freedreno/ir3: allow copy-propagate out of fanout

[mesa.git] / src / freedreno / ir3 / ir3_a4xx.c

diff --git a/src/freedreno/ir3/ir3_a4xx.c b/src/freedreno/ir3/ir3_a4xx.c
index 426a143acfbb3899246837bdd7d368834838edee..dd654bacc7cdf9e0c723e5916eabc6a96c8a1457 100644 (file)
--- a/src/freedreno/ir3/ir3_a4xx.c
+++ b/src/freedreno/ir3/ir3_a4xx.c
@@ -217,14 +217,15 @@ get_image_offset(struct ir3_context *ctx, const nir_variable *var,
        /* to calculate the byte offset (yes, uggg) we need (up to) three
         * const values to know the bytes per pixel, and y and z stride:
         */
-       unsigned cb = regid(ctx->so->constbase.image_dims, 0) +
-               ctx->so->const_layout.image_dims.off[var->data.driver_location];
+       struct ir3_const_state *const_state = &ctx->so->shader->const_state;
+       unsigned cb = regid(const_state->offsets.image_dims, 0) +
+               const_state->image_dims.off[var->data.driver_location];
 
-       debug_assert(ctx->so->const_layout.image_dims.mask &
+       debug_assert(const_state->image_dims.mask &
                        (1 << var->data.driver_location));
 
        /* offset = coords.x * bytes_per_pixel: */
-       offset = ir3_MUL_S(b, coords[0], 0, create_uniform(b, cb + 0), 0);
+       offset = ir3_MUL_S24(b, coords[0], 0, create_uniform(b, cb + 0), 0);
        if (ncoords > 1) {
                /* offset += coords.y * y_pitch: */
                offset = ir3_MAD_S24(b, create_uniform(b, cb + 1), 0,
@@ -316,10 +317,12 @@ emit_intrinsic_atomic_image(struct ir3_context *ctx, nir_intrinsic_instr *intr)
        case nir_intrinsic_image_deref_atomic_add:
                atomic = ir3_ATOMIC_ADD_G(b, image, 0, src0, 0, src1, 0, src2, 0);
                break;
-       case nir_intrinsic_image_deref_atomic_min:
+       case nir_intrinsic_image_deref_atomic_imin:
+       case nir_intrinsic_image_deref_atomic_umin:
                atomic = ir3_ATOMIC_MIN_G(b, image, 0, src0, 0, src1, 0, src2, 0);
                break;
-       case nir_intrinsic_image_deref_atomic_max:
+       case nir_intrinsic_image_deref_atomic_imax:
+       case nir_intrinsic_image_deref_atomic_umax:
                atomic = ir3_ATOMIC_MAX_G(b, image, 0, src0, 0, src1, 0, src2, 0);
                break;
        case nir_intrinsic_image_deref_atomic_and: