freedreno: a2xx: use nir_lower_alu_to_scalar instead of lowering pass

nir_lower_alu_to_scalar can now be used to only lower certain ops, so we
don't need the custom pass. And we can lower fall_equal/fany_nequal with
lower_vector_cmp instead.

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

4 files changed, 12 insertions, 178 deletions

diff --git a/src/gallium/drivers/freedreno/Makefile.sources b/src/gallium/drivers/freedreno/Makefile.sources
index d316bf56804..85d38186aa5 100644
--- a/src/gallium/drivers/freedreno/Makefile.sources
+++ b/src/gallium/drivers/freedreno/Makefile.sources
@@ -73,7 +73,6 @@ a2xx_SOURCES := \
 	a2xx/ir2_assemble.c \
 	a2xx/ir2_cp.c \
 	a2xx/ir2_nir.c \
-	a2xx/ir2_nir_lower_scalar.c \
 	a2xx/ir2_private.h \
 	a2xx/ir2_ra.c
 
diff --git a/src/gallium/drivers/freedreno/a2xx/ir2_nir.c b/src/gallium/drivers/freedreno/a2xx/ir2_nir.c
index 980ab7b1013..21ecf3a7003 100644
--- a/src/gallium/drivers/freedreno/a2xx/ir2_nir.c
+++ b/src/gallium/drivers/freedreno/a2xx/ir2_nir.c
@@ -41,6 +41,7 @@ static const nir_shader_compiler_options options = {
 	.vertex_id_zero_based = true, /* its not implemented anyway */
 	.lower_bitops = true,
 	.lower_rotate = true,
+	.lower_vector_cmp = true,
 };
 
 const nir_shader_compiler_options *
@@ -1084,8 +1085,17 @@ ir2_nir_compile(struct ir2_context *ctx, bool binning)
 	OPT_V(ctx->nir, nir_lower_bool_to_float);
 	OPT_V(ctx->nir, nir_lower_to_source_mods, nir_lower_all_source_mods);
 
-	/* lower to scalar instructions that can only be scalar on a2xx */
-	OPT_V(ctx->nir, ir2_nir_lower_scalar);
+	/* TODO: static bitset ? */
+	BITSET_DECLARE(scalar_ops, nir_num_opcodes);
+	BITSET_ZERO(scalar_ops);
+	BITSET_SET(scalar_ops, nir_op_frsq);
+	BITSET_SET(scalar_ops, nir_op_frcp);
+	BITSET_SET(scalar_ops, nir_op_flog2);
+	BITSET_SET(scalar_ops, nir_op_fexp2);
+	BITSET_SET(scalar_ops, nir_op_fsqrt);
+	BITSET_SET(scalar_ops, nir_op_fcos);
+	BITSET_SET(scalar_ops, nir_op_fsin);
+	OPT_V(ctx->nir, nir_lower_alu_to_scalar, scalar_ops);
 
 	OPT_V(ctx->nir, nir_lower_locals_to_regs);
 
diff --git a/src/gallium/drivers/freedreno/a2xx/ir2_nir_lower_scalar.c b/src/gallium/drivers/freedreno/a2xx/ir2_nir_lower_scalar.c
deleted file mode 100644
index 1af211e759e..00000000000
--- a/src/gallium/drivers/freedreno/a2xx/ir2_nir_lower_scalar.c
+++ /dev/null
@@ -1,174 +0,0 @@
-/*
- * Copyright (C) 2018 Jonathan Marek <[email protected]>
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the "Software"),
- * to deal in the Software without restriction, including without limitation
- * the rights to use, copy, modify, merge, publish, distribute, sublicense,
- * and/or sell copies of the Software, and to permit persons to whom the
- * Software is furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice (including the next
- * paragraph) shall be included in all copies or substantial portions of the
- * Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
- * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
- * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
- * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- *
- * Authors:
- *    Jonathan Marek <[email protected]>
- */
-
-/* some operations can only be scalar on a2xx:
- *  rsq, rcp, log2, exp2, cos, sin, sqrt
- * mostly copy-pasted from nir_lower_alu_to_scalar.c
- */
-
-#include "ir2_private.h"
-#include "compiler/nir/nir_builder.h"
-
-static void
-nir_alu_ssa_dest_init(nir_alu_instr * instr, unsigned num_components,
-					  unsigned bit_size)
-{
-	nir_ssa_dest_init(&instr->instr, &instr->dest.dest, num_components,
-					  bit_size, NULL);
-	instr->dest.write_mask = (1 << num_components) - 1;
-}
-
-static void
-lower_reduction(nir_alu_instr * instr, nir_op chan_op, nir_op merge_op,
-				nir_builder * builder)
-{
-	unsigned num_components = nir_op_infos[instr->op].input_sizes[0];
-
-	nir_ssa_def *last = NULL;
-	for (unsigned i = 0; i < num_components; i++) {
-		nir_alu_instr *chan =
-			nir_alu_instr_create(builder->shader, chan_op);
-		nir_alu_ssa_dest_init(chan, 1, instr->dest.dest.ssa.bit_size);
-		nir_alu_src_copy(&chan->src[0], &instr->src[0], chan);
-		chan->src[0].swizzle[0] = chan->src[0].swizzle[i];
-		if (nir_op_infos[chan_op].num_inputs > 1) {
-			assert(nir_op_infos[chan_op].num_inputs == 2);
-			nir_alu_src_copy(&chan->src[1], &instr->src[1], chan);
-			chan->src[1].swizzle[0] = chan->src[1].swizzle[i];
-		}
-		chan->exact = instr->exact;
-
-		nir_builder_instr_insert(builder, &chan->instr);
-
-		if (i == 0) {
-			last = &chan->dest.dest.ssa;
-		} else {
-			last = nir_build_alu(builder, merge_op,
-								 last, &chan->dest.dest.ssa, NULL, NULL);
-		}
-	}
-
-	assert(instr->dest.write_mask == 1);
-	nir_ssa_def_rewrite_uses(&instr->dest.dest.ssa, nir_src_for_ssa(last));
-	nir_instr_remove(&instr->instr);
-}
-
-static bool lower_scalar(nir_alu_instr * instr, nir_builder * b)
-{
-	assert(instr->dest.dest.is_ssa);
-	assert(instr->dest.write_mask != 0);
-
-	b->cursor = nir_before_instr(&instr->instr);
-	b->exact = instr->exact;
-
-#define LOWER_REDUCTION(name, chan, merge) \
-	case name##2: \
-	case name##3: \
-	case name##4: \
-		lower_reduction(instr, chan, merge, b); \
-		return true;
-
-	switch (instr->op) {
-		/* TODO: handle these instead of lowering */
-		LOWER_REDUCTION(nir_op_fall_equal, nir_op_seq, nir_op_fmin);
-		LOWER_REDUCTION(nir_op_fany_nequal, nir_op_sne, nir_op_fmax);
-
-	default:
-		return false;
-	case nir_op_frsq:
-	case nir_op_frcp:
-	case nir_op_flog2:
-	case nir_op_fexp2:
-	case nir_op_fcos:
-	case nir_op_fsin:
-	case nir_op_fsqrt:
-		break;
-	}
-
-	assert(nir_op_infos[instr->op].num_inputs == 1);
-
-	unsigned num_components = instr->dest.dest.ssa.num_components;
-	nir_ssa_def *comps[NIR_MAX_VEC_COMPONENTS] = { NULL };
-	unsigned chan;
-
-	if (num_components == 1)
-		return false;
-
-	for (chan = 0; chan < num_components; chan++) {
-		assert(instr->dest.write_mask & (1 << chan));
-
-		nir_alu_instr *lower = nir_alu_instr_create(b->shader, instr->op);
-
-		nir_alu_src_copy(&lower->src[0], &instr->src[0], lower);
-		lower->src[0].swizzle[0] = instr->src[0].swizzle[chan];
-
-		nir_alu_ssa_dest_init(lower, 1, instr->dest.dest.ssa.bit_size);
-		lower->dest.saturate = instr->dest.saturate;
-		comps[chan] = &lower->dest.dest.ssa;
-		lower->exact = instr->exact;
-
-		nir_builder_instr_insert(b, &lower->instr);
-	}
-
-	nir_ssa_def *vec = nir_vec(b, comps, num_components);
-
-	nir_ssa_def_rewrite_uses(&instr->dest.dest.ssa, nir_src_for_ssa(vec));
-
-	nir_instr_remove(&instr->instr);
-	return true;
-}
-
-static bool lower_scalar_impl(nir_function_impl * impl)
-{
-	nir_builder builder;
-	nir_builder_init(&builder, impl);
-	bool progress = false;
-
-	nir_foreach_block(block, impl) {
-		nir_foreach_instr_safe(instr, block) {
-			if (instr->type == nir_instr_type_alu)
-				progress = lower_scalar(nir_instr_as_alu(instr), &builder)
-					|| progress;
-		}
-	}
-
-	nir_metadata_preserve(impl, nir_metadata_block_index |
-						  nir_metadata_dominance);
-
-	return progress;
-}
-
-bool ir2_nir_lower_scalar(nir_shader * shader)
-{
-	bool progress = false;
-
-	nir_foreach_function(function, shader) {
-		if (function->impl)
-			progress = lower_scalar_impl(function->impl) || progress;
-	}
-
-	return progress;
-}
diff --git a/src/gallium/drivers/freedreno/meson.build b/src/gallium/drivers/freedreno/meson.build
index 8bd251b5a23..c0dcc1ea316 100644
--- a/src/gallium/drivers/freedreno/meson.build
+++ b/src/gallium/drivers/freedreno/meson.build
@@ -91,7 +91,6 @@ files_libfreedreno = files(
   'a2xx/ir2_assemble.c',
   'a2xx/ir2_cp.c',
   'a2xx/ir2_nir.c',
-  'a2xx/ir2_nir_lower_scalar.c',
   'a2xx/ir2_private.h',
   'a2xx/ir2_ra.c',
   'a3xx/fd3_blend.c',