freedreno/ir3: use standard list implementation

Use standard list_head double-linked list and related iterators,
helpers, etc, rather than weird combo of instruction array and next
pointers depending on stage.  Now block has an instrs_list.  In
certain stages where we want to remove and re-add to the blocks list
we just use list_replace() to copy the list to a new list_head.

Signed-off-by: Rob Clark <[email protected]>

8 files changed, 161 insertions, 209 deletions

diff --git a/src/gallium/drivers/freedreno/ir3/ir3.c b/src/gallium/drivers/freedreno/ir3/ir3.c
index e015de91c33..84564a9eef7 100644
--- a/src/gallium/drivers/freedreno/ir3/ir3.c
+++ b/src/gallium/drivers/freedreno/ir3/ir3.c
@@ -81,7 +81,7 @@ void ir3_destroy(struct ir3 *shader)
 		shader->chunk = chunk->next;
 		free(chunk);
 	}
-	free(shader->instrs);
+	free(shader->indirects);
 	free(shader->baryfs);
 	free(shader);
 }
@@ -534,28 +534,32 @@ static int (*emit[])(struct ir3_instruction *instr, void *ptr,
 void * ir3_assemble(struct ir3 *shader, struct ir3_info *info,
 		uint32_t gpu_id)
 {
+	struct ir3_block *block = shader->block;
 	uint32_t *ptr, *dwords;
-	uint32_t i;
 
 	info->max_reg       = -1;
 	info->max_half_reg  = -1;
 	info->max_const     = -1;
 	info->instrs_count  = 0;
+	info->sizedwords    = 0;
+
+	list_for_each_entry (struct ir3_instruction, instr, &block->instr_list, node) {
+		info->sizedwords += 2;
+	}
 
 	/* need a integer number of instruction "groups" (sets of 16
 	 * instructions on a4xx or sets of 4 instructions on a3xx),
 	 * so pad out w/ NOPs if needed: (NOTE each instruction is 64bits)
 	 */
 	if (gpu_id >= 400) {
-		info->sizedwords = 2 * align(shader->instrs_count, 16);
+		info->sizedwords = align(info->sizedwords, 16 * 2);
 	} else {
-		info->sizedwords = 2 * align(shader->instrs_count, 4);
+		info->sizedwords = align(info->sizedwords, 4 * 2);
 	}
 
 	ptr = dwords = calloc(4, info->sizedwords);
 
-	for (i = 0; i < shader->instrs_count; i++) {
-		struct ir3_instruction *instr = shader->instrs[i];
+	list_for_each_entry (struct ir3_instruction, instr, &block->instr_list, node) {
 		int ret = emit[instr->category](instr, dwords, info);
 		if (ret)
 			goto fail;
@@ -581,14 +585,15 @@ static struct ir3_register * reg_create(struct ir3 *shader,
 	return reg;
 }
 
-static void insert_instr(struct ir3 *shader,
+static void insert_instr(struct ir3_block *block,
 		struct ir3_instruction *instr)
 {
+	struct ir3 *shader = block->shader;
 #ifdef DEBUG
 	static uint32_t serialno = 0;
 	instr->serialno = ++serialno;
 #endif
-	array_insert(shader->instrs, instr);
+	list_addtail(&instr->node, &block->instr_list);
 
 	if (is_input(instr))
 		array_insert(shader->baryfs, instr);
@@ -625,6 +630,8 @@ struct ir3_block * ir3_block_create(struct ir3 *shader,
 
 	block->shader = shader;
 
+	list_inithead(&block->instr_list);
+
 	return block;
 }
 
@@ -652,7 +659,7 @@ struct ir3_instruction * ir3_instr_create2(struct ir3_block *block,
 	instr->block = block;
 	instr->category = category;
 	instr->opc = opc;
-	insert_instr(block->shader, instr);
+	insert_instr(block, instr);
 	return instr;
 }
 
@@ -677,7 +684,7 @@ struct ir3_instruction * ir3_instr_clone(struct ir3_instruction *instr)
 	*new_instr = *instr;
 	new_instr->regs = regs;
 
-	insert_instr(instr->block->shader, new_instr);
+	insert_instr(instr->block, new_instr);
 
 	/* clone registers: */
 	new_instr->regs_count = 0;
diff --git a/src/gallium/drivers/freedreno/ir3/ir3.h b/src/gallium/drivers/freedreno/ir3/ir3.h
index f37dfab3341..edb5b49e23c 100644
--- a/src/gallium/drivers/freedreno/ir3/ir3.h
+++ b/src/gallium/drivers/freedreno/ir3/ir3.h
@@ -28,6 +28,7 @@
 #include <stdbool.h>
 
 #include "util/u_debug.h"
+#include "util/list.h"
 
 #include "instr-a3xx.h"
 #include "disasm.h"  /* TODO move 'enum shader_t' somewhere else.. */
@@ -290,7 +291,9 @@ struct ir3_instruction {
 	 */
 	struct ir3_instruction *fanin;
 
-	struct ir3_instruction *next;
+	/* Entry in ir3_block's instruction list: */
+	struct list_head node;
+
 #ifdef DEBUG
 	uint32_t serialno;
 #endif
@@ -321,8 +324,6 @@ static inline int ir3_neighbor_count(struct ir3_instruction *instr)
 struct ir3_heap_chunk;
 
 struct ir3 {
-	unsigned instrs_count, instrs_sz;
-	struct ir3_instruction **instrs;
 
 	/* Track bary.f (and ldlv) instructions.. this is needed in
 	 * scheduling to ensure that all varying fetches happen before
@@ -361,7 +362,7 @@ struct ir3_block {
 	/* only a single address register: */
 	struct ir3_instruction *address;
 	struct ir3_block *parent;
-	struct ir3_instruction *head;
+	struct list_head instr_list;
 };
 
 struct ir3 * ir3_create(void);
@@ -402,11 +403,8 @@ static inline void ir3_clear_mark(struct ir3 *shader)
 	 * a block, so tracking the list of instrs globally is
 	 * unlikely to be what we want.
 	 */
-	unsigned i;
-	for (i = 0; i < shader->instrs_count; i++) {
-		struct ir3_instruction *instr = shader->instrs[i];
+	list_for_each_entry (struct ir3_instruction, instr, &shader->block->instr_list, node)
 		instr->flags &= ~IR3_INSTR_MARK;
-	}
 }
 
 static inline int ir3_instr_regno(struct ir3_instruction *instr,
@@ -756,6 +754,7 @@ int ir3_block_flatten(struct ir3_block *block);
 /* depth calculation: */
 int ir3_delayslots(struct ir3_instruction *assigner,
 		struct ir3_instruction *consumer, unsigned n);
+void ir3_insert_by_depth(struct ir3_instruction *instr, struct list_head *list);
 void ir3_block_depth(struct ir3_block *block);
 
 /* copy-propagate: */
diff --git a/src/gallium/drivers/freedreno/ir3/ir3_cp.c b/src/gallium/drivers/freedreno/ir3/ir3_cp.c
index fa7d363be7b..350f7dd5e6b 100644
--- a/src/gallium/drivers/freedreno/ir3/ir3_cp.c
+++ b/src/gallium/drivers/freedreno/ir3/ir3_cp.c
@@ -354,13 +354,6 @@ instr_cp(struct ir3_instruction *instr, unsigned *flags)
 {
 	struct ir3_register *reg;
 
-	/* stay within the block.. don't try to operate across
-	 * basic block boundaries or we'll have problems when
-	 * dealing with multiple basic blocks:
-	 */
-	if (is_meta(instr) && (instr->opc == OPC_META_INPUT))
-		return instr;
-
 	if (is_eligible_mov(instr, !!flags)) {
 		struct ir3_register *reg = instr->regs[1];
 		struct ir3_instruction *src_instr = ssa(reg);
@@ -394,11 +387,11 @@ instr_cp(struct ir3_instruction *instr, unsigned *flags)
 	return instr;
 }
 
-static void block_cp(struct ir3_block *block)
+void ir3_block_cp(struct ir3_block *block)
 {
-	unsigned i;
+	ir3_clear_mark(block->shader);
 
-	for (i = 0; i < block->noutputs; i++) {
+	for (unsigned i = 0; i < block->noutputs; i++) {
 		if (block->outputs[i]) {
 			struct ir3_instruction *out =
 					instr_cp(block->outputs[i], NULL);
@@ -407,9 +400,3 @@ static void block_cp(struct ir3_block *block)
 		}
 	}
 }
-
-void ir3_block_cp(struct ir3_block *block)
-{
-	ir3_clear_mark(block->shader);
-	block_cp(block);
-}
diff --git a/src/gallium/drivers/freedreno/ir3/ir3_depth.c b/src/gallium/drivers/freedreno/ir3/ir3_depth.c
index b899c66b37e..601e14a1c85 100644
--- a/src/gallium/drivers/freedreno/ir3/ir3_depth.c
+++ b/src/gallium/drivers/freedreno/ir3/ir3_depth.c
@@ -84,25 +84,25 @@ int ir3_delayslots(struct ir3_instruction *assigner,
 	}
 }
 
-static void insert_by_depth(struct ir3_instruction *instr)
+void
+ir3_insert_by_depth(struct ir3_instruction *instr, struct list_head *list)
 {
-	struct ir3_block *block = instr->block;
-	struct ir3_instruction *n = block->head;
-	struct ir3_instruction *p = NULL;
-
-	while (n && (n != instr) && (n->depth > instr->depth)) {
-		p = n;
-		n = n->next;
+	/* remove from existing spot in list: */
+	list_delinit(&instr->node);
+
+	/* find where to re-insert instruction: */
+	list_for_each_entry (struct ir3_instruction, pos, list, node) {
+		if (pos->depth > instr->depth) {
+			list_add(&instr->node, &pos->node);
+			return;
+		}
 	}
-
-	instr->next = n;
-	if (p)
-		p->next = instr;
-	else
-		block->head = instr;
+	/* if we get here, we didn't find an insertion spot: */
+	list_addtail(&instr->node, list);
 }
 
-static void ir3_instr_depth(struct ir3_instruction *instr)
+static void
+ir3_instr_depth(struct ir3_instruction *instr)
 {
 	struct ir3_instruction *src;
 
@@ -123,42 +123,38 @@ static void ir3_instr_depth(struct ir3_instruction *instr)
 		instr->depth = MAX2(instr->depth, sd);
 	}
 
-	/* meta-instructions don't add cycles, other than PHI.. which
-	 * might translate to a real instruction..
-	 *
-	 * well, not entirely true, fan-in/out, etc might need to need
-	 * to generate some extra mov's in edge cases, etc.. probably
-	 * we might want to do depth calculation considering the worst
-	 * case for these??
-	 */
 	if (!is_meta(instr))
 		instr->depth++;
 
-	insert_by_depth(instr);
+	ir3_insert_by_depth(instr, &instr->block->instr_list);
+}
+
+static void
+remove_unused_by_block(struct ir3_block *block)
+{
+	list_for_each_entry_safe (struct ir3_instruction, instr, &block->instr_list, node) {
+		if (!ir3_instr_check_mark(instr)) {
+			/* mark it, in case it is input, so we can
+			 * remove unused inputs:
+			 */
+			instr->depth = DEPTH_UNUSED;
+			/* and remove from instruction list: */
+			list_delinit(&instr->node);
+		}
+	}
 }
 
 void ir3_block_depth(struct ir3_block *block)
 {
 	unsigned i;
 
-	block->head = NULL;
-
 	ir3_clear_mark(block->shader);
 	for (i = 0; i < block->noutputs; i++)
 		if (block->outputs[i])
 			ir3_instr_depth(block->outputs[i]);
 
 	/* mark un-used instructions: */
-	for (i = 0; i < block->shader->instrs_count; i++) {
-		struct ir3_instruction *instr = block->shader->instrs[i];
-
-		/* just consider instructions within this block: */
-		if (instr->block != block)
-			continue;
-
-		if (!ir3_instr_check_mark(instr))
-			instr->depth = DEPTH_UNUSED;
-	}
+	remove_unused_by_block(block);
 
 	/* cleanup unused inputs: */
 	for (i = 0; i < block->ninputs; i++) {
diff --git a/src/gallium/drivers/freedreno/ir3/ir3_legalize.c b/src/gallium/drivers/freedreno/ir3/ir3_legalize.c
index 61713c25e72..be0b5ce442c 100644
--- a/src/gallium/drivers/freedreno/ir3/ir3_legalize.c
+++ b/src/gallium/drivers/freedreno/ir3/ir3_legalize.c
@@ -51,12 +51,9 @@ struct ir3_legalize_ctx {
 static void legalize(struct ir3_legalize_ctx *ctx)
 {
 	struct ir3_block *block = ctx->block;
-	struct ir3_instruction *n;
-	struct ir3 *shader = block->shader;
-	struct ir3_instruction *end =
-			ir3_instr_create(block, 0, OPC_END);
 	struct ir3_instruction *last_input = NULL;
 	struct ir3_instruction *last_rel = NULL;
+	struct list_head instr_list;
 	regmask_t needs_ss_war;       /* write after read */
 	regmask_t needs_ss;
 	regmask_t needs_sy;
@@ -65,9 +62,13 @@ static void legalize(struct ir3_legalize_ctx *ctx)
 	regmask_init(&needs_ss);
 	regmask_init(&needs_sy);
 
-	shader->instrs_count = 0;
+	/* remove all the instructions from the list, we'll be adding
+	 * them back in as we go
+	 */
+	list_replace(&block->instr_list, &instr_list);
+	list_inithead(&block->instr_list);
 
-	for (n = block->head; n; n = n->next) {
+	list_for_each_entry_safe (struct ir3_instruction, n, &instr_list, node) {
 		struct ir3_register *reg;
 		unsigned i;
 
@@ -140,12 +141,12 @@ static void legalize(struct ir3_legalize_ctx *ctx)
 		}
 
 		/* need to be able to set (ss) on first instruction: */
-		if ((shader->instrs_count == 0) && (n->category >= 5))
+		if (list_empty(&block->instr_list) && (n->category >= 5))
 			ir3_NOP(block);
 
-		if (is_nop(n) && shader->instrs_count) {
-			struct ir3_instruction *last =
-					shader->instrs[shader->instrs_count-1];
+		if (is_nop(n) && !list_empty(&block->instr_list)) {
+			struct ir3_instruction *last = list_last_entry(&block->instr_list,
+					struct ir3_instruction, node);
 			if (is_nop(last) && (last->repeat < 5)) {
 				last->repeat++;
 				last->flags |= n->flags;
@@ -153,7 +154,7 @@ static void legalize(struct ir3_legalize_ctx *ctx)
 			}
 		}
 
-		shader->instrs[shader->instrs_count++] = n;
+		list_addtail(&n->node, &block->instr_list);
 
 		if (is_sfu(n))
 			regmask_set(&needs_ss, n->regs[0]);
@@ -192,35 +193,19 @@ static void legalize(struct ir3_legalize_ctx *ctx)
 		 * the (ei) flag:
 		 */
 		if (is_mem(last_input) && (last_input->opc == OPC_LDLV)) {
-			int i, cnt;
+			struct ir3_instruction *baryf;
 
-			/* note that ir3_instr_create() inserts into
-			 * shader->instrs[] and increments the count..
-			 * so we need to bump up the cnt initially (to
-			 * avoid it clobbering the last real instr) and
-			 * restore it after.
-			 */
-			cnt = ++shader->instrs_count;
-
-			/* inserting instructions would be a bit nicer if list.. */
-			for (i = cnt - 2; i >= 0; i--) {
-				if (shader->instrs[i] == last_input) {
-
-					/* (ss)bary.f (ei)r63.x, 0, r0.x */
-					last_input = ir3_instr_create(block, 2, OPC_BARY_F);
-					last_input->flags |= IR3_INSTR_SS;
-					ir3_reg_create(last_input, regid(63, 0), 0);
-					ir3_reg_create(last_input, 0, IR3_REG_IMMED)->iim_val = 0;
-					ir3_reg_create(last_input, regid(0, 0), 0);
+			/* (ss)bary.f (ei)r63.x, 0, r0.x */
+			baryf = ir3_instr_create(block, 2, OPC_BARY_F);
+			baryf->flags |= IR3_INSTR_SS;
+			ir3_reg_create(baryf, regid(63, 0), 0);
+			ir3_reg_create(baryf, 0, IR3_REG_IMMED)->iim_val = 0;
+			ir3_reg_create(baryf, regid(0, 0), 0);
 
-					shader->instrs[i + 1] = last_input;
-
-					break;
-				}
-				shader->instrs[i + 1] = shader->instrs[i];
-			}
+			/* insert the dummy bary.f after last_input: */
+			list_add(&baryf->node, &last_input->node);
 
-			shader->instrs_count = cnt;
+			last_input = baryf;
 		}
 		last_input->regs[0]->flags |= IR3_REG_EI;
 	}
@@ -228,9 +213,11 @@ static void legalize(struct ir3_legalize_ctx *ctx)
 	if (last_rel)
 		last_rel->flags |= IR3_INSTR_UL;
 
-	shader->instrs[shader->instrs_count++] = end;
+	/* create/add 'end' instruction: */
+	ir3_instr_create(block, 0, OPC_END);
 
-	shader->instrs[0]->flags |= IR3_INSTR_SS | IR3_INSTR_SY;
+	list_first_entry(&block->instr_list, struct ir3_instruction, node)
+		->flags |= IR3_INSTR_SS | IR3_INSTR_SY;
 }
 
 void ir3_block_legalize(struct ir3_block *block,
diff --git a/src/gallium/drivers/freedreno/ir3/ir3_print.c b/src/gallium/drivers/freedreno/ir3/ir3_print.c
index a5c5d3c8efa..755c0c23c36 100644
--- a/src/gallium/drivers/freedreno/ir3/ir3_print.c
+++ b/src/gallium/drivers/freedreno/ir3/ir3_print.c
@@ -186,9 +186,8 @@ void ir3_print_instr(struct ir3_instruction *instr)
 static void
 print_block(struct ir3_block *block, int lvl)
 {
-	struct ir3_instruction *instr;
 	tab(lvl); printf("block {\n");
-	for (instr = block->head; instr; instr = instr->next) {
+	list_for_each_entry (struct ir3_instruction, instr, &block->instr_list, node) {
 		print_instr(instr, lvl+1);
 	}
 	tab(lvl); printf("}\n");
diff --git a/src/gallium/drivers/freedreno/ir3/ir3_ra.c b/src/gallium/drivers/freedreno/ir3/ir3_ra.c
index 501352515b5..95f6a81861e 100644
--- a/src/gallium/drivers/freedreno/ir3/ir3_ra.c
+++ b/src/gallium/drivers/freedreno/ir3/ir3_ra.c
@@ -75,10 +75,10 @@ struct ir3_ra_ctx {
 #  define ra_debug 0
 #endif
 
-#define ra_dump_list(msg, n) do { \
+#define ra_dump_list(msg, ir) do { \
 		if (ra_debug) { \
 			debug_printf("-- " msg); \
-			ir3_print(n->block->shader); \
+			ir3_print(ir); \
 		} \
 	} while (0)
 
@@ -175,14 +175,13 @@ static void mark_sources(struct ir3_instruction *instr,
 static void compute_liveregs(struct ir3_ra_ctx *ctx,
 		struct ir3_instruction *instr, regmask_t *liveregs)
 {
-	struct ir3_block *block = instr->block;
-	struct ir3_instruction *n;
+	struct ir3_block *block = ctx->block;
 	regmask_t written;
 	unsigned i;
 
 	regmask_init(&written);
 
-	for (n = instr->next; n; n = n->next) {
+	list_for_each_entry (struct ir3_instruction, n, &instr->node, node) {
 		struct ir3_register *r;
 
 		if (is_meta(n))
@@ -411,9 +410,8 @@ static void instr_assign_src(struct ir3_ra_ctx *ctx,
 static void instr_assign_srcs(struct ir3_ra_ctx *ctx,
 		struct ir3_instruction *instr, unsigned name)
 {
-	struct ir3_instruction *n, *src;
-
-	for (n = instr->next; n && !ctx->error; n = n->next) {
+	list_for_each_entry (struct ir3_instruction, n, &instr->node, node) {
+		struct ir3_instruction *src;
 		foreach_ssa_src_n(src, i, n) {
 			unsigned r = i + 1;
 
@@ -424,6 +422,8 @@ static void instr_assign_srcs(struct ir3_ra_ctx *ctx,
 			if (src == instr)
 				instr_assign_src(ctx, n, r, name);
 		}
+		if (ctx->error)
+			break;
 	}
 }
 
@@ -589,14 +589,45 @@ static void instr_assign_array(struct ir3_ra_ctx *ctx,
 
 }
 
-static int block_ra(struct ir3_ra_ctx *ctx, struct ir3_block *block)
+static bool
+block_ra(struct ir3_block *block, void *state)
 {
-	struct ir3_instruction *n;
+	struct ir3_ra_ctx *ctx = state;
+
+	ra_dump_list("-------\n", block->shader);
+
+	/* first pass, assign arrays: */
+	list_for_each_entry (struct ir3_instruction, n, &block->instr_list, node) {
+		if (is_meta(n) && (n->opc == OPC_META_FI) && n->fi.aid) {
+			debug_assert(!n->cp.left);  /* don't think this should happen */
+			ra_dump_instr("ASSIGN ARRAY: ", n);
+			instr_assign_array(ctx, n);
+			ra_dump_list("-------\n", block->shader);
+		}
+
+		if (ctx->error)
+			return false;
+	}
+
+	list_for_each_entry (struct ir3_instruction, n, &block->instr_list, node) {
+		ra_dump_instr("ASSIGN: ", n);
+		instr_alloc_and_assign(ctx, ir3_neighbor_first(n));
+		ra_dump_list("-------\n", block->shader);
+
+		if (ctx->error)
+			return false;
+	}
+
+	return true;
+}
 
+static int
+shader_ra(struct ir3_ra_ctx *ctx, struct ir3_block *block)
+{
 	/* frag shader inputs get pre-assigned, since we have some
 	 * constraints/unknowns about setup for some of these regs:
 	 */
-	if ((ctx->type == SHADER_FRAGMENT) && !block->parent) {
+	if (ctx->type == SHADER_FRAGMENT) {
 		unsigned i = 0, j;
 		if (ctx->frag_face && (i < block->ninputs) && block->inputs[i]) {
 			/* if we have frag_face, it gets hr0.x */
@@ -608,31 +639,23 @@ static int block_ra(struct ir3_ra_ctx *ctx, struct ir3_block *block)
 				instr_assign(ctx, block->inputs[i], j);
 	}
 
-	ra_dump_list("-------\n", block->head);
-
-	/* first pass, assign arrays: */
-	for (n = block->head; n && !ctx->error; n = n->next) {
-		if (is_meta(n) && (n->opc == OPC_META_FI) && n->fi.aid) {
-			debug_assert(!n->cp.left);  /* don't think this should happen */
-			ra_dump_instr("ASSIGN ARRAY: ", n);
-			instr_assign_array(ctx, n);
-			ra_dump_list("-------\n", block->head);
-		}
-	}
-
-	for (n = block->head; n && !ctx->error; n = n->next) {
-		ra_dump_instr("ASSIGN: ", n);
-		instr_alloc_and_assign(ctx, ir3_neighbor_first(n));
-		ra_dump_list("-------\n", block->head);
-	}
+	block_ra(block, ctx);
 
 	return ctx->error ? -1 : 0;
 }
 
+static bool
+block_mark_dst(struct ir3_block *block, void *state)
+{
+	list_for_each_entry (struct ir3_instruction, n, &block->instr_list, node)
+		if (n->regs_count > 0)
+			n->regs[0]->flags |= IR3_REG_SSA;
+	return true;
+}
+
 int ir3_block_ra(struct ir3_block *block, enum shader_t type,
 		bool frag_coord, bool frag_face)
 {
-	struct ir3_instruction *n;
 	struct ir3_ra_ctx ctx = {
 			.block = block,
 			.type = type,
@@ -648,12 +671,10 @@ int ir3_block_ra(struct ir3_block *block, enum shader_t type,
 	 * NOTE: we really should set SSA flag consistently on
 	 * every dst register in the frontend.
 	 */
-	for (n = block->head; n; n = n->next)
-		if (n->regs_count > 0)
-			n->regs[0]->flags |= IR3_REG_SSA;
+	block_mark_dst(block, &ctx);
 
 	ir3_clear_mark(block->shader);
-	ret = block_ra(&ctx, block);
+	ret = shader_ra(&ctx, block);
 
 	return ret;
 }
diff --git a/src/gallium/drivers/freedreno/ir3/ir3_sched.c b/src/gallium/drivers/freedreno/ir3/ir3_sched.c
index 5ca6d7b62d5..fc41f93b884 100644
--- a/src/gallium/drivers/freedreno/ir3/ir3_sched.c
+++ b/src/gallium/drivers/freedreno/ir3/ir3_sched.c
@@ -88,26 +88,21 @@ deepest(struct ir3_instruction **srcs, unsigned nsrcs)
 	return d;
 }
 
-static unsigned distance(struct ir3_sched_ctx *ctx,
-		struct ir3_instruction *instr, unsigned maxd)
+static unsigned
+distance(struct ir3_sched_ctx *ctx, struct ir3_instruction *instr,
+		unsigned maxd)
 {
-	struct ir3_instruction *n = ctx->scheduled;
+	struct list_head *instr_list = &instr->block->instr_list;
 	unsigned d = 0;
-	while (n && (n != instr) && (d < maxd)) {
+
+	list_for_each_entry_rev (struct ir3_instruction, n, instr_list, node) {
+		if ((n == instr) || (d >= maxd))
+			break;
 		if (is_alu(n) || is_flow(n))
 			d++;
-		n = n->next;
 	}
-	return d;
-}
 
-/* TODO maybe we want double linked list? */
-static struct ir3_instruction * prev(struct ir3_instruction *instr)
-{
-	struct ir3_instruction *p = instr->block->head;
-	while (p && (p->next != instr))
-		p = p->next;
-	return p;
+	return d;
 }
 
 static bool is_sfu_or_mem(struct ir3_instruction *instr)
@@ -125,25 +120,11 @@ static void schedule(struct ir3_sched_ctx *ctx,
 	 * scheduling and depth calculation..
 	 */
 	if (ctx->scheduled && is_sfu_or_mem(ctx->scheduled) && is_sfu_or_mem(instr))
-		schedule(ctx, ir3_NOP(block), false);
+		ir3_NOP(block);
 
 	/* remove from depth list:
 	 */
-	if (remove) {
-		struct ir3_instruction *p = prev(instr);
-
-		/* NOTE: this can happen for inputs which are not
-		 * read.. in that case there is no need to schedule
-		 * the input, so just bail:
-		 */
-		if (instr != (p ? p->next : block->head))
-			return;
-
-		if (p)
-			p->next = instr->next;
-		else
-			block->head = instr->next;
-	}
+	list_delinit(&instr->node);
 
 	if (writes_addr(instr)) {
 		assert(ctx->addr == NULL);
@@ -157,7 +138,7 @@ static void schedule(struct ir3_sched_ctx *ctx,
 
 	instr->flags |= IR3_INSTR_MARK;
 
-	instr->next = ctx->scheduled;
+	list_addtail(&instr->node, &instr->block->instr_list);
 	ctx->scheduled = instr;
 
 	ctx->cnt++;
@@ -284,18 +265,6 @@ static int trysched(struct ir3_sched_ctx *ctx,
 	return SCHEDULED;
 }
 
-static struct ir3_instruction * reverse(struct ir3_instruction *instr)
-{
-	struct ir3_instruction *reversed = NULL;
-	while (instr) {
-		struct ir3_instruction *next = instr->next;
-		instr->next = reversed;
-		reversed = instr;
-		instr = next;
-	}
-	return reversed;
-}
-
 static bool uses_current_addr(struct ir3_sched_ctx *ctx,
 		struct ir3_instruction *instr)
 {
@@ -317,16 +286,14 @@ static bool uses_current_pred(struct ir3_sched_ctx *ctx,
  * other instructions using the current address register:
  */
 static int block_sched_undelayed(struct ir3_sched_ctx *ctx,
-		struct ir3_block *block)
+		struct list_head *unscheduled_list)
 {
-	struct ir3_instruction *instr = block->head;
 	bool addr_in_use = false;
 	bool pred_in_use = false;
 	bool all_delayed = true;
 	unsigned cnt = ~0, attempted = 0;
 
-	while (instr) {
-		struct ir3_instruction *next = instr->next;
+	list_for_each_entry_safe(struct ir3_instruction, instr, unscheduled_list, node) {
 		bool addr = uses_current_addr(ctx, instr);
 		bool pred = uses_current_pred(ctx, instr);
 
@@ -347,8 +314,6 @@ static int block_sched_undelayed(struct ir3_sched_ctx *ctx,
 
 			attempted++;
 		}
-
-		instr = next;
 	}
 
 	if (!addr_in_use)
@@ -408,7 +373,10 @@ static int block_sched_undelayed(struct ir3_sched_ctx *ctx,
 
 static void block_sched(struct ir3_sched_ctx *ctx, struct ir3_block *block)
 {
-	struct ir3_instruction *instr;
+	struct list_head unscheduled_list;
+
+	list_replace(&block->instr_list, &unscheduled_list);
+	list_inithead(&block->instr_list);
 
 	/* schedule all the shader input's (meta-instr) first so that
 	 * the RA step sees that the input registers contain a value
@@ -423,31 +391,22 @@ static void block_sched(struct ir3_sched_ctx *ctx, struct ir3_block *block)
 		}
 	}
 
-	while ((instr = block->head) && !ctx->error) {
-		/* NOTE: always grab next *before* trysched(), in case the
-		 * instruction is actually scheduled (and therefore moved
-		 * from depth list into scheduled list)
-		 */
-		struct ir3_instruction *next = instr->next;
+	list_for_each_entry_safe (struct ir3_instruction, instr, &unscheduled_list, node) {
 		int cnt = trysched(ctx, instr);
 
 		if (cnt == DELAYED)
-			cnt = block_sched_undelayed(ctx, block);
+			cnt = block_sched_undelayed(ctx, &unscheduled_list);
 
 		/* -1 is signal to return up stack, but to us means same as 0: */
 		cnt = MAX2(0, cnt);
 		cnt += ctx->cnt;
-		instr = next;
 
 		/* if deepest remaining instruction cannot be scheduled, try
 		 * the increasingly more shallow instructions until needed
 		 * number of delay slots is filled:
 		 */
-		while (instr && (cnt > ctx->cnt)) {
-			next = instr->next;
+		list_for_each_entry_safe (struct ir3_instruction, instr, &instr->node, node)
 			trysched(ctx, instr);
-			instr = next;
-		}
 
 		/* and if we run out of instructions that can be scheduled,
 		 * then it is time for nop's:
@@ -455,9 +414,6 @@ static void block_sched(struct ir3_sched_ctx *ctx, struct ir3_block *block)
 		while (cnt > ctx->cnt)
 			schedule(ctx, ir3_NOP(block), false);
 	}
-
-	/* at this point, scheduled list is in reverse order, so fix that: */
-	block->head = reverse(ctx->scheduled);
 }
 
 int ir3_block_sched(struct ir3_block *block)