ for (ch = 0; ch < 4; ch++) {
+ if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) {
+ int s1_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]);
+ int s2_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[1]);
+ int d_reg = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]);
+
+ /* d = (s1 < s2) */
+ spe_fcgt(gen->f, d_reg, s2_reg, s1_reg);
+
+ /* convert d from 0x0/0xffffffff to 0.0/1.0 */
+ /* d = d & one_reg */
+ spe_and(gen->f, d_reg, d_reg, get_const_one_reg(gen));
+
+ store_dest_reg(gen, d_reg, ch, &inst->FullDstRegisters[0]);
+ free_itemps(gen);
+ }
+ }
+
+ return true;
+}
+
+/**
+ * Emit set-if_greater-then-or-equal. See emit_SGT for comments.
+ */
+static boolean
+emit_SGE(struct codegen *gen, const struct tgsi_full_instruction *inst)
+{
+ int ch;
+
+ spe_comment(gen->f, -4, "SGE:");
+
+ for (ch = 0; ch < 4; ch++) {
+ if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) {
+ int s1_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]);
+ int s2_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[1]);
+ int d_reg = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]);
+
+ /* d = (s1 >= s2) */
+ spe_fcgt(gen->f, d_reg, s2_reg, s1_reg);
+
+ /* convert d from 0x0/0xffffffff to 0.0/1.0 */
+ /* d = ~d & one_reg */
+ spe_andc(gen->f, d_reg, get_const_one_reg(gen), d_reg);
+
+ store_dest_reg(gen, d_reg, ch, &inst->FullDstRegisters[0]);
+ free_itemps(gen);
+ }
+ }
+
+ return true;
+}
+
+/**
+ * Emit set-if_less-then-or-equal. See emit_SGT for comments.
+ */
+static boolean
+emit_SLE(struct codegen *gen, const struct tgsi_full_instruction *inst)
+{
+ int ch;
+
+ spe_comment(gen->f, -4, "SLE:");
+
+ for (ch = 0; ch < 4; ch++) {
+ if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) {
+ int s1_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]);
+ int s2_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[1]);
+ int d_reg = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]);
+
+ /* d = (s1 <= s2) */
+ spe_fcgt(gen->f, d_reg, s1_reg, s2_reg);
+
+ /* convert d from 0x0/0xffffffff to 0.0/1.0 */
+ /* d = ~d & one_reg */
+ spe_andc(gen->f, d_reg, get_const_one_reg(gen), d_reg);
+
+ store_dest_reg(gen, d_reg, ch, &inst->FullDstRegisters[0]);
+ free_itemps(gen);
+ }
+ }
+
+ return true;
+}
+
+/**
+ * Emit set-if_equal. See emit_SGT for comments.
+ */
+static boolean
+emit_SEQ(struct codegen *gen, const struct tgsi_full_instruction *inst)
+{
+ int ch;
+
+ spe_comment(gen->f, -4, "SEQ:");
+
+ for (ch = 0; ch < 4; ch++) {
+ if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) {
+ int s1_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]);
+ int s2_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[1]);
+ int d_reg = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]);
+
+ /* d = (s1 == s2) */
+ spe_fceq(gen->f, d_reg, s1_reg, s2_reg);
+
+ /* convert d from 0x0/0xffffffff to 0.0/1.0 */
+ /* d = d & one_reg */
+ spe_and(gen->f, d_reg, d_reg, get_const_one_reg(gen));
+
+ store_dest_reg(gen, d_reg, ch, &inst->FullDstRegisters[0]);
+ free_itemps(gen);
+ }
+ }
+
+ return true;
+}
+
+/**
+ * Emit set-if_not_equal. See emit_SGT for comments.
+ */
+static boolean
+emit_SNE(struct codegen *gen, const struct tgsi_full_instruction *inst)
+{
+ int ch;
+
+ spe_comment(gen->f, -4, "SNE:");
+
+ for (ch = 0; ch < 4; ch++) {
+ if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) {
+ int s1_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]);
+ int s2_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[1]);
+ int d_reg = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]);
+
+ /* d = (s1 != s2) */
+ spe_fceq(gen->f, d_reg, s1_reg, s2_reg);
+ spe_nor(gen->f, d_reg, d_reg, d_reg);
+
+ /* convert d from 0x0/0xffffffff to 0.0/1.0 */
+ /* d = d & one_reg */
+ spe_and(gen->f, d_reg, d_reg, get_const_one_reg(gen));
+
+ store_dest_reg(gen, d_reg, ch, &inst->FullDstRegisters[0]);
+ free_itemps(gen);
+ }
+ }
+
+ return true;
+}
+
+/**
+ * Emit compare. See emit_SGT for comments.
+ */
+static boolean
+emit_CMP(struct codegen *gen, const struct tgsi_full_instruction *inst)
+{
+ int ch;
+
+ spe_comment(gen->f, -4, "CMP:");
+
+ for (ch = 0; ch < 4; ch++) {
+ if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) {
+ int s1_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]);
+ int s2_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[1]);
+ int s3_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[2]);
+ int d_reg = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]);
+ int zero_reg = get_itemp(gen);
+
+ spe_xor(gen->f, zero_reg, zero_reg, zero_reg);
+
+ /* d = (s1 < 0) ? s2 : s3 */
+ spe_fcgt(gen->f, d_reg, zero_reg, s1_reg);
+ spe_selb(gen->f, d_reg, s3_reg, s2_reg, d_reg);
+
+ store_dest_reg(gen, d_reg, ch, &inst->FullDstRegisters[0]);
+ free_itemps(gen);
+ }
+ }
+
+ return true;
+}
+
+/**
+ * Emit trunc.
+ * Convert float to signed int
+ * Convert signed int to float
+ */
+static boolean
+emit_TRUNC(struct codegen *gen, const struct tgsi_full_instruction *inst)
+{
+ int ch;
+
+ spe_comment(gen->f, -4, "TRUNC:");
+
+ for (ch = 0; ch < 4; ch++) {
+ if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) {
+ int s1_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]);
+ int d_reg = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]);
+
+ /* Convert float to int */
+ spe_cflts(gen->f, d_reg, s1_reg, 0);
+
+ /* Convert int to float */
+ spe_csflt(gen->f, d_reg, d_reg, 0);
+
+ store_dest_reg(gen, d_reg, ch, &inst->FullDstRegisters[0]);
+ free_itemps(gen);
+ }
+ }
+
+ return true;
+}
+
+/**
+ * Emit floor.
+ * If negative int subtract one
+ * Convert float to signed int
+ * Convert signed int to float
+ */
+static boolean
+emit_FLR(struct codegen *gen, const struct tgsi_full_instruction *inst)
+{
+ int ch;
+
+ spe_comment(gen->f, -4, "FLR:");
+
+ int zero_reg = get_itemp(gen);
+ spe_xor(gen->f, zero_reg, zero_reg, zero_reg);
+
+ for (ch = 0; ch < 4; ch++) {
+ if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) {
+ int s1_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]);
+ int d_reg = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]);
+ int tmp_reg = get_itemp(gen);
+
+ /* If negative, subtract 1.0 */
+ spe_fcgt(gen->f, d_reg, zero_reg, s1_reg);
+ spe_selb(gen->f, tmp_reg, zero_reg, get_const_one_reg(gen), d_reg);
+ spe_fs(gen->f, d_reg, s1_reg, tmp_reg);
+
+ /* Convert float to int */
+ spe_cflts(gen->f, d_reg, d_reg, 0);
+
+ /* Convert int to float */
+ spe_csflt(gen->f, d_reg, d_reg, 0);
+
+ store_dest_reg(gen, d_reg, ch, &inst->FullDstRegisters[0]);
+ free_itemps(gen);
+ }
+ }
+
+ return true;
+}
+
+/**
+ * Emit frac.
+ * Input - FLR(Input)
+ */
+static boolean
+emit_FRC(struct codegen *gen, const struct tgsi_full_instruction *inst)
+{
+ int ch;
+
+ spe_comment(gen->f, -4, "FLR:");
+
+ int zero_reg = get_itemp(gen);
+ spe_xor(gen->f, zero_reg, zero_reg, zero_reg);
+
+ for (ch = 0; ch < 4; ch++) {
+ if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) {
+ int s1_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]);
+ int d_reg = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]);
+ int tmp_reg = get_itemp(gen);
+
+ /* If negative, subtract 1.0 */
+ spe_fcgt(gen->f, d_reg, zero_reg, s1_reg);
+ spe_selb(gen->f, tmp_reg, zero_reg, get_const_one_reg(gen), d_reg);
+ spe_fs(gen->f, d_reg, s1_reg, tmp_reg);
+
+ /* Convert float to int */
+ spe_cflts(gen->f, d_reg, d_reg, 0);
+
+ /* Convert int to float */
+ spe_csflt(gen->f, d_reg, d_reg, 0);
+
+ /* d = s1 - FLR(s1) */
+ spe_fs(gen->f, d_reg, s1_reg, d_reg);
+
+ store_dest_reg(gen, d_reg, ch, &inst->FullDstRegisters[0]);
+ free_itemps(gen);
+ }
+ }
+
+ return true;
+}
+
+
+/**
+ * Emit max. See emit_SGT for comments.
+ */
+static boolean
+emit_MAX(struct codegen *gen, const struct tgsi_full_instruction *inst)
+{
+ int ch;
+
+ spe_comment(gen->f, -4, "MAX:");
+
+ for (ch = 0; ch < 4; ch++) {
+ if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) {
+ int s1_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]);
+ int s2_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[1]);
+ int d_reg = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]);
+
+ /* d = (s1 > s2) ? s1 : s2 */
+ spe_fcgt(gen->f, d_reg, s1_reg, s2_reg);
+ spe_selb(gen->f, d_reg, s2_reg, s1_reg, d_reg);
+
+ store_dest_reg(gen, d_reg, ch, &inst->FullDstRegisters[0]);
+ free_itemps(gen);
+ }
+ }
+
+ return true;
+}
+
+/**
+ * Emit max. See emit_SGT for comments.
+ */
+static boolean
+emit_MIN(struct codegen *gen, const struct tgsi_full_instruction *inst)
+{
+ int ch;
+
+ spe_comment(gen->f, -4, "MIN:");
+
+ for (ch = 0; ch < 4; ch++) {
+ if (inst->FullDstRegisters[0].DstRegister.WriteMask & (1 << ch)) {
+ int s1_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[0]);
+ int s2_reg = get_src_reg(gen, ch, &inst->FullSrcRegisters[1]);
+ int d_reg = get_dst_reg(gen, ch, &inst->FullDstRegisters[0]);
+
+ /* d = (s2 > s1) ? s1 : s2 */
+ spe_fcgt(gen->f, d_reg, s2_reg, s1_reg);
+ spe_selb(gen->f, d_reg, s2_reg, s1_reg, d_reg);
+
+ store_dest_reg(gen, d_reg, ch, &inst->FullDstRegisters[0]);
+ free_itemps(gen);
+ }
+ }
+
+ return true;
+}
static boolean
emit_IF(struct codegen *gen, const struct tgsi_full_instruction *inst)
spe_comment(gen->f, -4, "IF:");
/* update execution mask with the predicate register */
- int tmp_reg = spe_allocate_available_register(gen->f);
+ int tmp_reg = get_itemp(gen);
int s1_reg = get_src_reg(gen, channel, &inst->FullSrcRegisters[0]);
/* tmp = (s1_reg == 0) */
spe_ceqi(gen->f, tmp_reg, s1_reg, 0);
/* tmp = !tmp */
- spe_complement(gen->f, tmp_reg);
+ spe_complement(gen->f, tmp_reg, tmp_reg);
/* exec_mask = exec_mask & tmp */
spe_and(gen->f, exec_reg, exec_reg, tmp_reg);
gen->if_nesting++;
- spe_release_register(gen->f, tmp_reg);
+ free_itemps(gen);
return true;
}
spe_comment(gen->f, -4, "ELSE:");
/* exec_mask = !exec_mask */
- spe_complement(gen->f, exec_reg);
+ spe_complement(gen->f, exec_reg, exec_reg);
return true;
}
{
switch (inst->Instruction.Opcode) {
case TGSI_OPCODE_MOV:
+ case TGSI_OPCODE_SWZ:
return emit_MOV(gen, inst);
case TGSI_OPCODE_MUL:
return emit_MUL(gen, inst);
case TGSI_OPCODE_ADD:
return emit_ADD(gen, inst);
+ case TGSI_OPCODE_SUB:
+ return emit_SUB(gen, inst);
+ case TGSI_OPCODE_MAD:
+ return emit_MAD(gen, inst);
+ case TGSI_OPCODE_LERP:
+ return emit_LERP(gen, inst);
+ case TGSI_OPCODE_DP3:
+ return emit_DP3(gen, inst);
+ case TGSI_OPCODE_DP4:
+ return emit_DP4(gen, inst);
+ case TGSI_OPCODE_DPH:
+ return emit_DPH(gen, inst);
+ case TGSI_OPCODE_XPD:
+ return emit_XPD(gen, inst);
+ case TGSI_OPCODE_RCP:
+ return emit_RCP(gen, inst);
+ case TGSI_OPCODE_RSQ:
+ return emit_RSQ(gen, inst);
+ case TGSI_OPCODE_ABS:
+ return emit_ABS(gen, inst);
case TGSI_OPCODE_SGT:
return emit_SGT(gen, inst);
+ case TGSI_OPCODE_SLT:
+ return emit_SLT(gen, inst);
+ case TGSI_OPCODE_SGE:
+ return emit_SGE(gen, inst);
+ case TGSI_OPCODE_SLE:
+ return emit_SLE(gen, inst);
+ case TGSI_OPCODE_SEQ:
+ return emit_SEQ(gen, inst);
+ case TGSI_OPCODE_SNE:
+ return emit_SNE(gen, inst);
+ case TGSI_OPCODE_CMP:
+ return emit_CMP(gen, inst);
+ case TGSI_OPCODE_MAX:
+ return emit_MAX(gen, inst);
+ case TGSI_OPCODE_MIN:
+ return emit_MIN(gen, inst);
+ case TGSI_OPCODE_TRUNC:
+ return emit_TRUNC(gen, inst);
+ case TGSI_OPCODE_FLR:
+ return emit_FLR(gen, inst);
+ case TGSI_OPCODE_FRC:
+ return emit_FRC(gen, inst);
case TGSI_OPCODE_END:
return emit_END(gen);
/* XXX lots more cases to do... */
default:
+ fprintf(stderr, "Cell: unimplemented TGSI instruction %d!\n",
+ inst->Instruction.Opcode);
return false;
}
* For each TGSI TEMPORARY we allocate four SPE registers.
*/
static boolean
-emit_declaration(struct codegen *gen, const struct tgsi_full_declaration *decl)
+emit_declaration(struct cell_context *cell,
+ struct codegen *gen, const struct tgsi_full_declaration *decl)
{
int i, ch;
switch (decl->Declaration.File) {
case TGSI_FILE_TEMPORARY:
-#if DISASSEM
- printf("Declare temp reg %d .. %d\n",
- decl->DeclarationRange.First,
- decl->DeclarationRange.Last);
-#endif
+ if (cell->debug_flags & CELL_DEBUG_ASM) {
+ printf("Declare temp reg %d .. %d\n",
+ decl->DeclarationRange.First,
+ decl->DeclarationRange.Last);
+ }
+
for (i = decl->DeclarationRange.First;
i <= decl->DeclarationRange.Last;
i++) {
* to SPU memory. someday...
*/
-#if DISASSEM
- printf(" SPE regs: %d %d %d %d\n",
- gen->temp_regs[i][0],
- gen->temp_regs[i][1],
- gen->temp_regs[i][2],
- gen->temp_regs[i][3]);
-#endif
+ if (cell->debug_flags & CELL_DEBUG_ASM) {
+ printf(" SPE regs: %d %d %d %d\n",
+ gen->temp_regs[i][0],
+ gen->temp_regs[i][1],
+ gen->temp_regs[i][2],
+ gen->temp_regs[i][3]);
+ }
}
break;
default:
spe_allocate_register(f, gen.outputs_reg);
spe_allocate_register(f, gen.constants_reg);
-#if DISASSEM
- spe_print_code(f, true);
- spe_indent(f, 8);
- printf("Begin %s\n", __FUNCTION__);
- tgsi_dump(tokens, 0);
-#endif
+ if (cell->debug_flags & CELL_DEBUG_ASM) {
+ spe_print_code(f, true);
+ spe_indent(f, 8);
+ printf("Begin %s\n", __FUNCTION__);
+ tgsi_dump(tokens, 0);
+ }
tgsi_parse_init(&parse, tokens);
break;
case TGSI_TOKEN_TYPE_DECLARATION:
- if (!emit_declaration(&gen, &parse.FullToken.FullDeclaration))
+ if (!emit_declaration(cell, &gen, &parse.FullToken.FullDeclaration))
gen.error = true;
break;
return emit_END(&gen);
}
-#if DISASSEM
- printf("cell_gen_fragment_program nr instructions: %d\n", f->num_inst);
- printf("End %s\n", __FUNCTION__);
-#endif
+ if (cell->debug_flags & CELL_DEBUG_ASM) {
+ printf("cell_gen_fragment_program nr instructions: %d\n", f->num_inst);
+ printf("End %s\n", __FUNCTION__);
+ }
tgsi_parse_free( &parse );
projects / mesa.git / blobdiff