summaryrefslogtreecommitdiffstats
path: root/src/gallium/drivers/vc4
diff options
context:
space:
mode:
Diffstat (limited to 'src/gallium/drivers/vc4')
-rw-r--r--src/gallium/drivers/vc4/Automake.inc4
-rw-r--r--src/gallium/drivers/vc4/Makefile.am1
-rw-r--r--src/gallium/drivers/vc4/vc4_nir_lower_blend.c4
-rw-r--r--src/gallium/drivers/vc4/vc4_nir_lower_io.c7
-rw-r--r--src/gallium/drivers/vc4/vc4_opt_algebraic.c2
-rw-r--r--src/gallium/drivers/vc4/vc4_program.c6
-rw-r--r--src/gallium/drivers/vc4/vc4_qir.c8
-rw-r--r--src/gallium/drivers/vc4/vc4_qir.h8
-rw-r--r--src/gallium/drivers/vc4/vc4_qir_lower_uniforms.c56
-rw-r--r--src/gallium/drivers/vc4/vc4_qpu_emit.c4
-rw-r--r--src/gallium/drivers/vc4/vc4_reorder_uniforms.c26
11 files changed, 86 insertions, 40 deletions
diff --git a/src/gallium/drivers/vc4/Automake.inc b/src/gallium/drivers/vc4/Automake.inc
index 6fa3e190cac..5664c2ab14e 100644
--- a/src/gallium/drivers/vc4/Automake.inc
+++ b/src/gallium/drivers/vc4/Automake.inc
@@ -6,8 +6,4 @@ TARGET_LIB_DEPS += \
$(top_builddir)/src/gallium/winsys/vc4/drm/libvc4drm.la \
$(top_builddir)/src/gallium/drivers/vc4/libvc4.la
-if USE_VC4_SIMULATOR
-TARGET_CPPFLAGS += -DUSE_VC4_SIMULATOR
-endif
-
endif
diff --git a/src/gallium/drivers/vc4/Makefile.am b/src/gallium/drivers/vc4/Makefile.am
index f4a57ba3404..a3bf72fc72a 100644
--- a/src/gallium/drivers/vc4/Makefile.am
+++ b/src/gallium/drivers/vc4/Makefile.am
@@ -23,7 +23,6 @@ include Makefile.sources
include $(top_srcdir)/src/gallium/Automake.inc
if USE_VC4_SIMULATOR
-SIM_CFLAGS = -DUSE_VC4_SIMULATOR=1
SIM_LDFLAGS = -lsimpenrose
endif
diff --git a/src/gallium/drivers/vc4/vc4_nir_lower_blend.c b/src/gallium/drivers/vc4/vc4_nir_lower_blend.c
index 373c9e12d11..0672a92226f 100644
--- a/src/gallium/drivers/vc4/vc4_nir_lower_blend.c
+++ b/src/gallium/drivers/vc4/vc4_nir_lower_blend.c
@@ -509,8 +509,8 @@ vc4_nir_lower_blend_instr(struct vc4_compile *c, nir_builder *b,
nir_ssa_def *dst_vec4 = nir_unpack_unorm_4x8(b, packed_dst_color);
nir_ssa_def *src_color[4], *unpacked_dst_color[4];
for (unsigned i = 0; i < 4; i++) {
- src_color[i] = nir_swizzle(b, intr->src[0].ssa, &i, 1, false);
- unpacked_dst_color[i] = nir_swizzle(b, dst_vec4, &i, 1, false);
+ src_color[i] = nir_channel(b, intr->src[0].ssa, i);
+ unpacked_dst_color[i] = nir_channel(b, dst_vec4, i);
}
vc4_nir_emit_alpha_test_discard(c, b, src_color[3]);
diff --git a/src/gallium/drivers/vc4/vc4_nir_lower_io.c b/src/gallium/drivers/vc4/vc4_nir_lower_io.c
index 7ea263afb68..1afe52a63f4 100644
--- a/src/gallium/drivers/vc4/vc4_nir_lower_io.c
+++ b/src/gallium/drivers/vc4/vc4_nir_lower_io.c
@@ -84,7 +84,7 @@ vc4_nir_unpack_16u(nir_builder *b, nir_ssa_def *src, unsigned chan)
static nir_ssa_def *
vc4_nir_unpack_8f(nir_builder *b, nir_ssa_def *src, unsigned chan)
{
- return nir_swizzle(b, nir_unpack_unorm_4x8(b, src), &chan, 1, false);
+ return nir_channel(b, nir_unpack_unorm_4x8(b, src), chan);
}
static nir_ssa_def *
@@ -326,9 +326,8 @@ vc4_nir_lower_output(struct vc4_compile *c, nir_builder *b,
intr_comp->const_index[0] = intr->const_index[0] * 4 + i;
assert(intr->src[0].is_ssa);
- intr_comp->src[0] = nir_src_for_ssa(nir_swizzle(b,
- intr->src[0].ssa,
- &i, 1, false));
+ intr_comp->src[0] =
+ nir_src_for_ssa(nir_channel(b, intr->src[0].ssa, i));
nir_builder_instr_insert(b, &intr_comp->instr);
}
diff --git a/src/gallium/drivers/vc4/vc4_opt_algebraic.c b/src/gallium/drivers/vc4/vc4_opt_algebraic.c
index f1bab810eff..07a92266dd2 100644
--- a/src/gallium/drivers/vc4/vc4_opt_algebraic.c
+++ b/src/gallium/drivers/vc4/vc4_opt_algebraic.c
@@ -144,6 +144,8 @@ qir_opt_algebraic(struct vc4_compile *c)
case QOP_SEL_X_Y_ZC:
case QOP_SEL_X_Y_NS:
case QOP_SEL_X_Y_NC:
+ case QOP_SEL_X_Y_CS:
+ case QOP_SEL_X_Y_CC:
if (is_zero(c, inst->src[1])) {
/* Replace references to a 0 uniform value
* with the SEL_X_0 equivalent.
diff --git a/src/gallium/drivers/vc4/vc4_program.c b/src/gallium/drivers/vc4/vc4_program.c
index a48dad804e2..197577b6c20 100644
--- a/src/gallium/drivers/vc4/vc4_program.c
+++ b/src/gallium/drivers/vc4/vc4_program.c
@@ -987,6 +987,10 @@ ntq_emit_alu(struct vc4_compile *c, nir_alu_instr *instr)
qir_SF(c, qir_SUB(c, src[0], src[1]));
*dest = qir_SEL_X_0_NC(c, qir_uniform_ui(c, ~0));
break;
+ case nir_op_uge:
+ qir_SF(c, qir_SUB(c, src[0], src[1]));
+ *dest = qir_SEL_X_0_CC(c, qir_uniform_ui(c, ~0));
+ break;
case nir_op_ilt:
qir_SF(c, qir_SUB(c, src[0], src[1]));
*dest = qir_SEL_X_0_NS(c, qir_uniform_ui(c, ~0));
@@ -1167,7 +1171,7 @@ emit_point_size_write(struct vc4_compile *c)
struct qreg point_size;
if (c->output_point_size_index != -1)
- point_size = c->outputs[c->output_point_size_index + 3];
+ point_size = c->outputs[c->output_point_size_index];
else
point_size = qir_uniform_f(c, 1.0);
diff --git a/src/gallium/drivers/vc4/vc4_qir.c b/src/gallium/drivers/vc4/vc4_qir.c
index 7894b081b19..f2855e159fc 100644
--- a/src/gallium/drivers/vc4/vc4_qir.c
+++ b/src/gallium/drivers/vc4/vc4_qir.c
@@ -69,10 +69,14 @@ static const struct qir_op_info qir_op_info[] = {
[QOP_SEL_X_0_NC] = { "fsel_x_0_nc", 1, 1, false, true },
[QOP_SEL_X_0_ZS] = { "fsel_x_0_zs", 1, 1, false, true },
[QOP_SEL_X_0_ZC] = { "fsel_x_0_zc", 1, 1, false, true },
+ [QOP_SEL_X_0_CS] = { "fsel_x_0_cs", 1, 1, false, true },
+ [QOP_SEL_X_0_CC] = { "fsel_x_0_cc", 1, 1, false, true },
[QOP_SEL_X_Y_NS] = { "fsel_x_y_ns", 1, 2, false, true },
[QOP_SEL_X_Y_NC] = { "fsel_x_y_nc", 1, 2, false, true },
[QOP_SEL_X_Y_ZS] = { "fsel_x_y_zs", 1, 2, false, true },
[QOP_SEL_X_Y_ZC] = { "fsel_x_y_zc", 1, 2, false, true },
+ [QOP_SEL_X_Y_CS] = { "fsel_x_y_cs", 1, 2, false, true },
+ [QOP_SEL_X_Y_CC] = { "fsel_x_y_cc", 1, 2, false, true },
[QOP_RCP] = { "rcp", 1, 1, false, true },
[QOP_RSQ] = { "rsq", 1, 1, false, true },
@@ -218,10 +222,14 @@ qir_depends_on_flags(struct qinst *inst)
case QOP_SEL_X_0_NC:
case QOP_SEL_X_0_ZS:
case QOP_SEL_X_0_ZC:
+ case QOP_SEL_X_0_CS:
+ case QOP_SEL_X_0_CC:
case QOP_SEL_X_Y_NS:
case QOP_SEL_X_Y_NC:
case QOP_SEL_X_Y_ZS:
case QOP_SEL_X_Y_ZC:
+ case QOP_SEL_X_Y_CS:
+ case QOP_SEL_X_Y_CC:
return true;
default:
return false;
diff --git a/src/gallium/drivers/vc4/vc4_qir.h b/src/gallium/drivers/vc4/vc4_qir.h
index a92ad93ee07..ddb35e41fcf 100644
--- a/src/gallium/drivers/vc4/vc4_qir.h
+++ b/src/gallium/drivers/vc4/vc4_qir.h
@@ -99,11 +99,15 @@ enum qop {
QOP_SEL_X_0_ZC,
QOP_SEL_X_0_NS,
QOP_SEL_X_0_NC,
+ QOP_SEL_X_0_CS,
+ QOP_SEL_X_0_CC,
/* Selects the src[0] if the ns flag bit is set, otherwise src[1]. */
QOP_SEL_X_Y_ZS,
QOP_SEL_X_Y_ZC,
QOP_SEL_X_Y_NS,
QOP_SEL_X_Y_NC,
+ QOP_SEL_X_Y_CS,
+ QOP_SEL_X_Y_CC,
QOP_FTOI,
QOP_ITOF,
@@ -567,10 +571,14 @@ QIR_ALU1(SEL_X_0_ZS)
QIR_ALU1(SEL_X_0_ZC)
QIR_ALU1(SEL_X_0_NS)
QIR_ALU1(SEL_X_0_NC)
+QIR_ALU1(SEL_X_0_CS)
+QIR_ALU1(SEL_X_0_CC)
QIR_ALU2(SEL_X_Y_ZS)
QIR_ALU2(SEL_X_Y_ZC)
QIR_ALU2(SEL_X_Y_NS)
QIR_ALU2(SEL_X_Y_NC)
+QIR_ALU2(SEL_X_Y_CS)
+QIR_ALU2(SEL_X_Y_CC)
QIR_ALU2(FMIN)
QIR_ALU2(FMAX)
QIR_ALU2(FMINABS)
diff --git a/src/gallium/drivers/vc4/vc4_qir_lower_uniforms.c b/src/gallium/drivers/vc4/vc4_qir_lower_uniforms.c
index f087c3b81b5..a57e100593c 100644
--- a/src/gallium/drivers/vc4/vc4_qir_lower_uniforms.c
+++ b/src/gallium/drivers/vc4/vc4_qir_lower_uniforms.c
@@ -22,14 +22,10 @@
*/
/**
- * @file vc4_opt_algebraic.c
+ * @file vc4_qir_lower_uniforms.c
*
- * This is the optimization pass for miscellaneous changes to instructions
- * where we can simplify the operation by some knowledge about the specific
- * operations.
- *
- * Mostly this will be a matter of turning things into MOVs so that they can
- * later be copy-propagated out.
+ * This is the pre-code-generation pass for fixing up instructions that try to
+ * read from multiple uniform values.
*/
#include "vc4_qir.h"
@@ -85,6 +81,33 @@ is_lowerable_uniform(struct qinst *inst, int i)
return true;
}
+/* Returns the number of different uniform values referenced by the
+ * instruction.
+ */
+static uint32_t
+qir_get_instruction_uniform_count(struct qinst *inst)
+{
+ uint32_t count = 0;
+
+ for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
+ if (inst->src[i].file != QFILE_UNIF)
+ continue;
+
+ bool is_duplicate = false;
+ for (int j = 0; j < i; j++) {
+ if (inst->src[j].file == QFILE_UNIF &&
+ inst->src[j].index == inst->src[i].index) {
+ is_duplicate = true;
+ break;
+ }
+ }
+ if (!is_duplicate)
+ count++;
+ }
+
+ return count;
+}
+
void
qir_lower_uniforms(struct vc4_compile *c)
{
@@ -98,13 +121,7 @@ qir_lower_uniforms(struct vc4_compile *c)
list_for_each_entry(struct qinst, inst, &c->instructions, link) {
uint32_t nsrc = qir_get_op_nsrc(inst->op);
- uint32_t count = 0;
- for (int i = 0; i < nsrc; i++) {
- if (inst->src[i].file == QFILE_UNIF)
- count++;
- }
-
- if (count <= 1)
+ if (qir_get_instruction_uniform_count(inst) <= 1)
continue;
for (int i = 0; i < nsrc; i++) {
@@ -140,23 +157,22 @@ qir_lower_uniforms(struct vc4_compile *c)
list_for_each_entry(struct qinst, inst, &c->instructions, link) {
uint32_t nsrc = qir_get_op_nsrc(inst->op);
- uint32_t count = 0;
- for (int i = 0; i < nsrc; i++) {
- if (inst->src[i].file == QFILE_UNIF)
- count++;
- }
+ uint32_t count = qir_get_instruction_uniform_count(inst);
if (count <= 1)
continue;
+ bool removed = false;
for (int i = 0; i < nsrc; i++) {
if (is_lowerable_uniform(inst, i) &&
inst->src[i].index == max_index) {
inst->src[i] = temp;
remove_uniform(ht, unif);
- count--;
+ removed = true;
}
}
+ if (removed)
+ count--;
/* If the instruction doesn't need lowering any more,
* then drop it from the list.
diff --git a/src/gallium/drivers/vc4/vc4_qpu_emit.c b/src/gallium/drivers/vc4/vc4_qpu_emit.c
index 133e1385178..e0d3633da42 100644
--- a/src/gallium/drivers/vc4/vc4_qpu_emit.c
+++ b/src/gallium/drivers/vc4/vc4_qpu_emit.c
@@ -311,6 +311,8 @@ vc4_generate_code(struct vc4_context *vc4, struct vc4_compile *c)
case QOP_SEL_X_0_ZC:
case QOP_SEL_X_0_NS:
case QOP_SEL_X_0_NC:
+ case QOP_SEL_X_0_CS:
+ case QOP_SEL_X_0_CC:
queue(c, qpu_a_MOV(dst, src[0]) | unpack);
set_last_cond_add(c, qinst->op - QOP_SEL_X_0_ZS +
QPU_COND_ZS);
@@ -324,6 +326,8 @@ vc4_generate_code(struct vc4_context *vc4, struct vc4_compile *c)
case QOP_SEL_X_Y_ZC:
case QOP_SEL_X_Y_NS:
case QOP_SEL_X_Y_NC:
+ case QOP_SEL_X_Y_CS:
+ case QOP_SEL_X_Y_CC:
queue(c, qpu_a_MOV(dst, src[0]));
if (qinst->src[0].pack)
*(last_inst(c)) |= unpack;
diff --git a/src/gallium/drivers/vc4/vc4_reorder_uniforms.c b/src/gallium/drivers/vc4/vc4_reorder_uniforms.c
index 7f11fba2340..85a0c95e851 100644
--- a/src/gallium/drivers/vc4/vc4_reorder_uniforms.c
+++ b/src/gallium/drivers/vc4/vc4_reorder_uniforms.c
@@ -44,18 +44,28 @@ qir_reorder_uniforms(struct vc4_compile *c)
uint32_t next_uniform = 0;
list_for_each_entry(struct qinst, inst, &c->instructions, link) {
+ uint32_t new = ~0;
+
for (int i = 0; i < qir_get_op_nsrc(inst->op); i++) {
if (inst->src[i].file != QFILE_UNIF)
continue;
- uint32_t new = next_uniform++;
- if (uniform_index_size <= new) {
- uniform_index_size =
- MAX2(uniform_index_size * 2, 16);
- uniform_index =
- realloc(uniform_index,
- uniform_index_size *
- sizeof(uint32_t));
+ if (new == ~0) {
+ new = next_uniform++;
+ if (uniform_index_size <= new) {
+ uniform_index_size =
+ MAX2(uniform_index_size * 2, 16);
+ uniform_index =
+ realloc(uniform_index,
+ uniform_index_size *
+ sizeof(uint32_t));
+ }
+ } else {
+ /* If we've got two uniform references in this
+ * instruction, they need to be the same
+ * uniform value.
+ */
+ assert(inst->src[i].index == uniform_index[new]);
}
uniform_index[new] = inst->src[i].index;
generated by cgit v1.2.3 (git 2.39.1) at 2024-12-02 15:41:57 +0000